Plunder a country

Kidnap their people to make slaves

Impose anti-gay laws on them

200 years later: They are expanding the anti-gay laws! Why are they so backward?!

Pay reparations or shut up.

They said they're working on Orthus for Qwen 3.5. It'll be amazing!

My oversimplified and possibly wrong understanding: this is like speculative decoding, but instead of a separate draft model (which does its own prompt processing), they use some diffusion thing strapped on top of the main model. The diffusion reuses the high-quality prompt processing result of the main model.

The 7.8x faster claim sounds almost too good to be true. But even if we get like 3x then this is still a huge revolution in localLLMing.

1.5 hours runtime for like half a liter of gasoline?? That's unbelievably inefficient. A half-liter of gasoline is like 15MJ, should power a laptop drawing 30W for a week.

Maybe it would be better with a fuel cell.

Reinforcement learning makes the model better over time, so why should there be fewer and fewer good results?

If you're talking about the rate of improvement going down, then yes, of course. That's bound to happen (unless you have an actual intelligence explosion, but in that case you won't know what "good results" even mean anyway).

Yes :)

No one feeds random LLM output straight back though. The whole idea of reinforcement learning is you take some ML model output, check if it is good, and push the model in that direction if it is good.

As long as you believe that e.g. it's easier to verify a mathematical result than to come up with one, then RL should work.

You took those quotes wildly out of context. Of course there is a hard limit on how much information can be extracted from data. Clever processing won't break that limit. But only in basic cases have we seen proofs that certain statistical inference methods make optimal use of the data. In complicated systems like neural nets it is basically impossible to prove such optimality. In fact the models are almost definitely not using the data optimally. Processing can help. A lot.

How does the crosspost listing on Lemmy work?

The original post (just 10 minutes earlier) is here but is not shown in the list of crossposts for this post. Why?

On a separate note, I hope the author of this post and his colleagues are safe. His university got bombed by America and Israel earlier this month.

This is great! The value of having your computational pipeline nicely debuggable and visualizable cannot be overstated.

The float32 limitation sounds a bit arbitrary. Would be cool if blender allows float64 in geo nodes in the future.