Sriram Krishnan

Tomorrow should be a huge day for American AI. 🇺🇸

it's key for the US to win on open weights/open source models. It's very clear there is a distinct need for this.

Learning what it means to be “bitter lesson pilled”. Not a week goes by when I’m not sending someone Sutton’s essay.

An important thing to track is: marketshare of America AI vs Chinese AI. Right down to tokens / month inferenced and the models + hardware + stack used.

Struck by how similar Goodhart’s Law - familiar to anyone who has worked in a corporate job- is to reward hacking / gaming as a problem in RL. Though one key difference perhaps is in Goodhart’s Law it is the very act of making a metric the target that makes it cease to be a…

Really good post from @dwarkesh_sp on continuous learning in LLMs dwarkesh.com/p/timelines-ju… Also see @karpathy response to this.

What are good ways to get kids under 8-9 into RTS ? Just rediscovering AoE ( AoE 4). Open to ideas of simple games but with core RTS mechanics. Factorio is more construction but similar enough in some mechanics ( tech trees, economy , supply chains).

I want to create a 3d model of my living room ( so our kids can walk around it on a computer). Say something to play with in Blender/Unity/Unreal. What’s the best app / model / toolchain to do that easily?

Think @dwarkesh_sp is onto something with “AI management” and seeing the future Silicon Valley organizations being much better run. Already seeing some small moves towards this. dwarkesh.com/p/ai-firm

A lot of people are pointing to @tobi’s AI memo and one easily overlooked part is - *learning to use AI well is a skill* I’ve learned a ton when I watch people use these tools and especially those who have gone all in on using them seriously. Deep knowledge of how to use all…

Now have an official account - @skrishnan47. Do follow for all policy work on AI. Will try and talk more about everything from pro wrestling to gaming here!

spending the weekend vibe coding.

find myself wanting this everyday. feels like a fundamental limit still of what you can squeeze into context. some good responses in the thread below.