Forgot your password?
typodupeerror

Comment Re:Anthropic PR (Score 1) 44

I'm not very up on this stuff, but I think(?) he's saying Hugging Face ("HF" in his comment) was considering writing a common API for various different kinds of hardware, to replace CUDA.

And that would be awesome, but also very threatening to Nvidia, which is probably part of why they bought Hugging Face: to kill such a project.

Comment Re:Future tense [Re:FARA Foreign Agents Registrat. (Score 1) 76

I like to think the delay is justified by the fact that IF you think in terms of 1800s tech, America is huge and it can take a while for horses to pull a wagon from sea to shining sea.

It's November 1860 and you just won your election to be the new Senator from Oregon. You have two months to get to DC. Can you do it? (Yes, you can do it in a "mere" 3-4 weeks thanks to rail. But before rail? No way two months is enough.)

Comment Re:Things are going to get pretty messed up soon (Score 0) 44

The first thing will probably be to do something think of the children bullshit to ban local AI instances.

Are you seeing a lot of support for the "people should be forced to use a data center to run AIs, instead of being allowed to do it at home using solar power" position?

What I saw last week, is that "think of the children" means that no one should ever fucking ever let a Republican anywhere within ten feet of a child, rather than it meaning we need to give data centers a monopoly on running AI.

The RAM crisis is it. This is the high-water mark on suppressing competition. It was a clever move, but the consumer market isn't going to tolerate electronic gizmos and video games being expensive forever. As soon as someone decides they want money and opens another RAM factory, the AI companies are in trouble. They have this brief pre-Personal-Computer-Revolution situation going on like 1960s IBM, and the 1970s are coming to fuck their shit up.

Comment Who is buying them ? (Score 3, Interesting) 59

Boston Dynamics has always made the most capable humanoid robots, but it's not clear what the market is for them. Their dog-like Spot seems to sell to police forces and military and serves a purpose to go where no human can safely go.

Musk is projecting a market in the millions for his TeslaBot, but various Chinese companies like Unitree already have humanoid robots for sale and don't seem to be finding much of a market for them yet outside of quirky hospitality/entertainment usage. Some companies evaluating them for industrial use, but AFAIK no large scale adopters.

Comment Accusations? (Score 2) 18

The law doesn't explicitly define the term repeat infringer, nor does it say how many piracy accusations it takes before a subscriber can be cut off.

[emphasis mine] Surely they mean convictions, no?

Why would accusations be relevant? I accuse every single member of the Mexican government of having violated the copyright on my work, four times each. They all did it four times. No, I can't prove it, but the accusation stands.

Comment Re:It is not AI (Score 1) 58

It's important to keep using accurate language if you want to think clearly about things.

That sounds like a good idea, until someone says the word "hacker" and then you realize you lost the debate decades ago.

Words' meanings change, and nobody can ever do anything about it.

Comment Re:correcting repetitive reasoning? (Score 3, Interesting) 45

Weight values control what tokens get predicted/emitted.

Reasoning (to be understood in context as LLM regurgative reasoning) is ended by emitting an "end-think" token, otherwise it will continue.

The overly dramatic headline might have fooled you into believing that Astra was programming Qwen directly, as a possible alternative to training, but if you actually read the article the reality is just that Astra fixed a bug in Qwen so that it predicted (emitted) the "end-think" token at the appropriate time.

Comment Re:Not as impressive as it sounds ... (Score 1) 45

The defect was noted on Qwen's model card - there is no indication that Astra figured out for itself. What Astra did was apply it's mech interp knowledge to locate and fix the source(s) of the bug.

Saying that Astra boosted Qwen's math score is like saying that I boosted the top speed of your car when all I really did was fix your flat tire (which did allow you to drive faster).

Comment Not as impressive as it sounds ... (Score 1) 45

Astra didn't do anything to specifically improve Qwen's math ability, but rather fixed an unwanted behavior of the model, noted on it's model card, of it continuing reasoning for too long when it should have stopped - failure to emit a token at the right time.

A less dramatic headline would have been "Astra fixes Qwen bug by modifying its weights". This is mechanistic interpret-ability stuff that researchers already know how to do.

Still cool, but isn't AI making AI smarter the way the headline invites you to think. It's a bug fix, but in a model's weights rather than in code.

Slashdot Top Deals

Take everything in stride. Trample anyone who gets in your way.

Working...