AI transcript
0:00:06 let’s like sort of rapid fire like best tool for that use case right for research yeah the ai tool
0:00:15 that i use the most more than any other ai tool hey welcome to the next wave podcast i’m matt wolf
0:00:21 i’m here with nathan lands and today we’re going to talk a little bit about the various large language
0:00:26 models that are out there i feel like one of the things that we hear a lot is like okay you’ve got
0:00:32 claud you’ve got chat gpt you’ve got llama you’ve got perplexity you’ve got grok you’ve got
0:00:38 you know the list goes on and on but the reality is some of these tools are really good at some
0:00:43 things some of them are really bad at things and so we wanted to spend a little bit of time on this
0:00:48 episode just kind of deep diving in like what’s this good for what’s this good for if i was doing
0:00:56 copywriting what would i use if i’m using research what would i use when all your marketing team does
0:01:01 is put out fires they burn out but with hub spot they can achieve their best results without the
0:01:08 stress tap into hub spots collection of ai tools breeze to pinpoint leads capture attention
0:01:14 and access all your data in one place keep your marketers cool and your campaign results hotter
0:01:24 than ever visit hub spot dot com slash marketers to learn more so just something i think it should
0:01:29 be a really valuable resource for people just hey i need to do this let me go double check with
0:01:36 that episode of the next wave on how i actually should do that like what tool should i be using
0:01:40 to go do that so that’s what this episode’s all about we don’t have a guest today it’s just me
0:01:46 and nathan chatting from our own experience about these tools just getting into it here one of the
0:01:52 biggest news in the ai world right now is from chat gpt from open ai right they just released
0:01:58 their new open ai 01 model maybe do you want to kind of break down a little bit i mean i mean some of
0:02:02 it we don’t fully know how it’s working yet there’s a lot of like debate about what’s actually going
0:02:07 on with the new 01 model but the way that they’re describing it is this is a new paradigm in in the
0:02:11 past you know the large language models they’re really great you know they’re trained on tons of
0:02:15 data you know they can make great stuff from that but they they can’t really reason about what they’re
0:02:20 seeing right and so they’re saying that this is now a basic reasoning that they kind of attached on
0:02:25 top of those large language models it’s not like from what i can tell so far like it’s not you know
0:02:29 there’s not great use cases for like regular people yeah it’s a little bit better than editing and
0:02:35 encoding a few things but the big deal probably is that now you can it’s almost like in the past
0:02:40 the large language models were like had like half a brain right like almost like the creative side
0:02:44 or something like this and now you have like the logic side is there now too it’s like you put those
0:02:50 two together like it’s amazing with the reasoning side it’s all at inference time like there’s a
0:02:54 whole another side of things that they can improve now we’re like yeah you can just throw more gpu’s
0:02:59 at this and so like yeah although talk about like oh you know we’ve hit a limit and it’s not going to
0:03:04 get better anytime soon like no this is there’s no end in sight now like this is going to continue
0:03:09 to get way better and uh and the people at open ai they’re saying like yeah it’s almost like this is
0:03:15 like the gpt two or three moment where they’ve built this amazing new they’ve had a breakthrough
0:03:20 but consumers can’t really realize that yet there’s not like if you just use it in chat gpt it’s
0:03:23 you’re not gonna like oh my god this is amazing it’s like oh it’s like slightly better at some
0:03:28 things but there’s more going on the site you know behind the hood it’s using what they call
0:03:33 chain of thought reasoning right where basically there was a technique that people used when they
0:03:38 were you know quote unquote prompt engineering called chain of thought reasoning where you would
0:03:43 ask it a question and then within that same prompt you would say think this through think it through
0:03:48 step by step you know and and that was how people would do this chain of thought reasoning to me
0:03:52 what this is is they just sort of bake that in so now you don’t have to ask for that chain of
0:03:58 thought reasoning you don’t have to tell me think this through step by step i guess what i’m still
0:04:05 uncertain about is like how is this new model super different yeah from just doing that sort of
0:04:09 prompting technique i mean the rumors i’ve heard from like friends in silicon valley who know people
0:04:15 there is like it’s based on that idea but there’s more to it than that i mean i get that impression
0:04:19 yeah and so i don’t know we could you know we could guess what it is but like we have no idea
0:04:25 yeah and when you use this new open ai model too it it actually will show like a little sort of
0:04:28 dialogue thing that says like thinking or whatever right it’s got a drop down arrow and if you open
0:04:34 up that drop down arrow it will actually show you sort of what it’s thinking but from what i understand
0:04:40 it’s actually not really showing you a thing it’s almost like yeah it’s almost as if it’s what it’s
0:04:45 thinking is a much longer prompt and then it’s using its own sort of like ai model to give you a
0:04:49 summarized version of what it’s thinking yeah well people were saying oh it’s amazing you actually
0:04:52 finally see what they’re thinking yeah and now we don’t have to worry about like yeah taking over the
0:04:57 world we can we can see what it’s thinking in real time and then opening out was like oh yeah you
0:05:02 know we’re not actually like necessarily showing that to consumers like like we have the capability
0:05:05 to do that now though apparently that they are saying that they have the ability to do that i believe
0:05:09 yeah but they’re not making all of that you know known to like regular people yeah so yeah i think
0:05:14 we’re getting some kind of like edited down version of what they decided to show people right
0:05:18 still pretty cool to see that you know and a lot of the rumors are right we’ve been hearing about this
0:05:24 strawberry right and sam altman trolled everybody by putting a image up on his twitter say oh i love
0:05:29 it killing as well to the guy i don’t know who it was a real person or yeah if it was so there’s
0:05:33 been a lot of talk around strawberry and then this came out so a lot of people were like hey
0:05:38 strawberries finally out right yeah but we’re here at inbound and we’ve been talking to a lot of people
0:05:42 who are deep into ai as well and somebody’s like yeah it’s not even strawberry it’s something else
0:05:47 it’s called it’s actually the orion model which is different than strawberry q star whatever you
0:05:52 want to call it you know yeah it’s hard to know like like also is this gpt 5 or is that something
0:05:56 else too yeah because maybe yeah the the the large language model itself is going to keep getting
0:06:00 better and better maybe they’re still working on that maybe there is another gpt 5 coming yeah
0:06:04 this reasoning thing is another thing yeah now they can improve upon so well the way they announced
0:06:09 it on their blog as well was like we’re resetting the counter i don’t remember the exact wording
0:06:13 right but they’re like we’re resetting this so we’re starting with 01 right so it’s super confusing
0:06:18 with like preview and mini and yeah that’s the other thing right so like the version of 01 that
0:06:24 we’re seeing right now is the preview version all of the benchmark test like if you look at their
0:06:29 blog you’ll see that it outperformed in like phd reasoning and math and science and all of these
0:06:35 like stem related topics right but everything that it’s showing that really really outperformed
0:06:41 is the open ai 01 model the full model yeah we’re just seeing the preview model which if you look at
0:06:47 those same benchmarks they’re better than what we were getting but they’re not like insanely better
0:06:53 they’re not like you know the the phd level they’re showing on these charts right yeah i mean it’s
0:06:56 kind of gets us closer to agi though too right because like open ai had put out there like
0:07:02 tier list of like getting to agi i think number two was reasoning and number three was agents
0:07:06 right and so now they’re saying they have reasoning they’re saying it’s not great but is the basic
0:07:11 foundation right how you would build reasoning into an ai system they’ve nailed that now yeah and
0:07:16 then apparently like a david sacks on the old end podcast he was saying some other day where
0:07:20 apparently the word is that the open ai investors have been told by sam altman
0:07:25 that they’ve nailed agents that it’s working like the new 01 like that they haven’t shown
0:07:30 people publicly yet but in private it’s working and agents are actually going off and doing work
0:07:35 it’s like work that companies would love to have the agents doing for them it’s actually working and
0:07:40 so you know agents in the past like you know we had yo hey akajima for the who did baby agi and
0:07:44 always on and great concept but like a lot of stuff just didn’t work like it would go up and try to
0:07:49 do stuff for you and it would like get confused and like and that’s kind of it yeah and you could
0:07:53 possibly even run up a huge bill for you too because it just keeps trying yeah and so if they’ve
0:07:57 actually nailed that i mean this is it’s gonna be a huge step forward when that comes out so just
0:08:01 for anybody who’s listening that maybe hasn’t listened to like previous episodes the when we
0:08:07 talk about agents and agent is essentially like it’ll do multiple steps right you can give it a
0:08:13 task and it will go and potentially use tools and do multiple steps for you to get to the result
0:08:19 that you wanted so like examples people might give of like an ai agent might be something where i go
0:08:24 and ask it to book my flight for me right and it goes and checks your calendar sees what times you
0:08:29 have available and then it goes to the airline website and sees if there’s any flights that are
0:08:34 available within that window that you want and then it books it for you and then sends you a slack
0:08:39 message to let you know that it booked it right yeah so an agent is like all of like a multi-step
0:08:45 process probably using different api right things done one example took me like if you’re doing in
0:08:49 marketing like doing a marketing report you could say go off and get me this data and like
0:08:53 you know create a pdf and like right you can do a report like that’s something used to take a lot
0:08:58 of time yeah like in theory agents should be able to easily do that kind of stuff yeah yeah well so
0:09:04 so sam altman said i don’t remember if we did a tweet or if it was on the blog post but he basically
0:09:09 said for most cases you still want to use gpt for yeah he said for most of the cases you still
0:09:15 want to go use gpt for you want to use open ai o one if you really need it to think through logic
0:09:21 or do a math problem or solve really complex like things that require more logic right right so that’s
0:09:28 why we wanted to do this episode is to kind of talk through what are some of the the large language
0:09:33 models pros and cons like what what what models should i be using if i need this task done which
0:09:38 models should be i’d be using for this task over here right and so we wanted to start by talking
0:09:44 about chat gpt and this new open ai o one because you probably don’t need o one for most things like
0:09:50 maybe go and ask it some complex questions just to see what is capable of and look at the way it’s
0:09:56 reasoning yeah but for 99% of use cases gpt for is probably going to be better for most of your
0:10:02 use cases but obviously chat gpt isn’t the only game in town right right we’ve got chat gpt we’ve
0:10:08 got claud we’ve got lama we’ve got gemini we’ve got grok we’ve got all of these tools so let’s
0:10:14 talk a little bit about claud for for me this is what i think claud is the best at i think claud
0:10:20 is really really good at like storytelling and i think it’s amazing this is what i use it for more
0:10:26 than anything else is summarization taking a a fairly long document now it does have like some
0:10:31 context token limits where some documents tend to be too big it seems to be based on file size
0:10:36 on claud though like on claud it’ll i don’t know the exact file size but if you go over like 20
0:10:40 megabytes or something you’ll say it’s too big right so if you have a long pdf you can actually
0:10:46 take that pdf run it through a compression tool shrink the file size down and then claud will
0:10:51 read it so i’ve actually downloaded some like pds from archive.org before tried to throw it in the
0:10:55 claud and it would say oh this file is too large and it would be like 20 megabytes yeah so i went on
0:11:01 google and i said pdf compression tool right and i found a free pdf compressor and it would compress
0:11:07 that you know 25 megabyte pdf down into a one megabyte pdf without changing any of the text in
0:11:12 the pdf and then i was able to pull it into claud so if you actually run into that issue with like
0:11:17 a pdf on claud yeah just go and compress it down into a smaller file size and it’ll typically work
0:11:24 but claud is amazing at summarizing documents like taking an archive.org document and just
0:11:28 shrinking it down yeah definitely claud’s a great at writing like it’s definitely the
0:11:33 best llm at writing right now for sure at summarization if i had to guess 01 is probably
0:11:38 better now you think i had to guess but like you know i just we just flew out to boston from you
0:11:42 know me from japan i haven’t had much time to actually like play with it right but you know from
0:11:46 people i trust on twitter and things like that who’ve been trying it out that’s what i’ve been
0:11:50 hearing is like yeah in terms of editing anything that’s complex so i assume like summarization
0:11:55 probably fit into that as well do you know what the context window is of open io one does it have
0:11:59 a limited context window i actually don’t no yeah i’m not sure either we did you know that
0:12:05 producer yeah so the other thing that i really so both claud and chat gpt have these features
0:12:10 where you can kind of create your own little like almost like mini apps inside of them right i don’t
0:12:14 know the better way to describe it but you have custom gpt’s over on chat gpt and then you’ve got
0:12:20 projects over inside of claud right and what this allows you to do is create like custom instructions
0:12:25 for a specific use case that you might use over and over again but also feed it some documents
0:12:30 where you want it to use those documents for context right every time you use it both chat gpt
0:12:35 and claud can do that but like we were talking about this off air claud just has a better user
0:12:42 interface it just feels a lot more into it yeah i mean it’s you know we’ll be right back but first
0:12:46 i want to tell you about another great podcast you’re going to want to listen to it’s called
0:12:52 science of scaling hosted by mark robert and it’s brought to you by the hub spot podcast network
0:12:58 the audio destination for business professionals each week hosts mark robert founding chief revenue
0:13:04 officer at hub spot senior lecturer at harvard business school and co-founder of stage two capital
0:13:10 sits down with the most successful sales leaders in tech to learn the secrets strategies and tactics
0:13:15 to scaling your company’s growth he recently did a great episode called how do you solve for a
0:13:20 siloed marketing and sales and i personally learned a lot from it you’re gonna want to check out the
0:13:29 podcast listen to science of scaling wherever you get your podcasts so opening i actually came up
0:13:33 with the first as far as we know they did yeah you know they they came up with custom instructions
0:13:37 which was great because like okay now you can actually give the you know chat gpt like context
0:13:40 but it was like okay it’s like very generic though because you can’t tell like oh because
0:13:43 sometimes i’m working on the my newsletter sometimes i’m doing twitter or sometimes
0:13:47 personal stuff whatever yeah like you you know you couldn’t give it to that you know you had to
0:13:51 give it all the context for all the stuff yeah and then with a then they came out with the custom
0:13:57 gpt’s and then you can do okay i have one for you know getting in shape i have one for my newsletter
0:14:00 you know all this kind of stuff so but the the user interface was always really awkward and then
0:14:04 there’s like and then there’s gonna be like a public store and then there’s gonna be your revenue
0:14:06 share so confusing i didn’t know they were actually doing the revenue share like on that one
0:14:10 episode we did with ziki we found out like oh they actually are paying people like we do share
0:14:15 revenue but i’m not allowed to tell you about it yeah he also told me you know behind the scenes
0:14:18 that he’s also been pretty good for for legion actually as well like people are just because
0:14:22 there’s a link to his website right so that’s that’s cool but like for like a regular person
0:14:25 claud came out of the projects and i think it’s just it’s so much better just like a whole interface
0:14:30 where it’s like okay here’s the here’s the project and i can put files in it i can put notes in it
0:14:34 whatever to teach it to give it all the context it needs to accomplish something yeah just having
0:14:38 to you know like when chat gpt first came out if you want to do that i’d like to be copying and
0:14:43 pasting stuff into it there’s all this stuff you need to know and here let me copy and paste that in
0:14:47 yeah yeah so like one example just to sort of make it more tangible for people and i’ve shared
0:14:51 this example on past episodes so if you’re hearing this again well i’m just solidifying the idea for
0:14:57 you but one of the ways that i use it lately is for youtube shorts and script writing for youtube
0:15:04 shorts and what i’ve done is i’ve created a custom project inside of claud where i went and found
0:15:09 about 15 different transcripts of shorts that i found on youtube that i thought were really really
0:15:14 good shorts they hooked me in i watched the whole time this this short worked for me it’s got you
0:15:19 know half a million a million views it’s working for other people this is a good short right i went
0:15:25 and took the transcripts from those shorts and i loaded those 15 transcripts into a project inside
0:15:31 of claud and then for the the custom instructions for that prompt i would say i in your knowledge base
0:15:36 i’ve given you a bunch of transcripts of youtube shorts that are really really effective i’m going
0:15:42 to give you a topic when i give you a topic write a short for me write a script for a short for me
0:15:49 that is very similar in style and flow to the information that i gave you so now i can go and
0:15:55 for instance grab a blog post about something that happened in the news right i can grab a blog
0:16:01 goes i wanted i just put published it’s about open ai 01 i went and took the article that explains
0:16:07 open ai 01 from their blog pasted it into this custom project that i made gave no other prompt
0:16:12 i just literally copied and pasted the article in it knows what it needs to do because of the
0:16:18 custom instructions and it wrote me a 59 second script that sounds like a lot of the scripts
0:16:24 that i uploaded it hits all the right beats all the right points yeah and so that’s sort of a
0:16:29 tangible example of like what projects can be used for your workflow that saves you so much time too
0:16:33 right yeah you know before i think it was good that we came out here to inbound like you know
0:16:36 especially for me to get out of japan for a bit and actually talk to people in english i was surprised
0:16:41 how many people like just don’t know like about projects or or even custom gpt’s and how that how
0:16:45 much time that can save you they’re still in the land of like they just go to chat to bt every time
0:16:50 and they’re starting fresh with like it having no context yeah every single time and then they’re
0:16:55 kind of confused by why you can’t do more and it’s like yeah if you provide it more context it can
0:17:00 do a whole lot more than you’re currently thinking yes even with the current abilities of it so
0:17:03 and so i think it’s good that people like learn that and like so yeah claud’s the best for custom
0:17:08 projects but you can still do it with chat to bt most people use chat bt yeah you can do it with
0:17:13 custom gpt yeah you don’t make that public either keep it private and it’s just for you yeah same
0:17:17 same with the claud projects you can actually i i believe you can share them i don’t it’s not public
0:17:20 like they don’t have a marketplace yeah but you can share a link and let other people use your
0:17:24 projects i believe i believe i believe you give me sure like the entire team yeah i believe so
0:17:28 the other really cool thing about claud is their artifacts yeah so if you are going to use it for
0:17:33 code it will actually show you the code in the same window and actually let you run the code if
0:17:37 it’s something that’ll run in a browser so if it’s like a java script or a python code or something
0:17:42 that can actually be executed in the browser right it will actually let you run it side by side with
0:17:47 your chat so you have your chat window on the left side and on the right side you’ve actually got
0:17:53 your code and there’s usually a little like toggle switch to actually view the code and if the website
0:17:58 populates on the right side and you can see oh that link is broken or this needs to be centered
0:18:03 you go to the left side of the chat box and say hey center the headline center the call the action
0:18:07 button or whatever and i always thought that was really cool to not have to like open it in a new
0:18:11 tab see what it looks like jump back it’s just all right there on the single window yeah they’re
0:18:15 really nailing it like on the user interface right because like why does open AI not have that
0:18:20 right like they should have that yeah like they were there at first and they just missed it and the
0:18:25 same thing like you know i mean i’m hearing that 01 or 01 preview is the best at coding yeah and like
0:18:29 well if they had the thing where you could see the artifacts and actually see it live that would
0:18:33 show to so many more people there’s like a lot of great engineers on twitter talking about you know
0:18:38 who’ve been like really skeptical of AI and they’ve now tried 01 preview and they’re like okay it’s
0:18:42 not amazing yet but like everyone was over hyping things before like it would give you code that
0:18:46 sometimes worked sometimes didn’t yeah this is like a decent not a great engineer but it’s like
0:18:52 a decent junior level engineer yeah yeah and that’s a huge step forward and especially now that you
0:18:56 know open AI is saying it’s going to get better and better faster and faster now so yeah yeah that’s
0:19:01 okay well and then and then we’ve also got google right so google has gemini yeah and one of the
0:19:05 beautiful parts about gentleman i is it’s got this monstrous context window i can’t remember
0:19:08 it’s a million or if it’s up to 10 million i know at one point they keep increasing it says like it’s
0:19:12 like it’s at some point it’s like it’s going i know at one point they said it’s going to be 10
0:19:16 million but when they launched it was a million i don’t know where it’s at now it might be at
0:19:20 one two million and they might have already bumped into 10 million i don’t remember it would like
0:19:24 ram with computers and like it’s okay yeah one gig two gig four or whatever and just to put that
0:19:31 into context right so so a 10 million context window means that you can upload about 7.5 million
0:19:36 words between between what you give it and what you get back the combination of the two can be
0:19:42 about 7.5 million words that’s like all of the harry potter books all of the game of thrones
0:19:48 books the bible all of that combined all into a single like yeah you can put all of that in there
0:19:55 all at once right and so when it comes to what is google gemini good at is the largest in terms of
0:20:01 context window and how much data you can upload into it but one thing that google did recently
0:20:07 with gemini that i think is really really cool is the the notebook lm that’s that’s the one where
0:20:12 you can actually upload all of these documents and it will create an audio podcast for you
0:20:17 out of these documents have you seen that yet i haven’t no my gosh oh you’re going to play with
0:20:22 this so notebook lm yeah you can go to it i believe i don’t know if you have to be a paid member of
0:20:29 gemini or not but you can go into this notebook lm and it’s you can drop files into a folder so
0:20:36 you can go to like archive.org download the most complex research paper you’ve ever seen
0:20:42 that you have no idea what it means right pull that pdf throw it into this notebook lm tool yeah
0:20:50 and notebook lm will actually generate an audio podcast for you that explains the pdf and it’s
0:20:54 it’s two hosts there’s a male and a female host their ai generated voices yeah and they have a
0:20:57 conversation back and forth kind of like the conversation we’re having now right they have
0:21:03 a conversation back and forth that explains the pdf in terms that like anybody can understand
0:21:08 and because they have such a large context window you can throw a ton of documents in here
0:21:14 right you can you can go and find like we can go and throw the blog post about chat gpt in there a
0:21:19 blog post explaining claud in there a blog post explaining grok a blog post explaining gemini
0:21:26 throw all of that data in there yeah and then it’ll it’ll create a podcast out of that information
0:21:31 that explains the pros and the cons and the the differences between the two yeah it seems like
0:21:35 google’s doing a lot of cool stuff it’s just like people don’t know about it and even me like i i
0:21:38 know about some of it but it’s like it’s hard to get the habit like i have the habit of using
0:21:44 chat gpt and even you know i use claud a bit now it’s so hard like there’s so many of them yeah yeah
0:21:50 yeah so i mean like i i have a feeling we’re going to start hearing a lot a lot of ai generated
0:21:56 podcasts because of it because you anybody can go and take the latest like 10 news articles in the
0:22:03 ai world throw them all into notebook lm and then have it spit out a podcast for you explaining
0:22:07 what’s going on in all of these news articles oh no our buddy is already doing the tit ai podcast
0:22:14 and then if you want to take it to a next level too you can create some like custom characters
0:22:19 inside of like mid journey or or you know flux or Leonardo or whatever ai image you use create
0:22:26 some images of some characters throw those into like a hey gen or a di and have them actually
0:22:31 talking so you have you can make video content of two people having a podcast conversation
0:22:36 based on all of the data that you uploaded and that’s all totally possible right now like today
0:22:41 with current the next level like vtuber or something too like there’s yeah except the vtuber
0:22:46 doesn’t even exist it’s all graduated with ai it’s it’s really wild notebook lm is one of the
0:22:49 things i’ve had a lot of fun playing with i think once you actually go mess with it you’ll be like
0:22:54 oh damn this is cool okay that’s what jimna is good at huge context and and making podcasts yeah
0:23:03 replacing us yeah the disclaimer this is actually not as talking this is yeah surprise this is
0:23:10 notebook lm all along you know and also jim and i does have their own version of gp custom gpt’s
0:23:17 and projects they call them gems but they’re not nearly as robust you can’t do as much with them
0:23:21 they don’t seem to connect to like apis or use tools or anything like that but you can give it
0:23:27 your own context and your own custom instructions similar to you can with projects and and with
0:23:34 custom gpt’s i haven’t played with gems at all yet because i have my needs met with claud and
0:23:39 chat gpt already but they have that as well if that’s something you’re interested in and then we
0:23:46 also wanted to talk quickly about llama so llama is open source right and llama is it doesn’t really
0:23:53 excel in any of the areas we’ve talked about right like claud sonnet 3.5 is going to be better than
0:23:59 the outputs for llama for the most part same with the latest models of chat gpt same with jim and i
0:24:05 all of those closed off models they’re going to be a little bit better but if you really really
0:24:10 want like your data to be secure you want to run it on your own hard drive you don’t want it to be
0:24:14 connected to the internet you don’t want your data going to open ai you don’t want your data going to
0:24:19 google or anthropic or any of these companies that’s where tools like llama come into play because
0:24:26 you can install them locally on your own computer and they’re pretty good i mean for 90 plus percent
0:24:30 of use cases for people they’re going to be good enough right you know and so that’s really where
0:24:35 llama comes into play the other thing about llama is because it’s open source people can build off
0:24:40 of it they can iterate off of it they can go and make their own sort of fine tunes of it and you
0:24:47 know tweak it and train it and sort of mold it so if you wanted an uncensored llama you can get
0:24:50 there if you wanted to you actually ever use it though if you just only tried it because i’ve only
0:24:55 tried it i haven’t i’ve never built a habit of actually using it so i use a llama a little bit
0:25:00 but it’s only because i have my meta ray band sunglasses and llama is built into it so i’ve
0:25:05 used llama like built into the sunglasses i’m glad it exists though like i’ve said before
0:25:09 like i hate the idea of everything being open ai and you know they’re like collaborating with the
0:25:13 us government as well it’s like in the future like the idea that like all of our intelligence
0:25:18 is in control by one company right it’s kind of scary so i’m glad it exists yeah and then and then
0:25:25 you know finally we’ve got grok right and grok again i don’t know if it’s like really better
0:25:29 than any of the other ones at almost anything well it has access to twitter data so that’s
0:25:34 that’s one thing yeah so it’s got the twitter data and it’s completely uncensored like completely
0:25:38 like well like nine point nine nine yeah i mean i’ve asked it questions that i thought for sure
0:25:43 there’s no way it was going to answer yeah and it answered it for me like broke down recipes it should
0:25:51 not be giving me things like that okay and now you’re like a fbi watcher but it is very very
0:25:56 uncensored it’s got a really cool image generation model built into it called flux one which is capable
0:26:02 of making really really realistic images yeah and also pretty uncensored i’m not gonna say fully
0:26:08 uncensored because it’s not going to do any like uh like super adult content right but it’s uncensored
0:26:13 in the terms that you can prompt anybody’s names like celebrities and you know like crazy political
0:26:17 memes or whatever right that’s one of the things people aren’t using it for a lot yeah which by
0:26:23 the way is illegal in california as of this week did you did you see that yeah so gabin Newsom
0:26:29 in california just signed a bill basically making making like political memes that were generated
0:26:32 with ai illegal in california but we don’t need to get there we don’t we don’t need to get i have a
0:26:35 feeling that’s gonna go to the supreme court but anyways yeah yeah that that’s gonna be interesting
0:26:41 to watch play out i mean elon must move to austin now so he’s not really under california rules
0:26:45 but as soon as that law came out elon went and posted some sort of crazy meme and said oh this
0:26:50 this would suck if this ended up being seen in california or something right so you know elon
0:26:56 he’s a troll yeah um yeah i mean so so grok is it’s it’s gotten better though like the fact like
0:27:00 they started for like nothing like what was like a year ago or it wasn’t that long and it was like
0:27:04 people are like oh there’s no way he’s gonna catch up it’s like well he hasn’t caught up but like
0:27:08 it’s pretty decent now yeah like you know i’m not going to use it over chat to bt obviously but it’s
0:27:12 it’s decent so what you’re talking about this before you hit record but what sort of like
0:27:18 use cases would you use it for like in terms of it like hunting twitter profiles for you yeah i’ve
0:27:21 used it a little bit like this i think it was robert scope it was someone like that who shared
0:27:26 how they were using it uh that you can like pull up data about users right like you can pull up like
0:27:30 like what kind of tweets do they do what are their top tweets just anything you know i’m not using
0:27:34 it a lot for that you know use case i don’t really have a huge need for that yeah but it works for
0:27:38 that it’s like it’s it’s the best for that right now yeah i mean also you can pull up things about
0:27:42 news too right because it’s got access to what people are tweeting about yeah information about
0:27:48 something that’s currently hot and happening right now is probably the best for that yeah but
0:27:53 there is also some downside so as well right like well yeah if people are posting memes as if
0:27:57 it’s real news it might think it’s real news and feed it back to you right so there’s that as well
0:28:02 and then the the very last one i want to touch on really quickly because i think when it when it
0:28:08 comes to research right when you want to like deep dive and learn on a topic or you know find
0:28:13 information about something specific it actually needs to search the web i still think perplexity
0:28:19 hands down is the best yeah now perplexity is model agnostic if you have a pro membership
0:28:24 for perplexity you can get in there and choose i want to use gpt for they actually just added open
0:28:29 ai o o one in there as well as one of the models you can use you can pick claud you can pick lama
0:28:34 you can pick any of these models and it will use whatever your model of choice is but what it’ll
0:28:39 do is it’ll search the internet find information for you and then once it finds the information
0:28:44 it uses its large language model to help sort of summarize and make the information that it
0:28:49 gathered for you a little bit easier to digest yeah i mean perplexity i mean i mean there was a
0:28:53 reason that you know we got air vans around about the ceo to come on he’s like one of our first two
0:28:57 guests i guess yeah yeah and that was the reason that i wanted to get him on because like yeah i’ve
0:29:02 been using perplexity for like since the early days of it’s like it’s been one of the most useful
0:29:06 things and like early on most people didn’t know about it and people in silicon valley were talking
0:29:09 about it but like outside of that it was like oh it was everything was chat to bt i’m like
0:29:13 people are kind of sleeping on perplexity it’s like pretty damn good yeah yeah yeah no i love
0:29:19 perplexity i mean honestly when it comes to using ai just yeah the ai tool that i use the most
0:29:24 more than any other ai tool right now for me it’s perplexity yeah i go to perplexity more than any
0:29:29 other ai tool because i just get random ideas like i’ll be i’ll be walking and i’ll see like a
0:29:33 tall building and i’ll think it’s like a fascinating architecture and i’ll just be like
0:29:38 who built this building and why does it look so cool or you know something really random like i
0:29:43 just asked so many random questions to perplexity it like feeds my curiosity it’s amazing for podcast
0:29:48 too right like so we you know here at inbound there was a event yesterday where like a bunch of
0:29:52 podcasts hosts were there and i talked to a few of them and they didn’t know about this they didn’t
0:29:55 know they’re like oh yeah when you’re when you’re interviewing somebody perplexity is by far the
0:29:58 best way to like pull up notes about what you should talk to them about yeah you just have to be
0:30:03 careful because sometimes it will pull up like linked in profiles of like the complete wrong person
0:30:08 like if you start asking questions about matt wolf yeah well there’s a professional golfer named
0:30:12 matt wolf also right and so talking about golf and you like what it’ll give you a mix right like
0:30:17 i’ve had it summarized and it’ll say you know matt wolf co-host the next wave podcast he’s got a
0:30:23 popular youtube channel he’s you know his handicap at golf is like and it will it’ll start to like
0:30:29 blend the two so you probably want to double check sometimes but for the most part it’s pretty good
0:30:35 right it’s pretty dang good but you know to go ahead and wrap this one up i want to kind of
0:30:41 let’s like sort of rapid fire like use cases best tool for that use case right so like
0:30:49 copywriting and storytelling for me i’m probably going to claud yeah claud yeah for coding for me
0:30:55 i’m probably still going to claud but maybe open aio one i haven’t tried it for coding yet myself
0:31:00 yeah i mean i i think i code more than you right so like i like i’m more you know for you i think
0:31:04 when you can use claud you can actually just see it right there right which is obviously
0:31:09 awesome yeah but you know as someone who codes like for me like you know o one you know preview
0:31:13 at least from what i’ve heard from people i trust if it’s better than yeah i’d definitely be going
0:31:17 with o one yeah yeah i’d be using in some like cursor yeah cursors for sure cursor is great
0:31:22 it’s a it’s a fork of visual studio code with like ai built into it yeah where you can like
0:31:26 read your whole code base and have it basically has context of your entire code base when you’re
0:31:30 working on something yeah amazing yeah i mean since open ai o one came out i’ve literally been traveling
0:31:36 the entire month so i have not had like a a focus session to play with it so my answer might change
0:31:43 on that one yeah um you want something completely uncensored anything goes i think groc is the
0:31:49 obvious answer for that one right yeah um also really really realistic images yeah great memes
0:31:53 yeah good for memes great for making memes great like the images it it’ll generate like
0:31:58 really realistic pictures of of people that either exist or don’t exist right it’s right really good
0:32:04 at that with flux one built into it yeah so math science logic that kind of stuff we’re looking
0:32:09 at open ai o one that’s really the only one that’s going to do that really well right now right right
0:32:14 for research for me perplexity like i said that’s the one i think we’re the same on most of these
0:32:18 like a major yeah disagreements and then if you need super super long context you need to upload
0:32:22 all the harry potter books and all the lord of the rings books and all the game of thrones books
0:32:28 and you need to find a good comparison between all of them you’re going to go to gemini and then
0:32:33 also gemini is really good to create this podcast yeah we’re for creating a podcast and replacing
0:32:39 our jobs yes yeah so that’s kind of the the breakdown i guess of like what these different
0:32:46 models are good at and what we use them for i think that about i think that covers everything
0:32:50 so you know if you’re you’re listening to this podcast and you like stuff like this
0:32:54 make sure you subscribe to us on youtube we’re going to keep on putting out videos like this
0:32:57 we’re going to get a lot more tactical we’re going to give you a lot more use cases a lot
0:33:02 more ways to actually implement this ai in your business so subscribe on youtube if you if you
0:33:07 prefer audio we’re wherever you find podcasts so make sure you subscribe on all of those platforms
0:33:20 and thank you so much for tuning in with us today yeah thank you
0:33:21 you
0:33:22 you
0:33:23 you
0:33:25 you
0:33:27 you
0:00:15 that i use the most more than any other ai tool hey welcome to the next wave podcast i’m matt wolf
0:00:21 i’m here with nathan lands and today we’re going to talk a little bit about the various large language
0:00:26 models that are out there i feel like one of the things that we hear a lot is like okay you’ve got
0:00:32 claud you’ve got chat gpt you’ve got llama you’ve got perplexity you’ve got grok you’ve got
0:00:38 you know the list goes on and on but the reality is some of these tools are really good at some
0:00:43 things some of them are really bad at things and so we wanted to spend a little bit of time on this
0:00:48 episode just kind of deep diving in like what’s this good for what’s this good for if i was doing
0:00:56 copywriting what would i use if i’m using research what would i use when all your marketing team does
0:01:01 is put out fires they burn out but with hub spot they can achieve their best results without the
0:01:08 stress tap into hub spots collection of ai tools breeze to pinpoint leads capture attention
0:01:14 and access all your data in one place keep your marketers cool and your campaign results hotter
0:01:24 than ever visit hub spot dot com slash marketers to learn more so just something i think it should
0:01:29 be a really valuable resource for people just hey i need to do this let me go double check with
0:01:36 that episode of the next wave on how i actually should do that like what tool should i be using
0:01:40 to go do that so that’s what this episode’s all about we don’t have a guest today it’s just me
0:01:46 and nathan chatting from our own experience about these tools just getting into it here one of the
0:01:52 biggest news in the ai world right now is from chat gpt from open ai right they just released
0:01:58 their new open ai 01 model maybe do you want to kind of break down a little bit i mean i mean some of
0:02:02 it we don’t fully know how it’s working yet there’s a lot of like debate about what’s actually going
0:02:07 on with the new 01 model but the way that they’re describing it is this is a new paradigm in in the
0:02:11 past you know the large language models they’re really great you know they’re trained on tons of
0:02:15 data you know they can make great stuff from that but they they can’t really reason about what they’re
0:02:20 seeing right and so they’re saying that this is now a basic reasoning that they kind of attached on
0:02:25 top of those large language models it’s not like from what i can tell so far like it’s not you know
0:02:29 there’s not great use cases for like regular people yeah it’s a little bit better than editing and
0:02:35 encoding a few things but the big deal probably is that now you can it’s almost like in the past
0:02:40 the large language models were like had like half a brain right like almost like the creative side
0:02:44 or something like this and now you have like the logic side is there now too it’s like you put those
0:02:50 two together like it’s amazing with the reasoning side it’s all at inference time like there’s a
0:02:54 whole another side of things that they can improve now we’re like yeah you can just throw more gpu’s
0:02:59 at this and so like yeah although talk about like oh you know we’ve hit a limit and it’s not going to
0:03:04 get better anytime soon like no this is there’s no end in sight now like this is going to continue
0:03:09 to get way better and uh and the people at open ai they’re saying like yeah it’s almost like this is
0:03:15 like the gpt two or three moment where they’ve built this amazing new they’ve had a breakthrough
0:03:20 but consumers can’t really realize that yet there’s not like if you just use it in chat gpt it’s
0:03:23 you’re not gonna like oh my god this is amazing it’s like oh it’s like slightly better at some
0:03:28 things but there’s more going on the site you know behind the hood it’s using what they call
0:03:33 chain of thought reasoning right where basically there was a technique that people used when they
0:03:38 were you know quote unquote prompt engineering called chain of thought reasoning where you would
0:03:43 ask it a question and then within that same prompt you would say think this through think it through
0:03:48 step by step you know and and that was how people would do this chain of thought reasoning to me
0:03:52 what this is is they just sort of bake that in so now you don’t have to ask for that chain of
0:03:58 thought reasoning you don’t have to tell me think this through step by step i guess what i’m still
0:04:05 uncertain about is like how is this new model super different yeah from just doing that sort of
0:04:09 prompting technique i mean the rumors i’ve heard from like friends in silicon valley who know people
0:04:15 there is like it’s based on that idea but there’s more to it than that i mean i get that impression
0:04:19 yeah and so i don’t know we could you know we could guess what it is but like we have no idea
0:04:25 yeah and when you use this new open ai model too it it actually will show like a little sort of
0:04:28 dialogue thing that says like thinking or whatever right it’s got a drop down arrow and if you open
0:04:34 up that drop down arrow it will actually show you sort of what it’s thinking but from what i understand
0:04:40 it’s actually not really showing you a thing it’s almost like yeah it’s almost as if it’s what it’s
0:04:45 thinking is a much longer prompt and then it’s using its own sort of like ai model to give you a
0:04:49 summarized version of what it’s thinking yeah well people were saying oh it’s amazing you actually
0:04:52 finally see what they’re thinking yeah and now we don’t have to worry about like yeah taking over the
0:04:57 world we can we can see what it’s thinking in real time and then opening out was like oh yeah you
0:05:02 know we’re not actually like necessarily showing that to consumers like like we have the capability
0:05:05 to do that now though apparently that they are saying that they have the ability to do that i believe
0:05:09 yeah but they’re not making all of that you know known to like regular people yeah so yeah i think
0:05:14 we’re getting some kind of like edited down version of what they decided to show people right
0:05:18 still pretty cool to see that you know and a lot of the rumors are right we’ve been hearing about this
0:05:24 strawberry right and sam altman trolled everybody by putting a image up on his twitter say oh i love
0:05:29 it killing as well to the guy i don’t know who it was a real person or yeah if it was so there’s
0:05:33 been a lot of talk around strawberry and then this came out so a lot of people were like hey
0:05:38 strawberries finally out right yeah but we’re here at inbound and we’ve been talking to a lot of people
0:05:42 who are deep into ai as well and somebody’s like yeah it’s not even strawberry it’s something else
0:05:47 it’s called it’s actually the orion model which is different than strawberry q star whatever you
0:05:52 want to call it you know yeah it’s hard to know like like also is this gpt 5 or is that something
0:05:56 else too yeah because maybe yeah the the the large language model itself is going to keep getting
0:06:00 better and better maybe they’re still working on that maybe there is another gpt 5 coming yeah
0:06:04 this reasoning thing is another thing yeah now they can improve upon so well the way they announced
0:06:09 it on their blog as well was like we’re resetting the counter i don’t remember the exact wording
0:06:13 right but they’re like we’re resetting this so we’re starting with 01 right so it’s super confusing
0:06:18 with like preview and mini and yeah that’s the other thing right so like the version of 01 that
0:06:24 we’re seeing right now is the preview version all of the benchmark test like if you look at their
0:06:29 blog you’ll see that it outperformed in like phd reasoning and math and science and all of these
0:06:35 like stem related topics right but everything that it’s showing that really really outperformed
0:06:41 is the open ai 01 model the full model yeah we’re just seeing the preview model which if you look at
0:06:47 those same benchmarks they’re better than what we were getting but they’re not like insanely better
0:06:53 they’re not like you know the the phd level they’re showing on these charts right yeah i mean it’s
0:06:56 kind of gets us closer to agi though too right because like open ai had put out there like
0:07:02 tier list of like getting to agi i think number two was reasoning and number three was agents
0:07:06 right and so now they’re saying they have reasoning they’re saying it’s not great but is the basic
0:07:11 foundation right how you would build reasoning into an ai system they’ve nailed that now yeah and
0:07:16 then apparently like a david sacks on the old end podcast he was saying some other day where
0:07:20 apparently the word is that the open ai investors have been told by sam altman
0:07:25 that they’ve nailed agents that it’s working like the new 01 like that they haven’t shown
0:07:30 people publicly yet but in private it’s working and agents are actually going off and doing work
0:07:35 it’s like work that companies would love to have the agents doing for them it’s actually working and
0:07:40 so you know agents in the past like you know we had yo hey akajima for the who did baby agi and
0:07:44 always on and great concept but like a lot of stuff just didn’t work like it would go up and try to
0:07:49 do stuff for you and it would like get confused and like and that’s kind of it yeah and you could
0:07:53 possibly even run up a huge bill for you too because it just keeps trying yeah and so if they’ve
0:07:57 actually nailed that i mean this is it’s gonna be a huge step forward when that comes out so just
0:08:01 for anybody who’s listening that maybe hasn’t listened to like previous episodes the when we
0:08:07 talk about agents and agent is essentially like it’ll do multiple steps right you can give it a
0:08:13 task and it will go and potentially use tools and do multiple steps for you to get to the result
0:08:19 that you wanted so like examples people might give of like an ai agent might be something where i go
0:08:24 and ask it to book my flight for me right and it goes and checks your calendar sees what times you
0:08:29 have available and then it goes to the airline website and sees if there’s any flights that are
0:08:34 available within that window that you want and then it books it for you and then sends you a slack
0:08:39 message to let you know that it booked it right yeah so an agent is like all of like a multi-step
0:08:45 process probably using different api right things done one example took me like if you’re doing in
0:08:49 marketing like doing a marketing report you could say go off and get me this data and like
0:08:53 you know create a pdf and like right you can do a report like that’s something used to take a lot
0:08:58 of time yeah like in theory agents should be able to easily do that kind of stuff yeah yeah well so
0:09:04 so sam altman said i don’t remember if we did a tweet or if it was on the blog post but he basically
0:09:09 said for most cases you still want to use gpt for yeah he said for most of the cases you still
0:09:15 want to go use gpt for you want to use open ai o one if you really need it to think through logic
0:09:21 or do a math problem or solve really complex like things that require more logic right right so that’s
0:09:28 why we wanted to do this episode is to kind of talk through what are some of the the large language
0:09:33 models pros and cons like what what what models should i be using if i need this task done which
0:09:38 models should be i’d be using for this task over here right and so we wanted to start by talking
0:09:44 about chat gpt and this new open ai o one because you probably don’t need o one for most things like
0:09:50 maybe go and ask it some complex questions just to see what is capable of and look at the way it’s
0:09:56 reasoning yeah but for 99% of use cases gpt for is probably going to be better for most of your
0:10:02 use cases but obviously chat gpt isn’t the only game in town right right we’ve got chat gpt we’ve
0:10:08 got claud we’ve got lama we’ve got gemini we’ve got grok we’ve got all of these tools so let’s
0:10:14 talk a little bit about claud for for me this is what i think claud is the best at i think claud
0:10:20 is really really good at like storytelling and i think it’s amazing this is what i use it for more
0:10:26 than anything else is summarization taking a a fairly long document now it does have like some
0:10:31 context token limits where some documents tend to be too big it seems to be based on file size
0:10:36 on claud though like on claud it’ll i don’t know the exact file size but if you go over like 20
0:10:40 megabytes or something you’ll say it’s too big right so if you have a long pdf you can actually
0:10:46 take that pdf run it through a compression tool shrink the file size down and then claud will
0:10:51 read it so i’ve actually downloaded some like pds from archive.org before tried to throw it in the
0:10:55 claud and it would say oh this file is too large and it would be like 20 megabytes yeah so i went on
0:11:01 google and i said pdf compression tool right and i found a free pdf compressor and it would compress
0:11:07 that you know 25 megabyte pdf down into a one megabyte pdf without changing any of the text in
0:11:12 the pdf and then i was able to pull it into claud so if you actually run into that issue with like
0:11:17 a pdf on claud yeah just go and compress it down into a smaller file size and it’ll typically work
0:11:24 but claud is amazing at summarizing documents like taking an archive.org document and just
0:11:28 shrinking it down yeah definitely claud’s a great at writing like it’s definitely the
0:11:33 best llm at writing right now for sure at summarization if i had to guess 01 is probably
0:11:38 better now you think i had to guess but like you know i just we just flew out to boston from you
0:11:42 know me from japan i haven’t had much time to actually like play with it right but you know from
0:11:46 people i trust on twitter and things like that who’ve been trying it out that’s what i’ve been
0:11:50 hearing is like yeah in terms of editing anything that’s complex so i assume like summarization
0:11:55 probably fit into that as well do you know what the context window is of open io one does it have
0:11:59 a limited context window i actually don’t no yeah i’m not sure either we did you know that
0:12:05 producer yeah so the other thing that i really so both claud and chat gpt have these features
0:12:10 where you can kind of create your own little like almost like mini apps inside of them right i don’t
0:12:14 know the better way to describe it but you have custom gpt’s over on chat gpt and then you’ve got
0:12:20 projects over inside of claud right and what this allows you to do is create like custom instructions
0:12:25 for a specific use case that you might use over and over again but also feed it some documents
0:12:30 where you want it to use those documents for context right every time you use it both chat gpt
0:12:35 and claud can do that but like we were talking about this off air claud just has a better user
0:12:42 interface it just feels a lot more into it yeah i mean it’s you know we’ll be right back but first
0:12:46 i want to tell you about another great podcast you’re going to want to listen to it’s called
0:12:52 science of scaling hosted by mark robert and it’s brought to you by the hub spot podcast network
0:12:58 the audio destination for business professionals each week hosts mark robert founding chief revenue
0:13:04 officer at hub spot senior lecturer at harvard business school and co-founder of stage two capital
0:13:10 sits down with the most successful sales leaders in tech to learn the secrets strategies and tactics
0:13:15 to scaling your company’s growth he recently did a great episode called how do you solve for a
0:13:20 siloed marketing and sales and i personally learned a lot from it you’re gonna want to check out the
0:13:29 podcast listen to science of scaling wherever you get your podcasts so opening i actually came up
0:13:33 with the first as far as we know they did yeah you know they they came up with custom instructions
0:13:37 which was great because like okay now you can actually give the you know chat gpt like context
0:13:40 but it was like okay it’s like very generic though because you can’t tell like oh because
0:13:43 sometimes i’m working on the my newsletter sometimes i’m doing twitter or sometimes
0:13:47 personal stuff whatever yeah like you you know you couldn’t give it to that you know you had to
0:13:51 give it all the context for all the stuff yeah and then with a then they came out with the custom
0:13:57 gpt’s and then you can do okay i have one for you know getting in shape i have one for my newsletter
0:14:00 you know all this kind of stuff so but the the user interface was always really awkward and then
0:14:04 there’s like and then there’s gonna be like a public store and then there’s gonna be your revenue
0:14:06 share so confusing i didn’t know they were actually doing the revenue share like on that one
0:14:10 episode we did with ziki we found out like oh they actually are paying people like we do share
0:14:15 revenue but i’m not allowed to tell you about it yeah he also told me you know behind the scenes
0:14:18 that he’s also been pretty good for for legion actually as well like people are just because
0:14:22 there’s a link to his website right so that’s that’s cool but like for like a regular person
0:14:25 claud came out of the projects and i think it’s just it’s so much better just like a whole interface
0:14:30 where it’s like okay here’s the here’s the project and i can put files in it i can put notes in it
0:14:34 whatever to teach it to give it all the context it needs to accomplish something yeah just having
0:14:38 to you know like when chat gpt first came out if you want to do that i’d like to be copying and
0:14:43 pasting stuff into it there’s all this stuff you need to know and here let me copy and paste that in
0:14:47 yeah yeah so like one example just to sort of make it more tangible for people and i’ve shared
0:14:51 this example on past episodes so if you’re hearing this again well i’m just solidifying the idea for
0:14:57 you but one of the ways that i use it lately is for youtube shorts and script writing for youtube
0:15:04 shorts and what i’ve done is i’ve created a custom project inside of claud where i went and found
0:15:09 about 15 different transcripts of shorts that i found on youtube that i thought were really really
0:15:14 good shorts they hooked me in i watched the whole time this this short worked for me it’s got you
0:15:19 know half a million a million views it’s working for other people this is a good short right i went
0:15:25 and took the transcripts from those shorts and i loaded those 15 transcripts into a project inside
0:15:31 of claud and then for the the custom instructions for that prompt i would say i in your knowledge base
0:15:36 i’ve given you a bunch of transcripts of youtube shorts that are really really effective i’m going
0:15:42 to give you a topic when i give you a topic write a short for me write a script for a short for me
0:15:49 that is very similar in style and flow to the information that i gave you so now i can go and
0:15:55 for instance grab a blog post about something that happened in the news right i can grab a blog
0:16:01 goes i wanted i just put published it’s about open ai 01 i went and took the article that explains
0:16:07 open ai 01 from their blog pasted it into this custom project that i made gave no other prompt
0:16:12 i just literally copied and pasted the article in it knows what it needs to do because of the
0:16:18 custom instructions and it wrote me a 59 second script that sounds like a lot of the scripts
0:16:24 that i uploaded it hits all the right beats all the right points yeah and so that’s sort of a
0:16:29 tangible example of like what projects can be used for your workflow that saves you so much time too
0:16:33 right yeah you know before i think it was good that we came out here to inbound like you know
0:16:36 especially for me to get out of japan for a bit and actually talk to people in english i was surprised
0:16:41 how many people like just don’t know like about projects or or even custom gpt’s and how that how
0:16:45 much time that can save you they’re still in the land of like they just go to chat to bt every time
0:16:50 and they’re starting fresh with like it having no context yeah every single time and then they’re
0:16:55 kind of confused by why you can’t do more and it’s like yeah if you provide it more context it can
0:17:00 do a whole lot more than you’re currently thinking yes even with the current abilities of it so
0:17:03 and so i think it’s good that people like learn that and like so yeah claud’s the best for custom
0:17:08 projects but you can still do it with chat to bt most people use chat bt yeah you can do it with
0:17:13 custom gpt yeah you don’t make that public either keep it private and it’s just for you yeah same
0:17:17 same with the claud projects you can actually i i believe you can share them i don’t it’s not public
0:17:20 like they don’t have a marketplace yeah but you can share a link and let other people use your
0:17:24 projects i believe i believe i believe you give me sure like the entire team yeah i believe so
0:17:28 the other really cool thing about claud is their artifacts yeah so if you are going to use it for
0:17:33 code it will actually show you the code in the same window and actually let you run the code if
0:17:37 it’s something that’ll run in a browser so if it’s like a java script or a python code or something
0:17:42 that can actually be executed in the browser right it will actually let you run it side by side with
0:17:47 your chat so you have your chat window on the left side and on the right side you’ve actually got
0:17:53 your code and there’s usually a little like toggle switch to actually view the code and if the website
0:17:58 populates on the right side and you can see oh that link is broken or this needs to be centered
0:18:03 you go to the left side of the chat box and say hey center the headline center the call the action
0:18:07 button or whatever and i always thought that was really cool to not have to like open it in a new
0:18:11 tab see what it looks like jump back it’s just all right there on the single window yeah they’re
0:18:15 really nailing it like on the user interface right because like why does open AI not have that
0:18:20 right like they should have that yeah like they were there at first and they just missed it and the
0:18:25 same thing like you know i mean i’m hearing that 01 or 01 preview is the best at coding yeah and like
0:18:29 well if they had the thing where you could see the artifacts and actually see it live that would
0:18:33 show to so many more people there’s like a lot of great engineers on twitter talking about you know
0:18:38 who’ve been like really skeptical of AI and they’ve now tried 01 preview and they’re like okay it’s
0:18:42 not amazing yet but like everyone was over hyping things before like it would give you code that
0:18:46 sometimes worked sometimes didn’t yeah this is like a decent not a great engineer but it’s like
0:18:52 a decent junior level engineer yeah yeah and that’s a huge step forward and especially now that you
0:18:56 know open AI is saying it’s going to get better and better faster and faster now so yeah yeah that’s
0:19:01 okay well and then and then we’ve also got google right so google has gemini yeah and one of the
0:19:05 beautiful parts about gentleman i is it’s got this monstrous context window i can’t remember
0:19:08 it’s a million or if it’s up to 10 million i know at one point they keep increasing it says like it’s
0:19:12 like it’s at some point it’s like it’s going i know at one point they said it’s going to be 10
0:19:16 million but when they launched it was a million i don’t know where it’s at now it might be at
0:19:20 one two million and they might have already bumped into 10 million i don’t remember it would like
0:19:24 ram with computers and like it’s okay yeah one gig two gig four or whatever and just to put that
0:19:31 into context right so so a 10 million context window means that you can upload about 7.5 million
0:19:36 words between between what you give it and what you get back the combination of the two can be
0:19:42 about 7.5 million words that’s like all of the harry potter books all of the game of thrones
0:19:48 books the bible all of that combined all into a single like yeah you can put all of that in there
0:19:55 all at once right and so when it comes to what is google gemini good at is the largest in terms of
0:20:01 context window and how much data you can upload into it but one thing that google did recently
0:20:07 with gemini that i think is really really cool is the the notebook lm that’s that’s the one where
0:20:12 you can actually upload all of these documents and it will create an audio podcast for you
0:20:17 out of these documents have you seen that yet i haven’t no my gosh oh you’re going to play with
0:20:22 this so notebook lm yeah you can go to it i believe i don’t know if you have to be a paid member of
0:20:29 gemini or not but you can go into this notebook lm and it’s you can drop files into a folder so
0:20:36 you can go to like archive.org download the most complex research paper you’ve ever seen
0:20:42 that you have no idea what it means right pull that pdf throw it into this notebook lm tool yeah
0:20:50 and notebook lm will actually generate an audio podcast for you that explains the pdf and it’s
0:20:54 it’s two hosts there’s a male and a female host their ai generated voices yeah and they have a
0:20:57 conversation back and forth kind of like the conversation we’re having now right they have
0:21:03 a conversation back and forth that explains the pdf in terms that like anybody can understand
0:21:08 and because they have such a large context window you can throw a ton of documents in here
0:21:14 right you can you can go and find like we can go and throw the blog post about chat gpt in there a
0:21:19 blog post explaining claud in there a blog post explaining grok a blog post explaining gemini
0:21:26 throw all of that data in there yeah and then it’ll it’ll create a podcast out of that information
0:21:31 that explains the pros and the cons and the the differences between the two yeah it seems like
0:21:35 google’s doing a lot of cool stuff it’s just like people don’t know about it and even me like i i
0:21:38 know about some of it but it’s like it’s hard to get the habit like i have the habit of using
0:21:44 chat gpt and even you know i use claud a bit now it’s so hard like there’s so many of them yeah yeah
0:21:50 yeah so i mean like i i have a feeling we’re going to start hearing a lot a lot of ai generated
0:21:56 podcasts because of it because you anybody can go and take the latest like 10 news articles in the
0:22:03 ai world throw them all into notebook lm and then have it spit out a podcast for you explaining
0:22:07 what’s going on in all of these news articles oh no our buddy is already doing the tit ai podcast
0:22:14 and then if you want to take it to a next level too you can create some like custom characters
0:22:19 inside of like mid journey or or you know flux or Leonardo or whatever ai image you use create
0:22:26 some images of some characters throw those into like a hey gen or a di and have them actually
0:22:31 talking so you have you can make video content of two people having a podcast conversation
0:22:36 based on all of the data that you uploaded and that’s all totally possible right now like today
0:22:41 with current the next level like vtuber or something too like there’s yeah except the vtuber
0:22:46 doesn’t even exist it’s all graduated with ai it’s it’s really wild notebook lm is one of the
0:22:49 things i’ve had a lot of fun playing with i think once you actually go mess with it you’ll be like
0:22:54 oh damn this is cool okay that’s what jimna is good at huge context and and making podcasts yeah
0:23:03 replacing us yeah the disclaimer this is actually not as talking this is yeah surprise this is
0:23:10 notebook lm all along you know and also jim and i does have their own version of gp custom gpt’s
0:23:17 and projects they call them gems but they’re not nearly as robust you can’t do as much with them
0:23:21 they don’t seem to connect to like apis or use tools or anything like that but you can give it
0:23:27 your own context and your own custom instructions similar to you can with projects and and with
0:23:34 custom gpt’s i haven’t played with gems at all yet because i have my needs met with claud and
0:23:39 chat gpt already but they have that as well if that’s something you’re interested in and then we
0:23:46 also wanted to talk quickly about llama so llama is open source right and llama is it doesn’t really
0:23:53 excel in any of the areas we’ve talked about right like claud sonnet 3.5 is going to be better than
0:23:59 the outputs for llama for the most part same with the latest models of chat gpt same with jim and i
0:24:05 all of those closed off models they’re going to be a little bit better but if you really really
0:24:10 want like your data to be secure you want to run it on your own hard drive you don’t want it to be
0:24:14 connected to the internet you don’t want your data going to open ai you don’t want your data going to
0:24:19 google or anthropic or any of these companies that’s where tools like llama come into play because
0:24:26 you can install them locally on your own computer and they’re pretty good i mean for 90 plus percent
0:24:30 of use cases for people they’re going to be good enough right you know and so that’s really where
0:24:35 llama comes into play the other thing about llama is because it’s open source people can build off
0:24:40 of it they can iterate off of it they can go and make their own sort of fine tunes of it and you
0:24:47 know tweak it and train it and sort of mold it so if you wanted an uncensored llama you can get
0:24:50 there if you wanted to you actually ever use it though if you just only tried it because i’ve only
0:24:55 tried it i haven’t i’ve never built a habit of actually using it so i use a llama a little bit
0:25:00 but it’s only because i have my meta ray band sunglasses and llama is built into it so i’ve
0:25:05 used llama like built into the sunglasses i’m glad it exists though like i’ve said before
0:25:09 like i hate the idea of everything being open ai and you know they’re like collaborating with the
0:25:13 us government as well it’s like in the future like the idea that like all of our intelligence
0:25:18 is in control by one company right it’s kind of scary so i’m glad it exists yeah and then and then
0:25:25 you know finally we’ve got grok right and grok again i don’t know if it’s like really better
0:25:29 than any of the other ones at almost anything well it has access to twitter data so that’s
0:25:34 that’s one thing yeah so it’s got the twitter data and it’s completely uncensored like completely
0:25:38 like well like nine point nine nine yeah i mean i’ve asked it questions that i thought for sure
0:25:43 there’s no way it was going to answer yeah and it answered it for me like broke down recipes it should
0:25:51 not be giving me things like that okay and now you’re like a fbi watcher but it is very very
0:25:56 uncensored it’s got a really cool image generation model built into it called flux one which is capable
0:26:02 of making really really realistic images yeah and also pretty uncensored i’m not gonna say fully
0:26:08 uncensored because it’s not going to do any like uh like super adult content right but it’s uncensored
0:26:13 in the terms that you can prompt anybody’s names like celebrities and you know like crazy political
0:26:17 memes or whatever right that’s one of the things people aren’t using it for a lot yeah which by
0:26:23 the way is illegal in california as of this week did you did you see that yeah so gabin Newsom
0:26:29 in california just signed a bill basically making making like political memes that were generated
0:26:32 with ai illegal in california but we don’t need to get there we don’t we don’t need to get i have a
0:26:35 feeling that’s gonna go to the supreme court but anyways yeah yeah that that’s gonna be interesting
0:26:41 to watch play out i mean elon must move to austin now so he’s not really under california rules
0:26:45 but as soon as that law came out elon went and posted some sort of crazy meme and said oh this
0:26:50 this would suck if this ended up being seen in california or something right so you know elon
0:26:56 he’s a troll yeah um yeah i mean so so grok is it’s it’s gotten better though like the fact like
0:27:00 they started for like nothing like what was like a year ago or it wasn’t that long and it was like
0:27:04 people are like oh there’s no way he’s gonna catch up it’s like well he hasn’t caught up but like
0:27:08 it’s pretty decent now yeah like you know i’m not going to use it over chat to bt obviously but it’s
0:27:12 it’s decent so what you’re talking about this before you hit record but what sort of like
0:27:18 use cases would you use it for like in terms of it like hunting twitter profiles for you yeah i’ve
0:27:21 used it a little bit like this i think it was robert scope it was someone like that who shared
0:27:26 how they were using it uh that you can like pull up data about users right like you can pull up like
0:27:30 like what kind of tweets do they do what are their top tweets just anything you know i’m not using
0:27:34 it a lot for that you know use case i don’t really have a huge need for that yeah but it works for
0:27:38 that it’s like it’s it’s the best for that right now yeah i mean also you can pull up things about
0:27:42 news too right because it’s got access to what people are tweeting about yeah information about
0:27:48 something that’s currently hot and happening right now is probably the best for that yeah but
0:27:53 there is also some downside so as well right like well yeah if people are posting memes as if
0:27:57 it’s real news it might think it’s real news and feed it back to you right so there’s that as well
0:28:02 and then the the very last one i want to touch on really quickly because i think when it when it
0:28:08 comes to research right when you want to like deep dive and learn on a topic or you know find
0:28:13 information about something specific it actually needs to search the web i still think perplexity
0:28:19 hands down is the best yeah now perplexity is model agnostic if you have a pro membership
0:28:24 for perplexity you can get in there and choose i want to use gpt for they actually just added open
0:28:29 ai o o one in there as well as one of the models you can use you can pick claud you can pick lama
0:28:34 you can pick any of these models and it will use whatever your model of choice is but what it’ll
0:28:39 do is it’ll search the internet find information for you and then once it finds the information
0:28:44 it uses its large language model to help sort of summarize and make the information that it
0:28:49 gathered for you a little bit easier to digest yeah i mean perplexity i mean i mean there was a
0:28:53 reason that you know we got air vans around about the ceo to come on he’s like one of our first two
0:28:57 guests i guess yeah yeah and that was the reason that i wanted to get him on because like yeah i’ve
0:29:02 been using perplexity for like since the early days of it’s like it’s been one of the most useful
0:29:06 things and like early on most people didn’t know about it and people in silicon valley were talking
0:29:09 about it but like outside of that it was like oh it was everything was chat to bt i’m like
0:29:13 people are kind of sleeping on perplexity it’s like pretty damn good yeah yeah yeah no i love
0:29:19 perplexity i mean honestly when it comes to using ai just yeah the ai tool that i use the most
0:29:24 more than any other ai tool right now for me it’s perplexity yeah i go to perplexity more than any
0:29:29 other ai tool because i just get random ideas like i’ll be i’ll be walking and i’ll see like a
0:29:33 tall building and i’ll think it’s like a fascinating architecture and i’ll just be like
0:29:38 who built this building and why does it look so cool or you know something really random like i
0:29:43 just asked so many random questions to perplexity it like feeds my curiosity it’s amazing for podcast
0:29:48 too right like so we you know here at inbound there was a event yesterday where like a bunch of
0:29:52 podcasts hosts were there and i talked to a few of them and they didn’t know about this they didn’t
0:29:55 know they’re like oh yeah when you’re when you’re interviewing somebody perplexity is by far the
0:29:58 best way to like pull up notes about what you should talk to them about yeah you just have to be
0:30:03 careful because sometimes it will pull up like linked in profiles of like the complete wrong person
0:30:08 like if you start asking questions about matt wolf yeah well there’s a professional golfer named
0:30:12 matt wolf also right and so talking about golf and you like what it’ll give you a mix right like
0:30:17 i’ve had it summarized and it’ll say you know matt wolf co-host the next wave podcast he’s got a
0:30:23 popular youtube channel he’s you know his handicap at golf is like and it will it’ll start to like
0:30:29 blend the two so you probably want to double check sometimes but for the most part it’s pretty good
0:30:35 right it’s pretty dang good but you know to go ahead and wrap this one up i want to kind of
0:30:41 let’s like sort of rapid fire like use cases best tool for that use case right so like
0:30:49 copywriting and storytelling for me i’m probably going to claud yeah claud yeah for coding for me
0:30:55 i’m probably still going to claud but maybe open aio one i haven’t tried it for coding yet myself
0:31:00 yeah i mean i i think i code more than you right so like i like i’m more you know for you i think
0:31:04 when you can use claud you can actually just see it right there right which is obviously
0:31:09 awesome yeah but you know as someone who codes like for me like you know o one you know preview
0:31:13 at least from what i’ve heard from people i trust if it’s better than yeah i’d definitely be going
0:31:17 with o one yeah yeah i’d be using in some like cursor yeah cursors for sure cursor is great
0:31:22 it’s a it’s a fork of visual studio code with like ai built into it yeah where you can like
0:31:26 read your whole code base and have it basically has context of your entire code base when you’re
0:31:30 working on something yeah amazing yeah i mean since open ai o one came out i’ve literally been traveling
0:31:36 the entire month so i have not had like a a focus session to play with it so my answer might change
0:31:43 on that one yeah um you want something completely uncensored anything goes i think groc is the
0:31:49 obvious answer for that one right yeah um also really really realistic images yeah great memes
0:31:53 yeah good for memes great for making memes great like the images it it’ll generate like
0:31:58 really realistic pictures of of people that either exist or don’t exist right it’s right really good
0:32:04 at that with flux one built into it yeah so math science logic that kind of stuff we’re looking
0:32:09 at open ai o one that’s really the only one that’s going to do that really well right now right right
0:32:14 for research for me perplexity like i said that’s the one i think we’re the same on most of these
0:32:18 like a major yeah disagreements and then if you need super super long context you need to upload
0:32:22 all the harry potter books and all the lord of the rings books and all the game of thrones books
0:32:28 and you need to find a good comparison between all of them you’re going to go to gemini and then
0:32:33 also gemini is really good to create this podcast yeah we’re for creating a podcast and replacing
0:32:39 our jobs yes yeah so that’s kind of the the breakdown i guess of like what these different
0:32:46 models are good at and what we use them for i think that about i think that covers everything
0:32:50 so you know if you’re you’re listening to this podcast and you like stuff like this
0:32:54 make sure you subscribe to us on youtube we’re going to keep on putting out videos like this
0:32:57 we’re going to get a lot more tactical we’re going to give you a lot more use cases a lot
0:33:02 more ways to actually implement this ai in your business so subscribe on youtube if you if you
0:33:07 prefer audio we’re wherever you find podcasts so make sure you subscribe on all of those platforms
0:33:20 and thank you so much for tuning in with us today yeah thank you
0:33:21 you
0:33:22 you
0:33:23 you
0:33:25 you
0:33:27 you
Episode 26: Which Large Language Model (LLM) Should Your Business Use? Matt Wolfe (https://x.com/mreflow) and Nathan Lands (https://x.com/NathanLands) dive deep into the pros and cons of various LLMs in this jam-packed episode.
This episode explores the capabilities of AI tools like ChatGPT, Claude, Grok, and Gemini, and highlights how businesses can leverage these technologies for tasks such as copywriting, research, coding, and creating custom content. Get ready for a detailed discussion on emerging AI trends, practical applications, and the ethical considerations surrounding AI usage.
Check out The Next Wave YouTube Channel if you want to see Matt and Nathan on screen: https://lnk.to/thenextwavepd
—
Show Notes:
- (00:00) Deep dive on tools for copywriting and research.
- (05:46) Preview version outperforms but below full model.
- (08:38) Use GPT-4 for most tasks; complex, OpenAI-01.
- (10:09) Compress PDFs for Claude to read easier.
- (15:11) Exploring English, introducing custom GPT projects’ benefits.
- (16:05) Cloud runs and displays code alongside chat.
- (18:56) Notebook LM converts complex PDFs into audio podcasts.
- (22:13) Llama is best for local, secure data processing.
- (26:56) Perplexity Pro: Choose and use various models.
—
Mentions:
- Inbound 2024: https://www.inbound.com/
- Claude: https://claude.ai/
- Midjourney: https://www.midjourney.com/
- Flux: https://www.flux.ai/
- NotebookLM: https://notebooklm.google/
- Gemini: https://www.gemini.com/
- Llama: https://www.llama.com/
- Grok: https://x.com/grok
- Perplexity AI: https://www.perplexity.ai/
—
Check Out Matt’s Stuff:
• Future Tools – https://futuretools.beehiiv.com/
• Blog – https://www.mattwolfe.com/
• YouTube- https://www.youtube.com/@mreflow
—
Check Out Nathan’s Stuff:
- Newsletter: https://news.lore.com/
- Blog – https://lore.com/
The Next Wave is a HubSpot Original Podcast // Brought to you by The HubSpot Podcast Network // Production by Darren Clarke // Editing by Ezra Bakker Trupiano