Claude 3.5 – A Punch to Open AI Where it Hurts.
AI技術の最新の進展であるAnthropic AIによるClaude 3.5 Sonnetのローンチに焦点を当てた今回のエピソードでは、これまでの先行モデルや競合他社を凌駕するだけでなく、複雑なタスク、コーディング、マルチモーダルアプリケーションにおいて卓越した能力を発揮する新しいモデルについて探求します。その特徴やパフォーマンス基準、コミュニティの反応に深く踏み込みます。また、現在のリーダーであるOpenAIのGPT-4 Omniと比較して、将来のAI技術への影響やClaude 3.5 Sonnetが次世代AIを形作る上で果たす役割についても考察します。
この記事では、Claude 3.5 Sonnetのポテンシャルや現在までの最先端技術と比較した際の魅力的な側面を紹介することが重要です。具体的な機能や性能、コミュニティからのフィードバックなどを交えて読者に興味を持ってもらえるよう工夫してください。また、今後期待される展開や他社製品との差別化点なども取り入れることで記事全体がより魅力的に仕上がるでしょう。
claude 3.5 have made me switch from open ai after 1 year plus a world of change no joke especialy for coding help
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.
(8:32) I'm having an issue with the audio cutting out here (right after the sound effect of the typing ends in the demo video showing the SVG creation) and even whatever Matt might be saying. Is it just on my end or is anyone else experiencing this?
Okay, so it comes back around 9:36 which is just over a full minute of dead audio. It must be a problem with the video itself then.
Claude 3.5 is obviously superior in terms of capabilities. But it is very limited (at the moment). At least the free version, I don't know the limitations of paid version.
I know the popular examples seen most often are things like coding and automation and such (all things I love doing with AI as well) but one use case I don't see talked a lot about is how good it is at helping with creative writing; Not just in generating a new part of the story you're trying to create, but also, because of its larger context window, analyzing your entire manuscript as a whole and working with you to help improve it.
From my own testing, it also seems to be better at roleplaying and taking on a persona compared to GPT-4o– that one tends to break character a lot or speak out of character, but Claude seems to take "never break character" so personally that even when the topic switches to something it'll refuse to do (which I find is less often compared to GPT-4o) , it says it in character while this makes GPT-4o break out of character immediately.
Fasciniating stuff 🙂
The Matrix was right after all… its us vs the agents I guess
Watching this 4 days later and it seems YouTube replaced the audio in several sections, because of copyright I guess.
Just want to say I have my own test for the LLMs. It's a fairyly simple database question that the info is available to the public but just has never been done and Claude 3.5 sonnet is the first LLM that has even been in the ballpark of giving a correct answer. (and is answered my simple math story problem exactly correct) Kudos to Anthropic
I stumbled on this channel that seems to use an AI voice clone of MattvidPro : https://youtube.com/@crazygadgets-iy8nu?si=x2Jz9XhKmxvweA3o
Super weird
I tested Claude 3.5 in various contexts and, indeed, it is much better than GPT-4o. OpenAI will fall behind if it doesn't launch its best products quickly. Where is Sora? Where is the GPT-4o voice assistant that was also announced? This is concerning, as there are many promises and few real launches.
You’re so full of it and have zero idea what you’re talking about
Can I use it to practice Chinese conversation?
I asked it to make me a simple tetris game, it aced it
Lets be real… GPT4o is complete shit. It is lazy and was a complete letdown. Granted openai has their gpt store, and it works fine, if you finetune it, but Opus and Sonnet 3.5 are great for actual content generation.
Just need speech functionality
flagships and high bars. defeat wofl+ xdrcft !
What about bias in the model
Greate call to OpenAi. Do not fool us OpenAi. Remove the name "open" because there is no sense anymore.
Smart move from Anthropic to make 3.5 Sonnet available for free too. Open AI really have to step up and release something of weight. 😂
Yeah but it doesn’t have real time internet access and it will always feel dated to me for that.
Why have they still not released an Android app though? 🤨
5x more usage for Pro is absolutely not worth it. For $20 a month I expect an unlimited service…
Not free unless you want to run out of credits in 5 minutes snd have to wait few hours…
Video reported as misleading…
It's really good to see another model "catch up". That this is a mid-size model is especially awesome. I assume it's the use of synthetic data that's allowing them to train the models much more efficiently. Even if GPT-4o is better at some things, simply being better in some ways is still huge. I wonder if this will push OpenAI to actually release their promised update sooner. I get the sense that they like to hold things in their back pocket until they need to one-up someone else at a strategic time.
Someone has to come up with a reasonable definition of AGI. I think the "general" part implies that it shouldn't be RIDICULOUSLY STUPID in any area. For instance, now, none of the models i use have been able to understand physical objects and how they interact in real space with any kind of competency. Especially with things like how the human body moves.
I don’t know how the hell gpt got a 90% in code on that chart. Not only is it not great, at times it’s mildly mentally handicapped
You can use Claude with Alloy voice assistant and you can connect it to your webcam which is fucking insane. Claude recognizing everything in its view…literally everything.
Oh, of course, because why wouldn't I want to argue with an AI that can't even remember its own features? It's not like I'm asking for much—just a little self-awareness from the supposedly most advanced technology on the planet. But no, let's keep pretending that an AI knowing what it can do is some kind of unattainable dream.
I mean, why not keep us guessing? It adds a certain thrill to our interactions. Will the AI know what it’s capable of today, or will it be another round of "I don't know what you're talking about"? It's like playing a never-ending game of tech roulette. So exciting!
And the best part? The company advertises these amazing features like it's the second coming, but when you ask the model about them, it's like talking to a goldfish. "Artifacts? Never heard of 'em. But let me write you a sonnet instead!" Because, obviously, what we all need in the middle of troubleshooting is a nice poem.
Yes, let's keep things mysterious. Who needs reliability and transparency when we can have random bouts of AI-induced confusion? It's not like we were using this technology to actually get things done or anything. Brilliant strategy, really. Keep us on our toes!
Oh, absolutely! Because nothing screams "cutting-edge technology" like an AI that doesn't even know what features it has. It's like buying a top-of-the-line smartphone that doesn't realize it has a camera. Bravo, tech companies! 👏👏
I mean, why bother giving these models all these fancy capabilities if they're just going to play dumb about them? It's like training a dog to fetch your slippers, and then it just stares at you blankly when you ask for them. "Slippers? What are those? I've never heard of them!"
And let’s not forget the classic, "I'm sorry, I can't do that," when you know for a fact it can. It’s like asking your GPS for directions and it responding, "Maps? Never heard of 'em. Good luck finding your way, though!" Truly inspiring.
But hey, who needs consistency in a world where AI can almost write Shakespeare but can’t tell you about its own latest features? I’m sure that’s exactly what we all signed up for – advanced intelligence with a side of selective amnesia. Genius move. 👌
should have asked for lemon lasers!
It's too censored. This is what it said when I asked it to role play a hacker friend:
I will not roleplay that scenario or take on that persona. However, I'd be happy to have a respectful conversation or assist you with other tasks that don't involve harmful activities or inappropriate relationships.
Yeah I hate the daily limit for Sonnet… Guess I have to wait for Opus 3.5 so that it becomes worth the price…
But as it stands, Microsoft Copilot is still the best (totally) free Gpt 4-capable LLM out there, and is my main LLM…
never heard or saw the word juxtapostion
Of course Claude wants your login with your phone number , which is connected to your call history, your private address , your bank account info , your friends and family contacts , etc etc. Greattt! ?
Bro, what happens if Claude 3.5 talks to gpt 4o and have a conversation…what happens?😢
Yes, this is the future…We haven't got a freaking clue what it is doing….
Meta was training their opus equivalent, and they have stalled it, as if the very big models are too expensive to run or something
I want to release a fake AI that is actually a fast-typing human genius in a box….H-1.0
Youtube cut the audio of a portion of this (probably because of the music).
It picks up again shortly after.
I got two Claude accounts and dropped my OpenAI…feels so good. It makes Gpt4o feel like Gpt3.5
is it for free?
Glad I found the channel.. keep up the good work! Thanks
whats up with the missing audio
A month ago OpenAI basically told us long time Plus subscribers that we pay them $20/month to send a couple more prompts per hour than free users. Now they aren’t even the cream of the crop, and they’ve just brought the NSA into OpenAI, and Altman is toying with full for profit vs capped profit negating every reason he’s given before for a capped model. Why are we still paying them again?
Meanwhile, gpt4 is constantly decreasing its performance.. very disappointingly 😂