naasking

> amazon would publish every single boon without giving a cent to writers. it would be reqlly hard for writers to make a living.

Not quite, Amazon would just become the publisher that buys new books/articles/whatever directly from the writers, one upfront fee. Not saying it's a sustainable living for writers, just saying that's probably what would happen.

If it's consistently bad with no signs of improving, banning makes sense. People made mistakes even before LLMs though, and you'd only ban them if they didn't improve and so were a giant waste of time. I don't see why that wouldn't translate here.

> I'm also concerned these things will mostly be used by Big Tech for their A.I. data-centers

Why is that a concern?

> I strongly disagree with this take. AI does not have the theory of mind to manage complex social environments

Does Musk have a complex theory of mind to manage his employees, or does he just have a brutally efficient organizational strategy and gets rid of anything that doesn't fit? I think it's more the latter, and an LLM can absolutely do that.

This guy claims 6% throughout hit for this approach:

https://x.com/0xBakeer/status/2092694905978237224?s=20

Crazy how fast things move these days.

That probably includes the 51b ngrams too. It's possible that those could be streamed from NVMe on-demand. The Engram paper that developed this technique streamed from RAM to VRAM at only ~1% performance degradation, but these strix halo boxes and the spark have much slower memory, so it's possible moving down another rung on the memory hierarchy wouldn't affect their performance too much.

This will almost certainly require changes to llama.cpp or vllm to do it right.

You can read posts you link to directly, but no comments. That's what makes it less useful.

Linking to X is less useful than it used to be.

Because nutritional value is not the only thing of value. We invented this clever way of uniformly valuing things called "money" that abstracts over all of those differences though.

> But that's tripled of water use across all sectors due to datacenter buildout. So no "comparison" to other uses is needed.

You're saying almond farm water use tripled due to data center buildout? What other meaning of "all sectors" could make sense of this claim? The report clearly says 17 billion while all other sectors are in the trillions.

> Yes. Actual real people believe that LLMs are actual, thinking, intelligences, perhaps even with consciousness.

Yes, because it's possible. They don't have human consciousness/thought/intelligence, but it's entirely possible they have some form of consciousness, some form of real thought and some form of intelligence.

Yes, government regulations are almost always behind commercial entities making seemingly irrational choices.

I don't think "don't overthink" is a useful guide, it's too ambiguous which leads to the model overthinking how to interpret the advice to not overthink.

I think guidelines have to be specific, such as "provide only the specific properties specified, and if other properties are required but are not specified, then ask for clarification or pick the simplest default possible". Even that's ambiguous on "simplest default possible".

I would be more specific in specific domains. The circle example from the article is clear that it defaults to elaborate stylistic choices rather than simple colours and shapes, so the prompt would have to specify "use only simple colours and the simplest possible shape that fulfills the request".

To what extent can the overthinking be reduced by a better system prompt? For instance, telling it not to overthink, embellish or snazz anything up, and just directly deliver the most what was requested without in the plainest form to build on in future iterations.

You'd need hundreds, if not thousands, of parts to justify a mold for casting. And even then, by the time you got the mold made you could have already printed hundreds of parts on a print farm, so you're losing time too.

Can anyone confirm whether this new Qwen release is any more concise when thinking? Overthinking was the biggest (only?) downside of the Qwen models.

You know free LLMs can help you understand that command line or design your own...

You need to do some reading:

"Is raising mental health awareness causing more harm than good?", https://www.clinicaltrialsarena.com/analyst-comment/is-raisi...

"Are We Overdoing Mental Health Awareness?", https://www.psychologytoday.com/us/blog/silicon-psyche/20250...

No, in the past you just had the possibility of childhood illness and death, the death of siblings, war, famine, drought, none of which are real problems for university students these days.

What's less certain about modern environments compared to past ones?

The coddling is a big problem these days, and this focus on mental health is itself unhealthy. The assessment that "they would otherwise do well" is speculative, at best.

People say Qwen overthinks because they analyzed the thinking traces, and Qwen finds the answer relatively quickly but then second guesses itself multiple times for another 20,000+ tokens. Regardless of what other models do, that's clearly overthinking.

The continued vagueness suggests this isn't going anywhere, so I'll just conclude by saying that you've given nobody any real reason to believe this.

No such thing has been shown.

Typical LLM pretraining is very inefficient with data. NanoGPT slowrun shows that data can be used much more efficiently.

I don't understand all this dismissal of Einstein. What you just described is that he published a unified theory of Brownian motion, developed a new way of conceptualizing scientific theories via principles with which he derived special and general relativity, contributed foundational work to quantum mechanics, and more, but no, he was not one of humanities' greatest scientists.

I use LLMs regularly for real work. I understand the current limitations, but you're talking about existing limitations that will never be overcome, because they cannot be overcome. I'm just asking for a couple of specific examples, and your evasions are increasingly suspicious.

Yes, it's worse because processors clocks are much faster than memory clocks.

Yes. That's why law is a different discipline from linguistics requiring its own rules of analysis, pedagogy, etc.

That's just vibes then. How is that supposed to be convincing?