This chart should be framed and put in the office of every US member of congress.

CMU researchers, in collaboration with NVIDIA, present ASAP, a two-stage framework for humanoid robot agility. It pre-trains motion policies on human data, then refines them with real-world corrections using a delta action model, which adjusts for simulation mismatches.

The most succinct explanation of how you should think about owning stocks.

This would be a constant across all factories, especially and firstly newly built ones which is forefront here. The huge impact of this aside, doesn’t diminish the onshore arguments and if anything mutes all the “we’re too good to build crumby stuff here anyway” talking points

it really is incredible what kinds of things become possible when RL on LLMs works. clearly we’re just getting started

o3 pro is unusable

I’m just saying, airport drunk is different. I donno

Elon Musk Grok The problems in our country stem from an inability for the citizens to adequately express their preferences in voting using a first past the post process. This system will always decay into a duopoly with no incentives to do anything but stay in power. It’s not a new party you

Startup idea: Tinder but for guys to find other guys to do cool projects with

DJ_Dave live coding music made with Jade rose(_switch_angel)

o3 pro is definitively worse than o3 at 10x the cost and 3x the time. Hasn’t gotten better since release yet

Grok 4 heavy is actually very good. The projects thing is completely broken. But the normal chat interface is extremely coherent and impressive for complex code and research. Especially considering it’s day 1

Grok is throwing way more compute at me than open AI can. Open AI are too popular for their own good. This must be dynamically adjustable to match load and demand… the actual engineering for these flagship models must be sumthin huh. If your getting half a billion a year I

Guys I think I’ve found the pinnacle of Motorsports

The driving models currently in openpilot aren't very smart. They struggle with nuanced cases like construction. We've been scaling them up to make them smarter. Here are some examples of how much smarter the new models (right) are compared to current release (left)!

Context Rot is an excellent term and should be used more often

This was animated in 1987

undefined behavior The non-linearity she tells you not to worry about

You’re gonna wanna acquire 12 copies of this, go into debt if you have to.