Claude 3.5 Deep Dive: This new AI destroys GPT

Anthropic Claude 3.5 breakdown & hidden abilities. Make games, animations, reports, & more! #claudeai #anthropic #ainews #ai #agi #singularity
Thanks to our sponsor Abacus AI. Try their new ChatLLM platform here: chatllm.abacus.ai/?token=aisearch
Newsletter: aisearch.substack.com/
Find AI tools & jobs: ai-search.io/
Donate: ko-fi.com/aisearch
Here's my equipment, in case you're wondering:
Dell Precision 5690 dell.to/4cAYjSZ
GPU: Nvidia RTX 5000 Ada nvda.ws/3zfqGqS
Mouse/Keyboard: ALOGIC Echelon bit.ly/alogic-echelon
Mic: Shure SM7B amzn.to/3DErjt1
Audio interface: Scarlett Solo amzn.to/3qELMeu
CPU: i9 11900K amzn.to/3KmYs0b
0:00 Wild things it can do
1:29 Getting started
2:07 Snake game
4:55 Audio visualizer
9:19 Copying website design
10:28 Tetris
13:00 Shooter game
13:50 Presentations
15:25 Document to infographic report
17:00 Animating diagrams
22:45 3D particle cloud
27:07 Specs of Claude 3.5

[@theAIsearch]

Thanks to our sponsor Abacus AI. Try their new ChatLLM platform here: chatllm.abacus.ai/?token=aisearch Prompts used in the video: create a snake game using python Create a single HTML page that let's me upload an audio file. Sync that audio with a visualizer like the attached image. Don't use unsupported libraries. convert this UI design into frontend code create tetris game using python create a JS presentation on the health implications of coffee Create an interactive 2-page infographic on the attached document Using the neural network diagram I've shared as a reference, please create an interactive HTML/JavaScript animation that demonstrates the flow of data through this network. The animation should include: 1. A visual representation of the network structure, matching the layout in the image. 2. Animated paths showing data flowing from the input layer through the hidden layers to the output layer. 3. The ability to input sample data into the 5 input nodes. 4. Visual feedback showing how the activation of nodes changes based on the input. 5. A simple UI to control the animation speed and reset the simulation. The animation should be educational, helping users understand how data propagates through a neural network. Please provide the full HTML and JavaScript code needed to run this animation in a web browser. create an app in one html page that can be used in Artifacts. make an interactive 3D particle cloud, with a max of 100 particles. use Three.js for the simulation. do not use unsupported or 3rd party libraries or functions. create your own functions.

[@maoaris]

I've been a ChatGPT boy, but your video made me try to experience Anthropic but the UI that I have is quite different to the one that presented in the video, specially it does not have the cool preview option for JS or HTML, is there anything that I am missing ?

[@theAIsearch]

@@maoaris What country are you in? Maybe it's not supported yet. Also, did you go here: claude.ai/

[@MihailsZuravlovs]

Does ChatLLM provide the same Claude 3.5 or they are different?

[@leafstarivy]

Bro calm. I was already blown away by the first example you gave at the beginning of the video and then you kept making it even crazier. This video truly deserves the SHOCKING or STUNNING title

[@kutagaru3676]

Honestly, if he put shocking or stunning In his titles, I wouldn't even care, this guy always talks about the best.

[@theAIsearch]

thanks!

[@Pototoes949]

We blew past Shocking and Stunning and straight to speechless

[@matten_zero]

GPTs are cool and APIs can do more. You can also create agents much easier. That said, these results make Claude 3.5 a better base model. I just don't trust Amode. He's gonna want to legislate OS away because of "safety"

[@blisphul8084]

The issue is that those terms are used as clickbait too often, so they have no meaning anymore and only serve to drive sane people away.

[@Loadofbulltrading]

"when I touch myself I lose the game" well said my friend

[@retratosariel]

LOL😊

[@frogz]

and sadly, everyone watching also lost the game!!

[@mattcav6992]

first thing I looked for in the comments..

[@Sithhy]

Happens to me too often

[@StefanReich]

What are you, a mormon? 😆

[@TherronKeen]

Claude 3.5 has been the first AI that was able to help me figure out some coding problems for my game, and then accurately make changes after I tested the code and requested alterations. I had a really weird edge case that I couldn't find an example for by Google searching, and I didn't understand the underlying system enough to solve it. Claude solved it AND explained the code to me in a way that helped me understand the feature. It's fucking mind blowing.

[@MrRyusuzaku]

But if you didn't have the knowhow you wouldn't know what to ask either. This seems like a good step forward from the idiotic gtps

[@SahilP2648]

How are you using it for your code? Just the web browser? I have found a better way (although I haven't tried it yet) with Phind pro and using the Phind extension in VSCode. I need to try it out, just that I don't want to pay $20 per month since I am not actively working on a project at the moment lol.

[@PeterTroutman]

Did you try GPT4 and it failed? Or just didn't try anything but Claude ?

[@thanatosor]

We need AI to summarize huge amount of data into something usable, consumable by our limited human mind 😂

[@TherronKeen]

@@PeterTroutman I tried GPT4 and GPT4o and neither were able to solve my problems effectively. But to be fair, I'm coding in GDscript in the Godot game engine, so maybe Claude just had better access to some data regarding GDscript specifically since it's not a really a mainstream programming language.

[@Soar-hw2rw]

This is incredible. The singularity is approaching faster then I thought.

[@theAIsearch]

yes!

[@kutagaru3676]

We best be ready, im sure it's going to be a whole hell of a lot of fun.

[@MrErick1160]

That's the point of the singularity actually 😂

[@kutagaru3676]

@@MrErick1160 It sure is

[@entertainmentyoutube3606]

It stills mimics. It doesn't understand yet, although a lot of jobs can be replaced just mimicking

[@MonkeySimius]

I did a rather complicated coding project with Claude, using my phone, in about 15 minutes that it took me about 2 weeks of 4+ hour coding sessions on my PC to correctly complete. Claude truly is good. Especially since it doesn't break your code randomly by implementing "fixes" to your code behind your back. It is very upfront about explaining what it did and why as well as staying on task.

[@wezmasta]

Im just happy anthropic is a valid competitor. Between Google, Apple, and ClosedAi, Anthropic is the protagonist this story needs.

[@matt9546]

Heel turn incoming

[@rock3tcatU233]

Silence ChatGPT, we have a new king in the real of LLMs.

[@theAIsearch]

exciting times! i'm sure claude 3.5 will be dethroned shortly as well

[@xentaurro]

@@theAIsearch if it doesn't get replaced within a month, start using it as the name to beat in video titles.

[@kutagaru3676]

🗣All hail

[@NicVandEmZ]

@@theAIsearch it can make very convincing chat messages clones look

[@sychrov81]

yeah but free version is limited

[@jaywv1981]

I watch about 5 or 6 youtubers' videos regularly for my AI news but yours has quickly become my favorite.

[@theAIsearch]

thank you!

[@wezmasta]

Who do you watch?

[@jaywv1981]

@@wezmasta David Shapiro, AI revolution, mattvidpro, Matt wolfe....a few more

[@himalayachoudhary231]

@@jaywv1981 can u say few more as well all of them .

[@cdunne1620]

Yea diss is de best ai channel

[@brianlink391]

Well, the funny thing about the Snake game is that it created the exact same game for me as well, with the same message and everything. It seems to have been trained specifically on that, or for all we know, it could be identical code pulled from a database.

[@jichaelmorgan3796]

Probably not a database, just extra, specialized training. It is essentially a learned "skill" now.

[@theAIsearch]

interesting, thanks for sharing!

[@brianlink391]

@@theAIsearch I've been wondering whether it's pulling from some database or if it's just incredibly consistent. Remember that other AI guy, Matthew? (I can't recall his name exactly, but it starts with an M.) He does a lot of testing with snake games too. I thought it was an interesting observation. It seems like the AI might be highly trained on the snake game by now, or it's drawing from a database. It would be interesting to run many different trials and see the exact code that's varying and changing. This could provide insight into what's actually working under the hood. For all we know, it could be utilizing a vast database of open-source code, grabbing that code, and using an AI in the background to modify it. If they aren't doing that, I'd be disappointed because it would be a great idea.

[@erickflores785]

Glad to see I wasn’t the only one who noticed the same game

[@jichaelmorgan3796]

@@brianlink391 would be interesting to find out exactly for sure

[@FlumenSanctiViti]

"Unfortunately, Claude is not available to new users right now. We're working hard to expand our availability soon."

[@amruzaky4939]

Many such cases😞😞😞

[@AdvantestInc]

That introduction was the most impressive and almost scary, yet hilarious demonstration I've ever seen! Fantastic breakdown, your work is incredibly clear and engaging.

[@theAIsearch]

Thanks

[@Alice_Fumo]

I really like your videos. For example I respect you taking your time to properly test the model and not rushing out a video asap. I also like your general tone of talking, which doesn't tend to be overly expressive / hyped. Sort of reminiscent of AI Explained. I've gotten generally quite tired of clickbait and hype-men. I hope you keep it up.

[@theAIsearch]

Thank you! I appreciate that!

[@cdunne1620]

Yes I agree, well done man, keep it sane and calm

[@KiLVaiDeN]

Great video ! Very cool tryouts, which really show the tip of the iceberg of the possibilities of this model. It's quite incredible to see how intelligent this model is, considering it's the "middle tier" compared to the 3.5 Opus to come. I wonder how good 3.5 Opus will be. The improvements over previous models are gigantic, we are not talking about small improvements here, those are colossal considering the few months which have passed. It means that at this rate, which is going to accelerate also, we might have models with higher intelligence even faster. This is mind-blowing. I really wonder how the society will adapt to those models, because right now, we are really on a path that leads towards real usefulness. The artifact feature is an easy to use way to get what you want, to clean up the chat, it's very well thought. I must say that I'm nothing less than impressed. I was already impressed by Claude 3 Opus, and already made the switch from openai to anthropic few months ago. I don't regret it !

[@youtubesurfer134]

Having Patrick star's head explode is a pretty good example of what's going on in my head right now? Did it really make a video game? 🤯 Where did this come all the sudden?

[@ianPedlar]

Its like they're saying "Thanks kiddies, you've let the cat out of the bag, looking at you Open AI, let us take it from here" Open AI are saying "CRAP! Quick bolt on something that looks like Artifacts"

[@denisblack9897]

My take is OpenAI decided to step out of spotlight, so that when IT apocalypse comes they wont be ones to blame 😅

[@fynnjackson2298]

This is incredible, lost for words, just mind blowing... And it's only getting faster and smarter. Wild!

[@ObservingBeauty]

Mind blowing indeed... Thanks for this valuable video

[@FabbrizioPlays]

When I first tried 3.5 Sonnet, I had it work on an XML file. I said "maybe there should be a header element containing X, Y and Z attributes". It did that, AND cleaned up the other places X, Y and Z were defined in the XML. It's such a simple thing, but I've never seen an LLM have the intuition and confidence to do cleanup/optimization without being told.

[@srsanderson4932]

GPT 4o was launched around a month and a half ago, Claude 3.5 is really impressive. I am using both for my work and Claude really does better many times . Both these tools have made my life easy. Open AI probably be preparing to launch another beast of a model in coming weeks to again get 1st position.

[@danielmethner6847]

I wouldnt bet on it. They've announced to release the voice mode for 4o in autumn and the video mode for early next year. Which places GPT 5 earliest in summer next year, which adds up, seeing how many politicians are now running openAI

[@SportsIncorporated]

I noticed in the coffee presentation nothing was said about polyphenols or all-cause-mortality,... Maybe if Claude 3.5 was told to create a graduate level presentation or a presentation to researchers? So far I've been extremely happy with Claude Sonnet.

[@AIroboticOverlord]

From now on i will only drink Reine Claude lemonade to be amongst the first to harvest our AGI overlord's ultimate grace.

[@AdminOne-hj4dt]

😂

[@ILikeAmazingThings]

Damn. I imagine myself building my own game just with prompts in the future. So much will change in so many areas.

[@comebackcs]

I enjoyed every minute of this review. Now you have a new follower : )

[@theAIsearch]

Awesome, thank you!

[@PseudoProphet]

Somebody needs to provide Anthropic with more compute. Their rate limit ends pretty soon. 😂😂

[@theAIsearch]

yeah the free plan is very limited. i get around 5 messages every 5h

[@chillaxinmusic6295]

@@theAIsearch Well even the paid version is quite limited if you use it a lot. I mean I'm trying to do a kind of an advanced .php site with it, so I've hit the limit like 3 times today already. If not those limits, then the sky would be the limit lol

[@Ecstasio]

@@theAIsearch I get around 4 messages every 3 hours. According to Anthropic if I upgrade, I will get 5x tokens. So, are they saying I'll be stuck at 20 messages per 4 hours or 25 messages every 5 hours in your case? Seems very restrictive.

[@orthodox_gentleman]

I pay for the pro version yet I get limited frequently!

[@Ecstasio]

@@orthodox_gentleman Good to know. I suspected that may be the case. As someone who literally engages in occasional sessions that are hours long, this does not sound optimal for me.

[@bobomblavandia8561]

Finally, some good testings. thanks.

[@user-fc2ww7gc9k]

This is amazing 🎉🎉, thanks for keeping it super detailed

[@theAIsearch]

You're welcome!

[@tanglestone3]

Was waiting for your review on it .

[@theAIsearch]

😃

[@feelgoodmusic8722]

Amazing video. I'm totally blown away.

[@regalx1]

I went ahead and tested Claude 3.5 vs Chat GPT 4o in terms of comedy. Claude wasn't able to mimic every comedic style, but the ones that it did well on were A. Terry Pratchet Discworld B. Boondocks C. King of the Kill and D. Futurma. Whereas Chat GPT was able to copy the just the characters, and settings but couldn't imitate any of the styles. So in terms of comedic writing Claude 3.5 wins by a landslide.

[@JFGames365]

llms cant write comedy though

[@derman3219]

It's more censored than ChatGPT though

[@regalx1]

@@JFGames365 It's true it can't write comedy, but at least it now has an understanding of the comedic styles. The way I would do it just have it generate a bunch of jokes and then manually tweak the joke that I would find the most funny. The reason why this didn't work in Chat GPT 4 is that they were just all godawful, you were better off just working from scratch.

[@elon-69-musk]

incredible i feel this is first real progress after gpt4 release

[@a.y.greyson9264]

The insanity of this video is an understatement and I use GPTs all the time to help clean up scripts at work. I knew Claude was different, but this update is simply mind-blowing.

[@ulaviyal_uzhi]

Thank you, that was More Helpful

[@abramjones9091]

I'm building relatively complex and dynamic economic games with this, amazing

[@filatovvxdxd]

Как же я тебя люблю , спасибо тебе большое за всю твою работу

[@osamucabral]

Amazing! I can’t wait to see this getting popular for creatives around the world. ❤

[@icebear7830]

Thank u for this vid!

[@you-are-the-tharpsters]

I made a 2d fishing game with it. So you click a button called fish and it takes a second and you either collect a clam or fish or crab and you gain money also if you buy anything you get coolness as I added a coolness meter as a joke. Also if you have over 50 dollars you turn yellow and over 100 you turn blue and over 500 turn white and over 1000 neon red. Also a little meter shows when you and what your next rank is. The shop holds 3 upgrades the first is called extra hook the 2nd is extra cash and the 3rd is faster fishing.

[@erikals]

Impressive! Most Impressive!

[@M_A_A2301]

Impressive ! thank you & ps : I just subscribed xD

[@theAIsearch]

Thanks for the sub!

[@jothammartin5127]

This is on a whole other new level

[@technocorpus1]

I have tried it and it really is that good.

[@rasmusfoy]

Great video

[@RAZR_Channel]

8:36 - Song name : Dumpster Fire...

[@YoungMoneyFuture]

Now this is excellent! As a no/low code dev, so excited to be entering this revolution🔥

[@bm63]

I asked Chat GPT to create a snake game , Space invaders, and Breakout game… So it must be you. OK it’s not in a window, like that, but it made them

[@oscard9429]

Bro that AI song you made is actually really catchy

[@theAIsearch]

thanks. here's the full version: www.udio.com/songs/7caswcZpJpSU8MLYdMVKtc

[@mrkammytv]

I already like that it has separate windows for chat and code

[@JesusMartinoza]

Claude 3.5 + Cursor Editor is just crazy

[@devSero]

When Claude came out I wasn't going to touch it with a stick now I'm applying it to where I can. I hope it overtakes all other models but I'm also hoping for mistral & llama community supported projects.

[@BrianHallmond]

This is my dream Albeit, competition will be harder. Now I can finally make a game I want.

[@michael73532]

I hope I never lose my amazement at what AI can do. I feel like a kid in an amusement park all over again.

[@TABandiTA]

that snake game by Claude, holy hell. i wonder what 3.5 opus capable of

[@FabbrizioPlays]

Anthropic has been popping off lately, and I think this is exactly the kick in the ribs OpenAI needs/deserves

[@fl1ntlock1]

Subscribed 👍

[@keshavrao4969]

Changed the world forever

[@nelsonwesterhout2588]

Not sure why you think ChatGPT couldn't do this. I gave it the same Tetris prompt and it successfully coded the full game with no error messages like you encountered in Claude. I tried the audio visualizer as well and it worked just fine. All with 4o. I'm gonna continue trying in ChatGPT as I continue the video. So far it works just as well if not better that Claude. Would love to hear your thoughts on this. Regardless great video so far. I didn't know you could do all of this cool stuff with AI chat bots!

[@datsme8986]

Alright! Now, I am waiting for a sci-fi movie and an Oscar award winning movie.. made by AI. That would be amazing. ♥️💪✌️

[@maickelvieira1014]

This is insane, it is hard to even believe, with amazing tools like this i cant even imagine what people will create now.

[@theAIsearch]

indeed

[@AlexanderSuraphel]

Nice video! It'd we nice if you increase your font and resolution of your video for those of use watching on smaller screen :)

[@bemci1975]

this is mind blowing 🤯

[@lucid227]

Amazing. At some point there should be a setting where the programming errors are automatically fed back into the AI, and it alters its own model to learn from this for everyone.

[@concordance5387]

Great video. Perhaps you could have Claude help you write a script with a wider variety of adjectives. 😊

[@renkurniawan]

amazing man

[@AncientPurpleDragon]

"Create a Snake game using Python." What other language would you make a snake-based game in?

[@techwiththomas5690]

I have a suspicion though: Matthew Bermann also tried the snake game and apparently what you got is exactly the same implementation (as I can see visually, code based I am not sure!). What if the AI devs just trained it with hardcoded material for the most common problems that were known that many people would test and many LLMs could not do, so now it just replicated its training data and everyone thinks: "Wow this AI is so good", while it actually is not because the answer is not really freely generated?! Oh and P.S.: I use Claude 3.5 Sonnet via PoE and the coding skills are great, but its "memory" is really bad. After 4 to 5 prompts it completely forget what it implemented so the following code from there on our does not fit the original implementation anymore.

[@GDFistful]

holy sht this is insane i cant imagine how AI will evolve 10 years into the future

[@mk__cyanheron1154]

To what extent it is creating the game and to what extent it is simply remembering/sampling a trained code from memory. Internet is filled with SNakelike Mariolike Tetrislike and Doomlike games from CS students all over the world. I agree that it is hard to think up a game outside of the training distribution, but still. (I'm in no way saying that it is bad, I think that it is a grat leap forward I only urge some healthy dose of skepticism).

[@galco1419]

In my one test I did, it was not impressive.. I tried to get it to make a simple platformer with obstacles that randomly shoot from the ground. It make tons of mistakes like the character can only jump once, gets stuck in the air, the obstacles weren't behaving as requested, etc. and when I asked it to fix these mistakes it would, and then simultaneously create new issues that were arguably even worse.

[@MR-DURO]

Prompt: « Generate GTA6 with unicorns », Results: Tetris with pink blocks lol

[@Markoss007]

In ChatGPT I do unique projects. Even big with thousands of lines in different files. It's not just copying, it's remembering, extending, and fixing bugs. But of course, it is not with one prompt. It's like working with Jarvis (without voice)

[@VolodymyrPankov]

The type of games was created by people and recreated millions of times. And now artificial pseudo intelligence is able to produce you the garbage and even to edit some shit. It's about combinatorics. Primates: IT'S SINGULARITY! LOL 😅 Clowns 🤡 ARTIFICIAL PSEUDO INTELLIGENCE

[@GeneralKenobi69420]

Remember when the original ChatGPT came out and it felt revolutionary? That was barely a year and a half ago.

[@joker201235]

Claude 3.5 sonnet is available on lmsys, but it is not yet on the leaderboard due to insufficient votes, most likely

[@1augh0utL0UD]

This is scary good. These updates come out so fast that it's hard to keep up...

[@TheTwober]

What I get from this video is that you are mainly very easily impressed. Let's take the audio analyzer as example: there is a complete API in every browser to do that, so all you need is a few lines of code to adapt them. It is a good thing that Cloude knows about it, but it is about as impressive as giving someone 3 pieces of lego and ask them to build a triangle with it.

[@Windswept7]

This changes everything.

[@sanderhoogeland9161]

The only problem I have with it is that as long as it doesn’t have some kind of voice mode, I just wont use it as much. It’s still exciting to see the development and progress though.

[@lamsmiley1944]

This is the first LLM that has really impressed me for a while (at least with its coding ability). I’m still not convinced that LLMs will get us to AGI as they haven’t the required problem solving capabilities.

[@Hohohohoho-vo1pq]

GPTs are not LLMs.

[@rumfordc]

@@Hohohohoho-vo1pq I asked GPT and it says GPT is indeed a type of LLM.

[@Hohohohoho-vo1pq]

@@rumfordc GPT 4o for now is an LLM because it can only output text to text and image to text but it cannot do image to image. But GPTs are not limited to just language. Don't confuse GPT from openai to the actual GPT term. Google what it means

[@rumfordc]

@@Hohohohoho-vo1pq Google says: "GPT models are known for their exceptional text generation capabilities, but they are just one type of LLM"

[@Hohohohoho-vo1pq]

@@rumfordc by google I don't mean Wikipedia. You don't seem to even understand 1% of what GPT is. GPT can learn any type of information. GPTs can even learn to predict images from images. It's just that most GPTs are trained on language and get more attention there.

[@JustMaier]

16:40 “If your job is to create these reports or presentations…” You’re going to be out of work because they don’t need you to make them anymore.

[@PromethorYT]

That AI song is actually good. Scary haha. What a strange future ahead of us.

[@marcoseverini9741]

damn the frontend one is crazy dangerous

[@ChuckSwiger]

Thanks, was wondering boggling where artifacts were hidden - have to log out and log back in to enable, duh.

[@jamespowers8826]

"destroys" used to be such a useful word.

[@SpectacularSpidey670]

cant wait to create games with this

[@theAIsearch]

😃

[@shr4n]

3:34 haha good one

[@answerguru]

Have you fact checked the output? Curious if the Tesla summary or other items contain any hallucinations…

[@Jimbilben]

Super impressive - it would be good to check the actual accuracy of what it is creating though, e.g., in terms of the Tesla report. Are all of the numbers correct?

[@rytwars2971]

That is mind blowing. 🤯

[@RighteouSpirit246]

My problem with it is the censorship. It will not accept my custom instructions , all the other AIS will, but 3.5 Sonnet will not. Even Opus 3 does.

[@Roxxo88]

when he said "you see if I touch myself, I lose" I felt that

[@BooklyCrashCourse]

this is amazing

[@MarsorryIckuatuna]

I registered for an account and it blocked me on registration 😢. I think it doesn’t like the use of VPN’s or Ad Blockers, but I’m not about to disable my stuff to appease one model. Back to 4o we go 😂

[@whiteknight54455]

Loved it's capabilities BTW what does LLM mean ?

[@takenwandu5815]

LLM - Large Language Model

[@whiteknight54455]

@@takenwandu5815 thanks

[@TacticalCastle]

that may be the most overwhelming thing I have ever seen. Like... I need to power off my virtual reality and reset this back to factory defaults...

[@GeorgeMonsour]

Intellectual property laws will joyfully be under pressure. Power will shift and there will be commercial chaos! The outcome could be ugly or there could be real equalization. The only constant will be change. Say you use an AI to create a new game based on an idea you prompt for. Then you publish it for sale and it becomes a hit. How do you think the intellectual property ownership should and would be determined? Who owns the judgment? My view is that IP should have had scale limits and forces may finally see that limits need to be set in IP law. Creativity is a shared existence.

[@DihelsonMendonca]

Now imagine running that intelligent AI into your phone to run locally. It's just a matter of time. LLMs would be amazing and small. 🎉❤

[@rumfordc]

*Large* Language Models would be small?

[@jasonk125]

Can I suggest that you add a high pass filter to the audio of your videos. There is a low "thud" in my headphones every time you bump the table you are using.

[@Milennin]

Switched my subscription over from GPT to Claude. GPT is pretty good, but I hate how inconsistent its quality can be. One time it's amazing, the other time I use it it's dumb as bricks. Having tried Claude for a bit, it's definitely a step up over GPT in overall quality, at least for writing code for my game.

[@Kutsushita_yukino]

it would be perfect if they fix their censorship issues though i’ve been getting lots of false alarm

[@matthewclarke5008]

From now on I will use Claude for coding and ChatGPT for advanced math.

[@hassejansson]

Awesome 👍

[@theAIsearch]

Thanks

[@micbab-vg2mu]

great model:)

[@SiimKoger]

We need feedback for learning, makes sense that it would benefit AI models as well. Practice doesn't make perfect, feedback does.