Excited watch @dwarkesh_sp live!

If I could only give people one tool for LLM Evals, it would be error analysis. Nothing else comes close This is what Look At Your Data ™️ means Links in reply

"Move 37" is the word-of-day - it's when an AI, trained via the trial-and-error process of reinforcement learning, discovers actions that are new, surprising, and secretly brilliant even to expert humans. It is a magical, just slightly unnerving, emergent phenomenon only

I don't have too too much to add on top of this earlier post on V3 and I think it applies to R1 too (which is the more recent, thinking equivalent). I will say that Deep Learning has a legendary ravenous appetite for compute, like no other algorithm that has ever been developed

Easy prediction for 2025 is that the gains in AI model capability will continue to grow much faster than (a) the vast majority of people’s understanding of what AI can do & (b) organizations’ ability to absorb the pace of change. Social change is slower than technological change.

Good day to remember just how big that green tree is

@lhl @swyx Yeah I have been using byobu. Started using it many years back when @mussondata showed me.

I have now had multiple instances where professors have told me that they are pretty sure that o1 found something novel in their field, but that the results are technical, non-obvious and complex enough that they can't be sure without considerable checking Interesting challenge

We're releasing a preview of OpenAI o1—a new series of AI models designed to spend more time thinking before they respond. These models can reason through complex tasks and solve harder problems than previous models in science, coding, and math.

soon ai will eat absolutely everything, but right now there's a sweet spot where you can leverage your domain knowledge in conjunction with ai to make something useful with minimal financial or labor investment. last chance to join the upper class before takeoff maybe

this is the most interesting year in human history, except for all future years

Stripe released their 2023 annual letter this week. I thought it would be a fun project to try out OpenAI's text-to-speech API so I could listen to it like an audiobook. OAI's TTS capabilities are amazing! Take a listen here:

Of the OpenAI Sora videos, this one blew my mind. Rendering this scene via a classic renderer is very hard. Sora doesn't model physics the way we do. It can definitely still get it wrong, but I wouldn't have predicted it'd be this convincing. https://t.co/l3blB8Xe49

this was one of those things you tell yourself is coming and you think you are ready for it and couldn't possibly be surprised by it, but then you see it and don't quite believe it and you're not sure why you didn't think you'd be surprised excited to share our text-to-video

Announcing Sora — our model which creates minute-long videos from a text prompt: https://t.co/SZ3OxPnxwz

In the latest @dwarkesh_sp podcast, @tylercowen shared an interesting thought: Some people will be good at delegating tasks to AI, and some won't be. This will lead to some people becoming much more productive than others.

I think GPT-4 still has a long runway of adoption ahead (both in tool-integration and new user adoption). It's just such a capable model that can do so many things. The interesting question is: What happens when even more capable models are released?

I feel like you have to do this when you visit this building 😅?

The biggest benefit GPT-4 gives for coding tasks is simply reducing cognitive overhead IMO. It removes instances of "ugh I know how to do this but it's going to take a bit of thought and 20 lines of code." Instead I just ask GPT-4 and it gets me 95% of the way there.

humans built machines that talk to us like people do and everyone acts like this is normal now. it's pretty nuts