I wonder if there's some amount of on-prem going on there

It might still be a little slow (I'm not sure if the 16 seconds to compute an action is fast enough for commercial use cases), but this is definitely exciting and seems like a great step forward.

Like others have said, its an interesting avenue for AGI. The joint embeddings would be closer to thinking than the current LLM token work. LLMs look like they have a lot limitations for AGI (although who knows if we have another crazy scale up? but that extra scale is looking difficult right now).

I really hope it's the latter, that would be quite brilliant

I don't mean intentionally.

I mean maybe he made that repo because he'd given up on his coding dreams. Almost.

That would be some interesting accidental meta commentary on the state of things.

Absolutely worth it. LLMs have saved me hundreds, if not thousands of hours already. They easily more than 3x my productivity, using pretty objective metrics like # of features shipped, etc.

Is there a particular one you pay for pro on? Or do you pay for several models?

Cursor and gpt-4.1 or o3. I hardly ever write code myself anymore.

I say this as someone with a bachelor's in CS from a top school and years of experience at the most selective companies.

Maybe I'll get back on the subscriptions then... appreciate the perspective

While I tend to agree, I wonder if synthetic data might be reaching a new high with concepts like Google's AlphaEvolve. It doesn't cover everything, but at least in verifiable concepts, I could see it produce more valuable training data. It's a little unclear to me where AGI will come from (LLMs? EBMs - @LeCun)? Something completely different?)

I know this is the terminology, but I'd argue that the activations are the actual thinking. It's probably too late to change that, but I wish people would refer to thinking as the work Anthropic and Deepmind are doing with their mech interp