Interesting news that we can no longer throw compute at the problem. Given that these companies have already used more than all the data they could legally get their hands on, are we expecti     See More ance to plateau then?    See Less

Man, thanks so much for making these videos. I don’t know how I’d keep up with all this without them! 🙏     See Less

Twenty…THOUSAND?!! 🫣     See Less

17:07 Grok has access to WAY more data now that it has direct access to all government databases.     See Less

https://www.youtube.com/watch?v=qZ_ew6XWMbc     See Less

'6:40' "it could parrot technical terms but lacks understanding".
Thats the fundamental issues with a     See More br>And it really makes them worse that useless, as when given power they make stupid mistakes, mistakes you might not notice until its too late.
Its quite scary how some people hype them, and try to give them more power in real world, when they are at best assistants to already professional people.    See Less

Gemini 2.5 pro recently told me that “4 is outside the range [0,7]”. This was part of an analysis of some code. Seems like some simple self consistency checking could avoid this kind of     See More r reasoning models.    See Less

Please don’t start a video that requires hi levels of focus and analysis and please don’t promote drinking alcohol to solve problems when your channel promotes AI for solving problems.     See Less

I think after benchmarks are fully saturated, one of the last milestones for AI will be when a model can set its own goals and design its own benchmarks.
Once it can both declare “I�     See More d this problem” and invent new challenges for itself, we’ll have crossed a decisive threshold.
It’s a bit like the way play helps children and animals acquire broad, transferable skills, where self‑directed exploration fuels deeper learning.    See Less

I watch these videos for the almost-too-subtle digs and pokes.     See Less