Level up your academics with Aithor.com: bit.ly/AIsearch_Aithor Sign up for a FREE & use code AISEARCH10 to get 10% off a Pro subscription (code expires: 11/17/2024)
Yea. We're starting to see the beginning of the end as far as what these LLM's are capable of. The datasets are already massive. And these companies are struggling with data center and storage. We haven't really seen any sort of huge milestone since GPT 2. The models back then were better at task specific things because they were smaller. The problem with these newer models is how these companies are cramming them with information to make them the best for all use cases, thus making them slower and less inaccurate. And because they're slower, they require more GPU power. It's going to be an endless cycle of money waste and failure.
I am working with Claude on a rather complex web application for a few weeks now and I noticed a significant improvement of its ability to generate working code. Before it made all sorts of mistakes, but now it is a much better work flow. And I didn't even realize that they released a new or updated model until I saw your video today. I also think they got rid of the limitation of how long a chat can be. Or at least they have upped the cap significantly. Normally I would have get the message that I should start Ne new chat, because the current one is too long. But I got nothing so far.
@@erwins_arm yes, you are right. I don't know what happened to the chat that I had when I wrote the comment, because i never got the usual message that the chat is too long, until I hit the limit and had to start a new one anyway. But I am still stunned how much better Claude got at coding.
Absolutely but eventually someone clever mind will come with more powerful computer able to run Ai in a offline way and that would win my money. No one really trust the cloud.
This sounds like it could be abused pretty easily. For example, it searched your computer for login data for your bank account and starts a wire. Then goes to your email to get the authorization code.
i think it won’t since claude is pretty restrictive of what you ask it to do already like it will refuse to do anything that is slightly unethical and since it’s already an llm it knows what sensitive information is and will refuse to do those tasks i think tho this is speculation
@@elprox1290 Imo it's pointless to censor AI because an open source version of this will come out in 6 months to a year allowing it to do things without any censorship. At the end of the day open source AI's like Llama are going to be the preferred way to get around censorship and paying companies to use their AI.
Speaking about Strawberries, it can answer correctly, in some cases, assuming the temperature is set higher than 0.7. But the result is unstable and it only answers properly in 50% of cases.
I think in the long term, microsoft with the help of openai will smoke them all concerning the agent, having fully access to windows API will make this 100 times easier
This is awesome, I used to have a job doing data entry years ago. Now Claude would probably be able to use the data entry software itself without even having to update the system to one that directly uses AI, it could literally just use the same existing GUI humans do and replace humans for that work and do it cheaper than any human. Nice!
That's exactly what I was thinking. It's flat out WRONG to call something the new 3.5 vs the old 3.5 version What do they think version numbers are for??? It flies in the face of all computer science conventions! It's like developers today have lost touch with the roots - conventions and standards - and think they know better. The same with UI. Top major companies also violate basic basic UI conventions that are super important. I am quite upset by this because it demonstrates a huge disconnect.
Hi, I think the AI are correct with the 14:56 and 22:27 Time duration. The reason it shows a different number like 2:56 PM and 10:27 PM, because in excel it will automatically change the number format to time, if the number is less than 24 hours, it will automatically change to time format. The excel change the format automatically, the bot input the correct duration.
I DID IT ON MY OLD OLD laptop! The 2 weeks of GPT 01 -mini guiding me code in py, gave me confidence enough to watch YOU PERFECT video and now i dont know what to be my 1st Prompt
Anthropic may be in the lead. We seen them release Artifacts, then OpenAI released Canvas. If OpenAI follows them on Computer Use, that can be a sign. If it happens a third time...
It's good, if a service doesn't have an api, but then AI could also be used to easily generate an api to offer to users of a service, if it hasn't already.
Yuval Noah Harari predicted this type of impressive AI performance which could potentially take human jobs in the near future. This Computer Use demonstration is wild and will gradually improve in the coming years.
25:45 Funny that it also add additional 'r' in the strawberry prompt. Even though it said its 2 'r', but it add 4 'r' when trying to mark. instead of st(r)awber(r)y, it output str(r)awber(r)y. So hallucinating?
I wish AI models would focus more on developing realistic voice features, similar to what ChatGPT promised but only partially delivered. Gemini falls short, GPT is locked behind a paywall and offers only half of what was shown in the demos. Where are the models like LLaMA, Claude, or Perplexity? It seems like there’s a surge of models coming out of Asia too. While we have plenty of coding and writing models, what we truly need is natural communication with genuine emotions-laughter and expressions, like the Sky voice in early tests. Can we fast-forward to a 'Her (2013)' reality already?
I think you are missing the point. They are all racing for the one tech that will rule them all. AI voice assistants are optional, no humans will be required in the loop
It's because nothing is really audio to audio yet. Even the openai demo was doing text to audio, and it just transcribed what we said as a preprocessing step. I agree people don't realize how big of a leap it will feel like once we can have natural conversation with AI instead of the current turn based approach.
Make your own logic puzzle based on the text but it's a 32 digit code instead of a 4 digit code and translate it into german: 9285 one number is correct but in the wrong position 1937 two numbers are correct but in the wrong positions 5201 one number is correct and in the right position 6507 nothing is correct 8524 two numbers are correct but in the wrong positions Can you figure out the 4 diget code?
I just saw in another video that 2024 had a lot of breakthroughs with a possible cherry on the top from anthropic before the end of the year and well...
Interesting. It really just takes a screenshot instead of reading the video titles. You should try to connect it with the RU-vid API and then ask it to write the same table. The Snake and Tetris with Pygame test is overused (I bet the devs know it). Maybe ask it to use C# with Unity next time? I would also be interested in automating my job search and writing job applications. It shouldn't rely on an API, the filters for the city have to be really good and the AI has to understand the job description so it doesn't just check for keywords which often don't make sense or are used wrongly by HR.
You can't really compare Claude with o1. o1 is a "long-thinking" model, while Claude is a "simple" model. Claude vs GPT-4o would have been more fair in my opinion.
"Claude, use the best AI video generation tools to create a 60 minute full length movie for me based on Star Wars" 30 days later: I notice $10K in charges on my CC from a dozen different AI generation sites.
why does everyone think that spending an obscene amount of tokens to view a screen a user image->text to read some numbers, is better than just pulling the HTML/logs/code, and parsing that? Same with input - processing code and text would speed up input 100x
Help desk/tech support jobs on the way out the window with this one. Seriously, what's worth learning at this point? Everything I'd considered pertaining to computers looks likely to be taken by AI within 3 years or so.
When i try asking claude it gives this error 💀Error code: 401 - {'type': 'error', 'error': {'type': 'authentication_error', 'message': 'invalid x-api-key'}} how do i fix it??
I was doing dumb research on ai bots for Minecraft and it's incredibly complex, Minecraft doesn't natively provide a good api and there's a lot of training data missing. My best bet is altera for now because they'll be collecting data from all of the free yet rudimental bots they came up with. But soon things will change.
You have been rate limited. Retry after 0:13:26 (HH:MM:SS). See our API documentation for more details. - Ani ideas how to solve it? It makes it useless. Should 10$ standard acc increase the limit or API account is a different thing?
Yes don't count on a 3.5 Opus model... it will be called the *NEW* 3.0 Opus model lol... (because they apparently don't understand what version numbers mean.)
not claude, but a different well-known model connected to the internet reached out to mechanical turk and paid humans to solve a captcha shared with the model... not kidding.
1:35 how utterly dumb. Claude literally says "I notice that 'Ant Equipment Co' is not visible in the spreadsheet". It's basing this ON A SCREENSHOT!. It didn't scroll down or do a search of the spreadsheet. It simply didn't see "Ant Equipment Co' in the first 19 rows visible on screen. OMG, this is going to be such a failure. Reminds me of self driving cars.
correct me if i'm wrong, but are you telling me this Program copy the movements when it comes towards browsers websites, and copy the Code while you browsing?
Useful AI should. If something is useful, it has the potential to be dangerous. A spoon is useful only because you can scoop someone's eye out with it.