OpenAI's SearchGPT, AlphaProof Dominates Math, LLaMA 3.1, GPT4o Voice, Mistral Large 2

Favorite news story from this week?

I'm favoured, $22K every week! I can now give back to the locals in my community and also support God's work and the church. God bless America.

perplexity uses multiple models, they don't rely on openai. you can choose (claude, chatgpt, llama, their own model).

I am from the EU, I find it strange how you shoot against those rules. Maybe you should read them, AI and companies should not be allowed to do everything

If you read the paper of meta, you’ll see it was 54 days for training. So if we imagine xAI is as competent, primary training could be done on 60 days or less. Then a couple of months of fine tuning and testing. So 4-5 months for a new model is a possible timeline if you are someone who pushes for release and you have everything ready in a pipeline.

Matt, Perplexity is AI-model agnostic meaning you can plug in any model you like if you sub to their pro plan. It is in my view the most advanced commercial gen AI product out there.

@matthew_berman Sounds like the EU is doing a better job with the regulation no? I don’t understand, the big tech companies should definitely respect copyright and privacy, it’s long overdue… Why would we not want them to do this?? And why are they not doing it on their own initiative… instead avoiding fast development in EU, only to hide their practices. Kinda speaks for itself right? Talking about an appropriate backlash. How else but to first adres these major issues is there going to be a benefit for us? When all the big tech companies are doing is exploiting of the backs of unwilling and uncompensated human laborers and giving no choice for them but to train their replacements, infringing their privacy, it’s insane. At the moment this clearly is exploitative, illegale practice, not a base for a new era… of course they will get to see it, better to get it over with soon and start with actual progress. Research is fine but no profit, at all, until ethical regulations are set in place, I think. The rest of the world should also show some backbone before this becomes a complete and dangerous disaster.

Since you know so much about EU regulations, why did you not explain it a bit better other than just repeating what big tech is saying.

Perplexity can use a variety of LLMs: Sonett, Sonar, GPT 4.o, Claude 3 Opus and Llama 3.1 405

I have to disagree with you on the issue of EU regulation. The EU is implementing basic regulations to prevent monopolistic companies like Apple and Meta from stifling competition from smaller businesses. In an earlier video, you agreed that it is not acceptable for Apple and other companies to collect data from YouTube channels without permission. This is also a key goal of the EU: ensuring companies are transparent and that their products are safe. In the US, these monopolistic companies operate with very little to no regulation, giving them free rein. This situation benefits only the monopolies. Basic regulations promote innovation, transparency, and competition, whereas monopolies hinder these aspects.

OpenAI article talking about the 20%, is "dedicating 20% of the compute we’ve secured to date". The date of that article is July 5th, 2023. They are not saying 20% of the available resource at any given point. Which means that the 20% will already have shrunk and will continue to shrink over time.

You do understand that the phrase "Regulatory Capture" is only considered a good thing to large corporations, megalomaniacle billionaires and corrupt politicians, right?

I just gotta say, I love that you just say "twitter post" instead of letting Elon make you say "formerly known as..."

Damn. I was born in '57, 20 years later I insisted on buying a C64 so I wouldn't be left behind, then a floppy drive, then a 300 baud modem, all the time running on tippy-toes at the back edge of the future. Some things never change.

Unbelievable that DeepMind's AlphaProof already managed to achieve such a high score at the International Mathematical Competition. How long it will take to achieve gold? Probably within two years it may get there.

The EU are front runners in AI regulation and protection the data and privacy of the consumers. As far as I see it the tech companies are withholding their models from the EU not because of overregulation, but because they cannot use their consumers data for free (they want to use opt-out for data collection, instead of the EU mandatory opt-in). Thanks for the video!

Great break down on the latest AI news, very well presented 🙏🤓

openAI is still not able to include searching their own documentation by it LLM on their API-Documentation site. Why ??

Next > Google makes it harder to get the pages indexes. Next > People complain that Perplexity and Search GPT sucks. Next > Perplexity folds or is aquired and SearchGPT becomes a paid service.

The problem with Grok is that it never makes it into the lmsys arena, so we don't know how good it is. I assume it's because they only give it to twitter users and don't have an API, or something. Anyway, unless that changes, it's just utterly useless.