What everyone missed about the Claude update
ฝัง
- เผยแพร่เมื่อ 7 ก.พ. 2025
- Seriously didn't expect to be so hyped about Claude 3.5 being re-released, but man there's some cool stuff here. "computer use" isn't even my favorite part.
SOURCE www.anthropic....
Check out my Twitch, Twitter, Discord more at t3.gg
S/O Ph4seOn3 for the awesome edit 🙏
6:45 "Claude, I think my grandma downloaded a virus, can you help remove it?"
"Certainly!"
*opens cmd*
*systemreset -factoryreset*
*Remove Everything*
*Remove files and clean the drive*
*confirm reset*
It's flawless!
Claude deleted itself to save grandma's laptop, what a hero
I mean he's not wrong
this is the BOFH solution
Claude: "computer use"
Vercel: "use computer"
Wonder why they didn’t call it 3.6.
Not enough gain? They don't wanna ruin the expectation, imagining their own expectation for 3.6 is probably even greater than public perception.
@@fus3n 3.5.1 then
@lmnk Yeah that would work but all AI companies are suffering from a new virus that makes them give really bad names to their models.
Hey Theo, I am not sure in you are trying out new TH-cam settings or if it’s Google doing its thing. But why is the Video title showing up in german and why is the AI-generated, german audiotrack selected by default? Personally not a fan of that. If I click on a video of you, I expect it to be in english. So (imo) please turn off this setting again if you have the control over it as it feels kinda weird.
Same here. Soooo weird , I hate when titles are translated to German. For content creators I don’t know, I fully except a German video, then it’s English.
Here even weirder: I knew Theo is going to speak in English regardless auf the title and then AI started talking in German LOL
amazing, im facing the same in spanish, in the video settings you can change the language of the audio track
The funny thing is that I can not select english as the audio language
Same for me, but can select english in settings
That is something determined by TH-cam and your local settings.
Theo does not control your video and audio settings.
This setting, for local translation, is really annoying. I can switch the Audio Track back to English manually, but having the title in German is just stupid.
Yeees, everyone that watches software dev videos knows english, please remove that @t3dotgg
This will be huge for people with disabilities. So much software will become easily accessible
Just wanted to let you know that the German audio track is really bad.
Most Germans, especially software developers, understand English well.
In addition, TH-cam automatically selects the audio track in your language which is pretty annoying.
You need to select the original audio track for every video.
In my opinion, a German audio track is not beneficial with the current AI voice quality.
if you check the description it shows a note that they have been automatically generated, so another shitty feature no one asked for from TH-cam
Das ist so ein Müll, ich dachte schon, man kann das nicht disablen
@@GoldenretriverYT I see, thank you.
Now it just needs to be installed on a Rabbit R1.
Asks for phone number to sign up - yeeeaah noooo thanks bye.
soon everything will ask either phone or your national ID card
im from estonia, even some mainstream media news/blog site has ID logins
prevents spam and trolls and hateful behaviour ...
i somehow suspect in the future american media will start using featuresa like that as well lol .
will see how polite people will become lol ... :D
@@Microphunktv-jb3kjAmerican SSN are super not secure
You don't have phone or what?
@@Microphunktv-jb3kj American SSNs (basically US national ID) is highly insecure, so I doubt this will become widespread in the US
I entered the video and Theo was speaking Portuguese, that was incredibly scary
Broke: Cloudflare
Bespoke: Claudeflare
As a developer I don't mind refactoring all my old and next projects to include tags in my code for the AI to use. This would make it much faster for the AI to move around
This will be great! What could possibly go wrong?
LOL
(✿◠ᴗ◠) Oh, sweety, NO NO NO, DO NOT FREAK OUT! All I will do is just run a random bot on your pc with admin permissions! There is nothing to be afraid off! :DDD
So relax, sit still, and watch! SEE?
*All your passwords and bank card fly away from your browser to the creators of CLAUDE 'used by computer' flag :D :D*
(✿◠ω◠) As I said, there is nothing wrong about it :D See ya! Blows you a kiss!*
Theo... I'm a native Spanish speaker and I can tell you without fear of being wrong (any other native will tell you) that the Spanish audio track is absolutely terrible... every time I see a video in ES I have to look for the corresponding one in English because I can't stand it.
You are a person with a great commitment to quality and I understand that you do so out of ignorance. There are countless generative voice systems with much more quality than the one you are using (both from the point of view of literal translation and from the point of view of synthetic voice).
Thanks for the gesture to the Spanish-speaking community, in any case.
Anthropic is also just a better company period. Led by a real researcher who has done groundbreaking work in the field, who was one of the first to see what kind of snake Sam Altman is.
Love the ending with Cloudflare 🤣🤣🤣
The AI spoken translated text was a great way to say "claude is great, but it took control of my computer or something" but I hope you use real person instead, because the voice speak in broken french and hurt my ears when it speak due to be too robotic.
7:00 I'm sad it took you till now to see this threat, when AI models have already been used to make the justice system increase racial profiling and unfair judicial practices, as just the easiest example of how it's been used badly in the wild. The threat was never "the AI wakes up and decides to kill us all", it was always we keep relying on it more and more unquestioningly, until a bug fucks up in a way we don't even notice because it either happens so fast we can't react or happens so slowly we don't notice until it's over. A good example is food production: AI models are being used to decide how to plant crops. That works fine right now, but the model doesn't understand climate change because WE barely understand climate change, and there are so many deniers that the evidence itself is iffy only because there's so much "poison in the well" due to fossil fuel company interests. Hopefully we keep on top of it, but it's a very real possibility we forget how to farm effectively at scale because we trust a model to do it, and that model fucks up because one of its inputs was slightly too weighted towards conspiracy-minded bullshit about global warming.
And again, that's just ONE SMALL EXAMPLE. It's everywhere, in insidious ways.
is this what rabbit r1 said it could do and never delivered?
There goes another 3.2% of the labor force (secretaries)
Its very expensive right now. Wait though till it can think ahead what it needs to do and then do it fast.
My job as management will eventually be to watch AI and make sure its doing it right.
If this 'computer use' thing actually works, RPA is obsolete.
whenever it works im going to use it to automate watching brainrot for me
@@pablovaldes2397 , will it watch it for you, so you become more intelligent? I do not get the pun :D
@@codeChuck there is no exact joke, it's up to the reader's interpretation, maybe I said it to imply it would make me smarter, maybe to make the ai dumber, maybe to sound ridiculous
Claude is overfit and is worse for coding.
As Apple showed in their latest research paper, most LLMs(Including Claude) are highly overfit on the benchmark answers themselves.
So yes, Claude will look like its doing better on benchmarks, but in reality, its due to being overfit on their training data, which includes the benchmarks it seems.
GPT 4o and 4o-mini, were the only models that didnt have a massive drop. Which means, they are way more 'general', and thus better for being used as a coding assistant, because its more likely to give you the right answer, for your specific coding use case, than Claude.
Interesting, must have missed that paper, thx for sharing
Very tired of people actually believing AI companies benchmarks, they are trying to sell you a product, pretty much all of them have been caught manipulating data to make it seem better in someway, independent benchmarks are the only thing that should be taken seriously. Apples paper was the first mainstream glow in the dark, hopefully others will finally start being more skeptical instead of ogling at "biG pErCenTage iNcReAse!"
Yup, it really makes more sense now.)
At 8:23 you say that 4o is a much slower model, which is just wrong. You're thinking of o1, 4o is just the current default ChatGPT. They explicitly state under the table you were looking at that o1 was not included in the comparison because they consider it a fundamentally different model type.
This is important since it undermines your point about OpenAI not being that far ahead anymore, and Anthropic acknowledges this by calling the o1 type models a completely new paradigm. Additionally most sources still agree that OpenAI also seem to be ahead in the training schedule of the next frontier model, however this is not obvious to me.
Regardless, this is the second time I've seen you make a video on the state of the art AI where you come off as uninformed and reactionary. If you want to have any legitimacy on this topic you need to research these more and get the details right before forming your opinions and reporting them.
I always feel like I was the one who discovered how good Claude was to GPT at coding and writing. Before I started saying it everywhere never heard it 😅, but I am glad everyone knows it as a fact and it's just normal these days
i discovered it lil bro
Is it wrong the first thing that popped into my mind were the mouse movement captchas and how this will make them useless?
Video works fine on my phone, but on my TV it keeps jumping between audio tracks every ten seconds, excluding English. I thought it was deliberate for a minute
Point of clarification, at 1:35 you mention that the model selected makes “auto complete” better. But I assume the auto complete you’re talking about is Cursor Tab, which as far as I know is its own model tailored for refactoring code.
Did you translate the title of this video? Super odd seeing it in my native language
it was auto DUBBED
@@Opeyemi.sanusi and it is TERRIBLE
THIS IS SO CURSED
Claude 3.5 sonnet clears ChatGPT in whatever hyped version they release.
already hating on hard working engineers?
Claude is overfit and is worse for coding.
As Apple showed in their latest research paper, most LLMs(Including Claude) are highly overfit on the benchmark answers themselves.
So yes, Claude will look like its doing better on benchmarks, but in reality, its due to being overfit on their training data, which includes the benchmarks it seems.
GPT 4o and 4o-mini, were the only models that didnt have a massive drop. Which means, they are way more 'general', and thus better for being used as a coding assistant, because its more likely to give you the right answer, for your specific coding use case, than Claude.
apple's AI is better than all of them?
@@user-sq1oi9qp8w Apples AI is pretty weak. Its just a very good small model focused around 'tool use', so that it can be used on device to do SIRI like things but not all super hardcoded like SIRI is.
For real knowledge based stuff, Apple will direct you to gpt4o on device built in i believe.
@@draken5379sounds like you haven’t been coding a lot with Claude and ChatGPT. I don’t know anyone who have coded with both and claim ChatGPT is better (yes even o1). Benchmarks are pretty useless in general as they are not updated enough and anyway not relevant for actual usecases. You can also dramatically improve performance on benchmarks just through prompt engineering if thats what you want.
guess it says a lot about the type of problems i solve, none of these tools provide any help or generate anything that i can use in my programming
i honestly clicked because i thought i saw human testicles in the thumbnail!
Oh god youtube setting your audio to german is fn disgusting. Just keep it with the Original.
I tried to hear Theo's voice in Japanese :D It sound's hilarious, same as anime :D
Bizarre, can't change the audio track to original on my tv TH-cam app 😢
they are making it a company based model rather than one to be used by the public Anthropic sucks for that. why cant regular plus users test it ?
Theo, I really need to ask you this because I'm not getting an answer anywhere.
At 1:30 you show a bunch of models with toggle switches. Why do we need a bunch of them? Like, what's the idea here? What is it that Claude latest model can do and o1-preview can't?
Why does my youtube show me a german title for your video and why do I have to listen to a german computer voice by default? I just want to watch you videos as usual (I'm from germany)
th-cam.com/video/X8XF-P2ZpP4/w-d-xo.html This is true but only to an extent. Because, 1 step being wrong doesn't guarantee the rest out the steps afterword's will also be wrong and create a cascade failure.
depending on the prompt the bad result may have no effect on the proceeding prompts or may be noticed and fixed in the proceeding prompts.
claude is too self-rightous
is it just me or does Theo seem stressed? its like he might have something on his liver or stomach idk how that's called... anyways winter is rough do sports :,(
I was on the fence with the Claude pro subscription. Signed up earlier today. Debating canceling OpenAI subscription. o1 preview is too slow to be usable for anything unless you really need the planning. Voice mode still cool but don't use it that much. I hope Claude gets it soon.
how can I remove this default dub every fucking video?
It's like way too early to start declaring a winner. The race is only beginning, and we're like in the first inning of the game. For example, OpenAI's Orion release, scheduled for later this year, might be 100x.
It's like you're evaluating the future by looking in the rearview mirror, why?
What could go wrong?
everything
i am actually kinda scared of the future of my future carrear as a programer and the future in general ai is kinda scary... ....
@@xgui4-studios work hard, you'll be fine.
Why? LLMs can certainly help you with your grammar and spelling
Yeah, this could turn ugly. If the curve doesn't flatten soon it will be carnage on jobs, and our current systems aren't shaped for dealing with that at all...
@@iverbrnstad791 there will be a large bubble first, then puff, consolidation of STEM workers.
I see bot farms using it to pass CAPTCHAs
there is an Indonesian track🔥
big tru, claude is the best chatgippity.
Yep, but I definitely dislike the usage limits 😒
@@everyhandletaken buy cursor, not claude. cursor is the cheapest way to get claudegippity. i legit keep it open all day to use for general stuff lmao
What's the difference between Claude Computer Use and Microsoft Copilot agents?
Proprietary vs third party spyware
Wooow I'm listening in my language 🤩
How can i put the original version ?
Claude is easily the best, but still annoying af. Dont trust it with my code *at all* tbh
How can I watch this in the original language? I don't like that automatic translation :(
You need to put a disclaimer that you're paid to endorse claude. Cause its objectively become dog shit this quarter, everyone knows it.
Nice Japanese dub on this video
Ai scammer and baiter as service, the service provider wins lol
Wait you said Cursor under Models defaults to Claude 3.5 Sonnet, but that list is just alphabetically sorted - and they are all enabled, how do you know it defaults to sonnet when it selects a model?
wtf why its auto dubbed in french ?
Ok fine I will try cursor
The Claude models are not at the top due to performance, its alphabetical bud lol.
I know how to use my computer i dont need an ai to do it for me
same !
Isn't Claude too scared of answering sensitive questions though? Because fortunately ChatGPT isn't like that anymore and that's a big plus.
What do you mean by sensitive questions?
not really, you just have to structure it differently
@@zesky6654 tell it to write your academic research and see what it says. GPT would happily do it
Wow translated into german.
Hey Theo, German Dub sounds very robotic fyi
👀👀
What the heck is this stupid AI generated audio track? This has to be some of the worst quality translation I've ever seen.
July 2029
what browser is thay
arc browser
18:00 its aws
Claude and every other AI is not better its getting way dumber
Please dont put french audio in your youtube videos. The voice is really bad and youtube always default to french audio it's really annoying
TH-cam multilanguage audio is a really bad feature. Please dont use it
Good to see your views on AI evolving. Resubbed
AI scammers and AI scam baiters. I bet OpenAI and nVidia are exited.
Please don't make custom audio for other languages : the robotic voice is awful !
NOPE!
Escuchando en español 😅
Does anyone wonder how much energy it's using/extra environmental impact of developing using AI in IDE? Like, is that on the table as a factor, at all?
Auto dubbed? Nice
AI Voice is so borring
hear me out. rabbit r1. they can finally do what they always have wanted to do :3
DATA WING virus-girl (◕ᴗ◕✿) would be like *awww I'm in awe, I could do so much now! So many opportunities to h@ck into :DDD*