MasterScrat

I’d be curious to see how reliable it is in practice, it sounds like a cheap and (hopefully) easy benchmark

I fully agree with this quote, it really crystallized something i felt but couldn’t put into words.

I don’t think it’s a good analogy.

Having your fly open is a harmless mistake that has little impact on your peers. People may or may not mention it to you but it’s not something they’ll hold against you.

Posting LLM slop under your name is a deliberate act. You decide to damage your message by taking a shortcut.

The obvious analogy was speaking while chewing.

I don't think this is bad. I want tools that deliver what I envision. The fact that given the same input, I get similar output is something I would rather consider a positive. The imagination should come from the human at the wheel.

> he just read out something like "You're right to push back on this. If you'd like, I can reword this..." and didn't even flinch

Can you please provide sources for this? My understanding is that he used the expression "I appreciate the pushback" in a context in which it made sense (responding to criticism), and people flagged that it sounded LLM-like.

There’s some poetry seeing people complaining about how LLM provide unreliable information in a confident tone, who then go on to share incorrect information in a confident tone.

I don’t see how this market ("mostly" AI-written books) won’t get arbitraged to death with the cheaper and open-weight models that keep coming out

Damn the pelican guy can’t get no sleep

> xAI is renting out their DC space. I think I read that Google is, too.

You’re probably thinking of Meta. Google renting GPUs is a less recent phenomenon, to say the least.

From HF:

> The practical lesson for defenders: have a capable model you can run on your own infrastructure vetted and ready before an incident, both to avoid guardrail lockout and to keep attacker data and credentials from leaving your environment.

This sounds like a reasonable measure. Any recommendations regarding the most suitable models for this? GLM? Kimi?

How is what you describe "YC-ish"?

Playing backward is actually a great idea for testing generalization

Would be a great idea to see how much we could manipulate the latent space and whether it has some internal structure w.r.t the physical state. I guess the only unknown is how the world model would show robustness to latent states that are transformed through this network

Hey all, happy to see this here! This was a colab between General Intuition (that I’m part of), Kyutai and Epic Games.

You can read plenty of details in the blog post and tech report but the TLDR is that we trained a multiplayer world model on 10k hours of Rocket League data. We optimized it to be playable at 20fps on a single GPU.

So what you see in the demo is fully generated: there’s no graphics or physics engine. Instead it’s a 5b neural network that takes actions in and gives pixels out.