AI News: The Best Chat Tool Got SO Much Better!

[@Diamonddavej]

1:50 I just taught Claude to read crystallographic data from scientific papers and generate valid CIF files (Crystallographic Information File) that are used to visualised the arrangement of atoms in crystals. I wanted to do this for years. Many many times I found papers that describe new crystal structures, but no associated CIF file. I did this by using Projects, uploading papers and documents describing how CIF files are put together, their structure, formatting and rules. This is great, as the alternative is buying expensive $$$ crystallographic software, or using a rubbish old utility someone wrote 20 years ago that no one remembers how to use. But now I just upload a paper and a correctly formatted CIF file appears like by magic, it's not magic of course, but it's still amazing. Oh, and it also knows how to compare papers, and make relevant comments, describe the differences between the same crystal / mineral examined by different researchers who use techniques or conditions, fascinating and a great time saver.

[@jl6165]

This seems awesome. I’m experienced with AI art, but just started generating reports. But while lot of AI programs do amazing stuff, getting the to export it so I can share it with someone is a real challenge. Are you able to print these reports?

[@TheDontletmedie]

I've watching almost every video you put out Matt and I will say there is not a SINGLE ad read that I've ever skipped --and the reason why? You're genius at making the products engaging, interesting and something relatable with examples given throughout. Not sure if your background is in advertising, but man -- you are truly great at presenting the products you are sponsored by! only editing this b/c somehow YT drew a line through the fact that I was calling you a genius!

[@BunnyOfThunder]

I used Suno to make a rock album that I absolutely love. The record labels are right to be worried.

[@guardiantko3220]

Suno made me drop my music subs. Tbh I haven't listened to non AI music besides from anime/video games, for a couple months now

[@r34ct4]

We're doomed.

[@metron0m]

@@r34ct4 capitalists and labels are doomed

[@Arthur-jg4ji]

yes , personally i never really liked music that much but with AI i now have music that i like and listen to.

[@brianckelley]

Absolutely never occurred to me that this was the target use case. I've always seen these music apps as super neat, but largely a novelty or "writers block" therapy for musicians.

[@kevinbee5080]

I appreciate how when you said "Let's get right into it" the video took off with no brakes!

[@PazLeBon]

there was an advert 'brake' tho ;)

[@GringaCR]

Was a diehard fan of Chat GPT but after working today on Anthropic's Claude producing months worth of work in a few hours, agree w/ Matt 100%. So. Much. Better. 😊

[@kozinoart]

Great video, as always, Matt! Honestly I watch these videos 60% to stay in the loop and 40% because Matt just has such a nice personality X)

[@rebeccamiller8772]

Every time I get excited about a new tool I start planning a fun little project to make with it, but then something better comes along before I finish and project A is scratched and the new fun project begins. These AI advancements are just hell for this work from home AI obsessed ADHD girl.

[@bethgetzelmann1530]

I agree! I need a good avatar one that isn’t too expensive and does custom characters I designed and to be consistent and used in various ways

[@ronilevarez901]

But I bet you used to do the same _before_ AI, isn't? Cause I did 😅

[@PazLeBon]

and each one of those projects has a milion other folk doing thesame thing. billions of wasted man hours, the irony

[@rebeccamiller8772]

@@ronilevarez901 yeah.. busted ;-)

[@ronilevarez901]

@@PazLeBon not always. And each project is slightly different to the other anyway. Also, you never know who is the one that will actually finish their project, so the more the better. Sometimes.

[@MisterTwit]

"We're improving the model's ability to detect and refuse certain content" I love that "refusing content" is now something they're trying to play up as an important feature. Imagine if other things worked like this. "We're delaying the rollout of our new car because we're improving its ability to not drive"

[@spadaacca]

I'm very much for more unfiltered AI, but your analogy is inaccurate. To play devil's advocate, it's their version of saying "We're delaying the rollout of our new car because we're improving its ability to reduce what could potentially be (according to us) dangerous/harmful behavior". Is OpenAI a bit too censored - yes in my opinion. Should OpenAI readily provide data to facilitate and optimize user success in harmful activity - inclined to say otherwise. Is this what they're referring to? Who knows.

[@ronilevarez901]

@@spadaacca some people is fine with crashing into a crowd at full speed as long as they're allowed to drive while speeding like crazy 😑

[@ratsock]

“We’re delaying launching the new car while we figure out how to detect if the driver is drunk and prevent the car from starting”

[@cesarsantos854]

We are delaying this product to find more ways to save us from legal problems. Screw you user.

[@PazLeBon]

we are changing the car because that pedestrian looks like a banana to ai

[@JamesOKeefe-US]

Always the best recaps Matt. Appreciate you and your newsletter to stay up to date! You rock!

[@skelious]

Thank you so much for gathering this relevant info in a sea of news.

[@extremelylucky999]

PRO TIP: Convert PDF files to txt. Or, compress them to allow more space per project.

[@raj-rn5sv]

Project features are yet not available for me in India

[@tiltok]

Top man, everything to the point, a lot of information which always has references and time codes. Arguably the best reporter in its field. Keep up the good work, you do save a lot of time for sheer amount of people🎉

[@NESDUB]

If Claude offers text-to-speech functionality comparable to GPT-4 and provides the ability to output JavaScript-generated downloadable files, I'm all in. I often use GPT to create iCal calendar schedules, Excel spreadsheet files, and JS scripts for neat automation tools. I like the interface of Claude more, and the fact that it’s statistically smarter is very attractive.

[@DeathHeadSoup]

I like how the project feature shows you how much memory the feature has used. The best part of using LLMs locally is the insights you get with token rates and memory usage. This is something that is lacking with most AI SaaS products and is very disappointing when using them.

[@aiforculture]

Thanks Matt! I recommend your channel in pretty much every talk I do :)

[@erniea5843]

Ive mostly only used Anthropic’s Claude models for the past few months. Overall they’re the best IMO and Sonnet 3.5 just makes it better

[@emmanuelgoldstein3682]

I can imagine a future where social media has an auto-generated voice trained on the person who makes a post, or an option to pick any voice as the one that represents your "voice" that other people hear when they see your posts.

[@alelondon23]

keep going. dream big. that is already implementable.

[@EJ-vf9qf]

I imagine a future where google replaces KZbin content creators with AI allowing them to stop paying humans for content... making channels like this obsolete eventually.

[@PazLeBon]

i can imagine when social media is even more huge and no humans actually are using it

[@tobiasjakobprivate]

🎯 Key points for quick navigation: 00:00 *🆕 Introduction and Claude's New Features* - Overview of Claude's new project feature, - Interface updates and new collaborative tools. 02:18 *📁 Creating and Using Projects in Claude* - Demonstration of creating and using a project, - Uploading and managing research papers, - Custom instructions for projects. 04:07 *🧩 Gemini 1.5 Pro and Google's AI Updates* - Introduction to Gemini 1.5 Pro's large context window, - Comparison of context window sizes with popular book series, - Other updates from Google, including Gemini's code execution and new languages in Google Translate. 06:40 *🔊 OpenAI Voice Assistant and Critic GPT* - Delay in the rollout of the voice assistant feature, - Introduction of Critic GPT for code review. 09:11 *💻 OpenAI's New Acquisitions and Partnerships* - Acquisition of the remote collaboration platform Multi, - Speculations on future integrations, - OpenAI's partnership with Time Magazine. 12:02 *🎬 AI in Commercials and LTX Studio Updates* - Use of AI by Toys R Us for commercial creation, - New features in LTX Studio for video editing. 13:23 *📰 OpenAI and Time Magazine Partnership* - Details of the strategic content partnership, - Media companies' changing perspectives on AI. 15:01 *🔉 11 Labs Text-to-Speech Innovations* - Partnership with Time Magazine, - Launch of a new reader app for listening to text. 16:38 *📞 Character AI Voice Interaction* - New feature allowing voice calls with AI avatars, - Demonstration of character AI's functionality. 18:01 *🎵 AI Music Tools and Legal Challenges* - Lawsuits against AI music tools by record labels, - KZbin's approach to AI music licensing. 19:54 *🎨 Figma AI Updates* - New AI-driven design features in Figma, - Demonstration of generating designs from text prompts. 21:05 *🤖 AI Avatar Enhancements* - Synthesia announced updates including more expressive avatars and personal avatars. - Introduction of full-body avatars. - Link to further details about Synthesia's updates. 21:47 *🎬 AI Video Editing Tools* - New AI video editing tool from Captions with features like green screen effects and after-effects animations. - Potential future review of the tool. 22:15 *🖼️ AI Image Upscaling and Relighting* - Magnific's new relight feature allows creative restyling of images. - Example of relighting an image with an underwater scene. 23:07 *🏆 Leadership Changes in AI Companies* - Stability AI appoints new CEO from Weta Digital. - Weta Digital's notable works include "The Lord of the Rings" and "Avatar". 23:36 *🐇 Rabbit R1 Data Breach* - Rabbit R1 faces issues from Rabbit Tude exposing API keys and security flaws. - Risks include leaking personal information and altering AI responses. 24:31 *💾 Specialized AI Chips* - Etched announced the world's first specialized chip for Transformers, significantly faster than Nvidia's h100s. - Competition from Groq achieving 400 tokens per second with the Llama 370b model. 25:26 *🗣️ User-Created AI Chat Bots on Instagram* - Meta testing custom AI chat bots on Instagram with early tests in the US. - Examples include chat bots modeled after celebrities like Snoop Dogg and Mr. Beast. 26:09 *🏅 AI in the Olympics* - AI-driven personalized Olympic recaps with AI version of commentator Al Michaels. - Ability to customize updates based on sports interests. Made with HARPA AI

[@dave2431]

Thanks Matt! I really appreciate your up to date content. Just wanted to let you know. Great Work! Thanks!

[@davidtindell950]

Very Comprehensive A.I. Review. Thank You. Also, your own opinions are helpful.

[@nycgweed]

Claude is looking better ! Will have to try

[@MeinDeutschkurs]

Claude is great! Can’t wait for opus 3.5… I‘m so disappointed by OpenAI, so I ditched it. If I had known that Claude is that great, I had left OpenAI earlier.

[@N30Dallyr]

The spider man voice message caught me off guard. It lacks emotion but the voice itself resembles the one from the PS4 game.

[@Pusty159]

In the same way, in the past, a lobby of transport owners using carts and horses fought against cars. And now we have road signs prohibiting the entry of horse-drawn carriages. If they want to survive, music publishers must adapt to changes and accept the existence of AI. And not the other way around.

[@shadowproductions969]

That claude update is pretty huge, it's by far my fave llm for creative writing. One of the biggest problems i has was the limited message length when working on a novel.. maybe this prokect ability will help keep that under control now

[@SoverineSR]

Man, I haven't heard anyone talk about Character AI since everyone complained that they lobotomized it and took the personality and weird realism away from it in an effort to restrict it. I don't even know how long it's been anymore, maybe over a year now? They might have been the first major case of community kickback over dumbing down AI, not counting Tay since that got shut down shortly after being dumbed down. I'm guessing there's a good reason I continued to not hear about CAI since then until now.

[@mickelodiansurname9578]

If I had an AI that can read a workflow and carry out all the tasks I do and redo every day, then thats what it would do while I drink beer and goof off on cat vids

[@handpowers]

Try fabric coz its kinda like that

[@PazLeBon]

why woud anyone pay you to do that. they wouldnt need you

[@mickelodiansurname9578]

@@PazLeBon I'd have the AI pay me... to watch cat vids... important data collection right there

[@skierpage]

​@@PazLeBon companies will pay 1 person to supervise, guide, and collaborate with AI(s) cranking out what used to be done by 20 people. And that one employee will be tired and demoralized.

[@PazLeBon]

@@skierpage it can crank oout junlk and besides, if one employess can do te workof 20 people then 20 employees will do the workof 400people sooooo

[@aldorodriguez7310]

Claude Projects are cool, but the context window is still small for me, I always run out.

[@tiagotiagot]

Oh great, now they're making the NSA backdoor a feature!

[@dixalex02]

Have you already done a deep dive on websim? Sorry if I missed it but that seems extremely relevant and powerful.

[@rpeart73]

Really great work, Matt.

[@cdmichaelb]

Clauds new projects is amazing, and so is claud 3.5. But even paying 20$/m for premium we just don't get enough usage for it to be useful, so it's hard to say it's good.

[@xwdarchitect]

Try HARPA and pay for what you use, without rate limits

[@PazLeBon]

still better to learn to do all these tasks yourself lol

[@bigglyguy8429]

@@xwdarchitect If you use a wrapper you lose those extra features

[@shukrantpatil]

Also I believe none of the Claude models have internet access, correct me if i am wrong

[@xwdarchitect]

@@bigglyguy8429 sure, there is a feature trade-off. Wrappers have feature overlaps.

[@KingJ64]

The biggest issue I run into with Claude is that when you upload PDF's even with the Pro model it gives you an extremely limited number of questions you're allowed to ask. When I have PDF's I am limited to less than 10 every 6 hours. Is this something that has been updated or anyone knows a way around?

[@Milennin]

Isn't the limit based on how much content you make it read? Like, if you give it a ton to read with each prompt, it will go through its limit much faster than chatting with it without any attached documents.

[@omegadecisive]

Don't like that Magnific don't have a free tier to test or try. Nearly £40 for one month is an expensive starting price too.

[@KimmieJohnny]

I love your nuanced long forms. Please consider developing short form video clips from your longer forms. More viewers. More shares. I really want you, need you to become a major influence.

[@kel4242]

Thanks Matt, it seems like a big week for lots of updates but no big tech conference telling us about things that we can't play with yet. I am excited for Claude to have another new feature to play around with.

[@DJ369-Miami]

The real productivity bombshell is projects by anthropic. while open ai promises but under delivers Anthropic with Projects, Artifacts and Sonnet 3.5 is killing it.

[@shijilkumar]

00:01 Anthropic's ChatGPT introduces projects for customizing GPT conversational models. 02:00 AI tool enhances research paper analysis 06:18 OpenAI's advanced voice assistant is being rolled out further, with improved content detection and user experience. 08:23 ChatGPT is now available as a Mac desktop app for all users 12:36 LTX Studio offers a pitch deck feature for exporting storyboards. 14:44 Time partners with 11 Labs for realistic text-to-speech technology 18:32 Record labels losing relevance, KZbin making AI music deals 20:20 Figma introduces AI-enhanced design features 24:01 API Keys exposure in Rabbit R1 poses data leak risk. 25:59 AI technology enhancing Olympic viewing experience

[@TheBlessingReport]

great video as always

[@koganboss4874]

I'm surprised by companies that are trying to fight AI. How do they not realize that it is impossible? AI can no longer be stopped, you can only learn how to use it or be left behind.

[@hawaiianbabyrose]

it's interesting to note that with every piece of disruptive new tech there's a crowd that tries to downplay, oppress, fight, deny, and eliminate it yet they've _never_ been successful, except maybe a handful of hot button issues (idk gene editing in humans? someone more aware pls provide examples😅) i'd argue this is for several reasons. people tend to see the world through the lens of what's familiar to them, and some people are more open to new information than others, and so, some are more informed than others, even among big players as far as i see, conservatives in this regard are always the losing party in history lately, but that doesn't mean they go out without a fight that said, there are some things that disruptive tech has yet to challenge, that is disparate concentrations of power and wealth, which prove to be incredibly robust configurations given our current social embeddings dude, did i just whip out a random autistic rambling in yt comments? whatever

[@skierpage]

​@@hawaiianbabyrose ask a GPT to tighten up your long comment 😉. Music and film companies were successful in blocking file sharing and BitTorrent from widespread adoption until they came up with paid and advertising-supported streaming models that most people are fairly happy with.

[@raj-rn5sv]

Yeah, as matt said initially big companies refuse or even boycott ai are now partnering with it.

[@bkbrg9330]

been waiting for ltx and nothing i got a email saying they sent me a login code but it does nothing and they say that its not the code i need to get access, weird

[@HorrorSFManiac]

I've been waiting since March to be given access to LTX, why do they need more people to join the waitlist if they haven't even given access to the ones that are already on it?

[@haroldpierre1726]

While the models are only getting a tad better, what can be done with them have tremendously improved.

[@emanuelmma2]

Very nice and solid.

[@drednac]

Having huge context windows is massively useful for codebases. I wish I could just upload my entire codebase instead of creating isolated parts.

[@sanjesco]

Thanks Matt!

[@brianlink391]

2:14 That's roughly 6,291,456 million tokens for a 30MB text file(s)

[@PazLeBon]

lol bargain

[@MediaCreators]

Excellent, Matt!

[@ufologistsguide]

Hi Matt, love your shows, keep up the good work. I’m an old guy, but I’ve learnt so much from you and your channel, so a big thank you for that. Using the products I saw on your channel I made an all A.I. You Tube Show using SID, a Ufology & Sci-Fi Bot I created. I hope you don’t mind but I gave your show a mention in my 'Favorite You Tube Creators' feature. There are no videos in my show, only AI generated photos, voices and avatars. Hope you like it. Steve (Charlie)

[@TheRemarkableN]

Much thanks for highlighting that Eleven Labs app. I’ve been hoping for a really good AI reader like that.

[@Slvl710]

yeah Ive heard straight up sabaton and blind guardian on suno, if you listen to genres with a fewer amount of artists it becomes really obvious, another way to tell is put it actual song lyrics, which with very popular songs, almost whatever you put into the style it still sounds like the original song, which doesnt happen with lyrics from fringe bands

[@badashphilosophy9533]

As we saw with chatgpt higher context window increases dont increase output. U mentioned it does in gemini 1.5pro but thats important info i just hope it is correct

[@Silverscribe]

LTX is a rip off. Their "Beta" doesn't even provide enough to produce a single video without subscribing. So despite being in the "Beta" I'm unable to even make a single video to even see how their stuff works. And their pricing is insane.

[@badashphilosophy9533]

Yeah ltx was overhyped, I think it was timing because you could join the waitlist not long after we started waiting for sora. When we actually were able to use it it wasn't really what people advertised it to be. It may get better and the features Matt referenced look cool, but your right about the pricing which is a bit of a turnoff.

[@PazLeBon]

we will soon need 3000 20 bucks a month subs

[@b.j.freedman]

I have found the same at LTX and similar sites. At least you get five 4-second or extended videos at Runway for free, and it's $15 a month. Though I have to add I tried Runway 3 and it couldn't follow the simplest instruction in the prompt. "Character walks to center screen and faces the camera" had the guy turning his back, etc.

[@ENGLISHtranslationsAandS]

Isn't it a problem that for no coders most good working ai tools cost around 20 dollars per month.. I am not rich that is why I want learn how to earn with AI but I can't pay 3000 times 20 bucks per sub😂

[@kurocknotabi]

What are the alternatives?

[@sirhammon]

8:00 I actually sent a proposal for a democratic platform to Open AI with this exact feature, but it was for search. Search would create a summary of every single unique idea on the planet and have references for it. The query would show the ideas as single entries so you could get a list of every single idea about the origin of the universe in a list, even is some kid said it was the spaghetti monster. A separate AI would then critically analyse the idea for biases and emotional language and manipulation techniques. Another AI would list every single problem that the result has. You could then click the problem and it would list all the possible solutions that could solve that problem. Combine that with personal bias controls (Extremist Right Wing content only? White supremacy content only?) and it was the ultimate way to allow ideas from anyone to be searched for and found. It's not 200 pages of the same stupid corporate idea.

[@chakrameditation6677]

Please give us a Gemma 2 review ❤

[@brianlink391]

11:40 Do you really play with your mouse without even looking at the screen? That's pretty cool. Let's go.

[@brianlink391]

SSL?

[@I_Mackenzie]

I think all this free tier subscription stuff is heading in one direction; advertising. They could even copy the KZbin models of pelting free tier people with tons of ads until they pay for a subscription. It’s definitely about getting the numbers up first so they can hit the free tier with something to convert them. They will also be aggregating that data for advertisers.

[@spadaacca]

I am so at two minds about these developments. On one side I'm super-duper excited when I see these advancements and developments. But on the other side all the exciting creative stuff I have/had ideas for like/liked doing I feel like now are less and less special as it just takes a press of a button to create something as good or better by literally anyone. I don't think this is the kind of progress we can or should stifle, but we really don't know what the world will look like on the other side of this.

[@b.j.freedman]

Part of my work is producing content for science videos and overseeing social media content. I can make something in Leonardo, add stuff in Photoshop, extend and add text, use ChatGPT or a freebie sidebar in Chrome to generate or correct content, and wrap things up in a few minutes. But I'm working overseas and no one else on the staff has the English chops (or creativity) to write a good prompt. (I don't use prompt help, maybe they can.) So for some of these tools it is still an uphill battle. Eager youngsters (I'm 70) get the concept but can't make it work. Wait! I just had a non-AI-generated idea! I'll teach a live class!

[@brianlink391]

4:14 2 million tokens? That's a lot. But with vector databases and extended context, how many do we really need? An intelligent search algorithm is key. It'll sift through the vector database and tap into vast knowledge. But here's the hitch: with too much context, hallucinations start creeping in. So, how effective is that context length, really?

[@KiteTurbine]

the bike shop sells bikes with 2 saddles and no handlebars? sora?

[@drrains]

New Claude is awesome. But still stumped when asked how many R's in the word strawberry

[@RodskiDimitri]

Nice, I tried it and provide wrong answer but it gave the right answer whwn i prompted this 'Ok give each letter in vertical form. And count the numbers of R's' and got it right

[@drrains]

Yes. I got the vertical count form too, eventually

[@jackpisso1761]

That's a limitation of tokenisation. The LLM doesn't actually see the letters. Has nothing to do with actual intelligence.

[@bustedshark5559]

Thanks for the info, Matt. Excellent as usual, but another waitlist 'teaser'! Really? Of the many I signed up for over the past year I've yet to receive a single invite to join before the programs went to public release.

[@shewhoisaudacious]

Mr Wolfe, the Captions AI sounds really interesting. Would love a deep dive video on this. 😊. I took a look on their website and it looks like the next generation of video creation and editing.

[@achliscantplay4202]

Subjectively, all freshly-dropped AI versions behave amazingly for a couple of days and then... Get nerfed? Gpt 4o was atrocious from the start with the simplest tasks - straight up ignoring follow-up prompts, gasligting ("yep, I did it!" while not having done jack), being super-lazy... Claude 3.5 is the same now, after being awesome for 2 days 🤦‍♀️🤷‍♀️🤔😢

[@AskTheRealDrBrown]

Is Claude linked live to the Net now?

[@BuPhoonBaba]

No.

[@ThoughtFission]

Ooooo, imagin being able to watch any NFL game with Madden doing the commentating. I see big things for this AI path.

[@mr.hampton4982]

FWIW, I asked Gemini what NKo is. Gemini replied: "N'Ko script: This is the most likely meaning you're interested in. N'Ko is an alphabetic writing system created in 1949 for the Manding languages of West Africa. The word "NKo" itself means "I say" in Manding languages [Wikipedia: N'Ko script]."

[@timothyhartridge3824]

Just a heads up, the ChatGPT Mac desktop version needs the newer macOS and M-chip ... "*The desktop app is only available for macOS 14+ with Apple Silicon (M1 or better). " 😞Time for an upgrade, I guess.

[@Atreyuwu]

Dirty sneaky OpenAI gets everyone excited for the new 'advanced GPT' voice chat, makes things just vague enough that many people who *wouldn't otherwise* bother with a pro subscription jumps in expecting things will be there (I've seen so many people who thought they were interacting with the advanced version) - then they turn around at the end of June and tell everyone it will be pushed back. Will be cancelling my pro subscription today. lol (but I'm sure I'll re-subscribe when it's finally here, because it looks absolutely amazing! XD).

[@scottyharp]

26:15 this is actually pretty cool

[@JNET_Reloaded]

cant save to list

[@TrevorFosterTheFosterDojo]

WHERES THE CLAUDE PHONE APP THO !! NEED IT ! Matt any updates on this?

[@LarryAtherton]

It would be dope if we could use a project like a git repo…. Can we?

[@tmendoza6]

If I upload my math formulas and examples, I can assume that Anthropic now has that data for free

[@ONDANOTA]

7:44 so now that scene in which C3PO narrates a story at the end of a Star Wars movie is not scifi any more

[@boerbenlp8659]

I guess Software developer cant wait for big context windows. Huge Applications and software projects can get manageable with AI.

[@mustaphanasir9448]

Thank you Lord Jesus for the gift of life and blessings to me and My family $14,120.47 weekly profit our lord Jesus have lifted up my Life!!!'

[@SunstateBank]

As a beginner what do I need to do? How can I invest, on which platform? If you know any please share.

[@aisle_of_view]

The AI voice of Al Michaels will be used during the summer Olympics.

[@swstephe]

If my company adopts "Multi", I'm going to go into full "Leroy Jenkins!" mode.

[@omaritwan]

Nice video Matt, but, Woah... I disagree that Claude is the best AI Chat Tool. ChatGPT is way better for a variety of reasons... but mainly convenience and usefulness. I can go into more detail but I guess my point is although Claude's models may so slightly better on a few benchmarks, it doesn't really matter - there are so many features and capabilities that put ChatGPT way ahead in the AI race.

[@ronilevarez901]

But not free.

[@omaritwan]

@@ronilevarez901 GPT3.5 is free w/ unlimited messages and claude free has only 10 messages so its basically useless. if you're not going to pay for any, you might as well use 3.5 cuz its not practical to use a more powerful model when you can barely use it. basically for average consumers 3.5 wont be that different than the free version of claude

[@PazLeBon]

@@omaritwan theres a huge difference

[@ronpetersen2317]

I went to give Claude a try .... it wants my phone number ... There is no good reason for that ... I am not interested in getting my number on a bunch of list to get more nonsense calls than I already do. I am trying to use it on my computer. Guess I am hard passing until the get some options.

[@theterminaldave]

But is the projects feature in Claude only available to paid users? Please keep the free tiers in mind. Also, I think I'd prefer content to be about 15 minutes long, I'd rather visit two separate videos, then watch half of one, and then remember to come back and watch the second half. A half hour of varied content is quite a bit to consume.

[@mshonle]

I totally want that Groq pen!

[@KennylexLuckless]

If I had downloaded several terabytes with copyrighted music from any site it had been seen as "piracy" whatever I had claimed to do with the files, I would never get any "fair usage" exception even if I claimed I used them to learn to play "ukulele".

[@GaryMillyz]

Correct, what I heard re: those Suno songs under scrutiny couldn't be farther from fair use. They are straight up reworking a copyrighted work.

[@BoHaMo]

synthesia body scan... from the waist up of course :)

[@omaritwan]

As Fireship says, "Another day, another controversy for the Rabbit R1." Truly a rookie and VERY dangerous mistake they made by hard coding API keys (who does that lol). They're lucky the organization who found the mistake had good intentions.

[@PazLeBon]

rabbit is copywritten sex toy in europe

[@omaritwan]

@@PazLeBon uhhhhh wut lmao

[@PazLeBon]

@@omaritwan lol its trie so found it funny when they called it that

[@LIVE781REDRUM]

17:22 doesn't sound like spider man but I wanna talk to Freddy Kruger

[@Diablodasne]

Hey Matt I watch most of your videos.. google is charging me $30 a month for Gemini I just want to know if it's worth it cause it's crap.. wondering if you have a highly recommend AI? I'm a mechanic so I'm not on my phone constantly

[@I_Am_Jimloo]

I'll only be truly satisfied with AI and technology when I can download a delicious pizza straight into my mouth!😅

[@IceMetalPunk]

You're not quite right in saying there's no precedent for the music copyright lawsuit. There's *tons* of precedent about infringement vs sampling vs inspiration, and all of it ultimately has the same conclusion: the *input* doesn't matter as long as the *output* is sufficiently transformed. The training data for Suno and Udio are irrelevant to copyright, unless the models are directly producing copyrighted elements in their outputs -- which they're not. The record labels are fighting because they're scared, and using ignorance to get people on their side, but they have no legal standing here, and if they win, it'll be a very problematic day for copyright law.

[@handpowers]

I have anythingllm and can chat to files locally also

[@lioneldasilva4889]

10:28 nan it's not about coding... Co pilot can already do that... It's about creating agents that will act like if you had 4 hands... Imagine trading in real time with gpt without having to create a python script for it... Or ask your computer to play LoL for you while you take a nap

[@sveinndagur]

GPTs and now so-called 'projects' are gimmicks. You can already just load the PDFs manually. Furthermore, my experience is that these AIs cannot reliably consider all the information in large or even just medium sized PDFs despite the supposedly large context windows.

[@Todestuete]

They aren't gimmicks, they are tooling. We'll need a lot more tooling to make AI more useful.

[@GregNTech]

Honestly, with some many options out there. Who in their right mind uses Gemini?

[@shewhoisaudacious]

Love you Matt. Thanks for bringing the latest to us. You are doing the Lord’s work.

[@TheAutisticRebel]

I thought Claude was crap first time I tried. Now I realize it blows GPT-4o... *_Out of the Water!_* Wow. Matt was *_right again..._* Claude is TRULY remarkable!!! Can wait to see what they do next... these guys are good! *_REAL GOOD!_* Not to mention OpenAi and the whole Government War Machine Projects. I'd much rather pay Anthropic then OpenAi... not that I trust either... lol

[@cbnewham_ai]

When it comes to record labels and the RIAA, we are seeing a repeat of the late 90s and the days of Napster. They kicked against that until they realised the ton of money to be made from streaming - which is now the dominant form of selling tracks. The same will happen with AI - once they realise the ton of money to be made from churning out formulaic songs using AI, not having to pay artists, and getting their tracks at the top of the charts by manipulation (which they already do), they'll be full on using AI.

[@ronilevarez901]

They don't care about them using AI. All they want is to stop users from using AI.

[@cbnewham_ai]

​@@ronilevarez901​@ronilevarez901 Incorrect. When you don't have to pay a singer or band then you make more money. When you don't have to pay a director for a music video for your AI artist, then that's even more money. When you can use a digital avatar to do a live performance of your AI artist then that's even more money. They certainly won't care about stopping people generating their own music using AI - they control the charts and what you create in your bedroom won't have the huge marketing and chat manipulation so will remain pretty much unheard.

[@MeinDeutschkurs]

The Labels: I see it like this. If we buy music, we do not owe it. We just have the right to listen to it in a private context. Every other usage is not licensed. I cannot see fair use, but I‘m also not able to see restrictions on AI training. It was just no part of the licensing.

[@LIVE781REDRUM]

8:01 The birth of skynet so it begins