willy_k

> To me, I think these folks see that the music is getting harder to hear and might stop entirely

At the frontier companies, but not for hugging face or nvidia. Their business are going to be around for a while considering how local models are performing now.

> I would buy it and it was fun to do recreationally which supposedly means you don't have ADHD.

That’s something some people with ADHD need to hear to be ok with taking what is a powerful drug. A comorbidity of ADHD is substance abuse, and stimulants are very rewarding neuobiologically speaking (because they, especially amphetamine, promote dopamine release and persistence in the reward areas near the brainstem, as opposed to non-stimulants like Strattera which does effect dopamine reuptake but only in the prefrontal cortex, the primary area of impairment in ADHD [it does this by exploiting the fact that the NET is used for Dopamine reuptake exclusively in the PFC]), especially when treated like a recreational substance to be taken freely in a dose designed to be fun.

There is a lot of variety in people with ADHD, but I think that perception is more due to how the drugs work when taken medicinally vs recreationally and individual propensity for abuse.

NVIDIA is a lot more than a “lab”.

Because deleting model weights after training is likely to cause knock-on effects in model knowledge and/or behavior. Targetting it might mitigate this but it’s

a) not guaranteed that only censor-ey parameters get removed, and b) likely that removing those parameters still has effects on the effectiveness of related parameters.

> knowingly

Intent or negligence.

My understanding is that the current effort settings is a part of the system prompt, and that the levels and their intended results are a part pf the training process. The effect is more or less tokens spent reasoning before the model outputs a stop token. However it is as consistent as any other aspect of LLM behavior is..

IIRC responding to effort level settings appropriately is part of the (post)-training. In that case it could be considered another instance of the Bitter Lesson. Uplifting.

Is there a specific book from 40 years ago you have in mind? Asking out of curiosity.

Which is an opinionated rephrasing of the question whether ends justify the means. It’s not clear cut.

Not a direct response to your question; I figure you or others reading this may be interested in the thoughts of the host of The War on Beauty (best found on youtube).

Have you used the Google search engine in the past couple years?

You’re painting a dichotomy that doesn’t exist and hasn’t for a good while.

Edit: I missed your final paragraph. I still disagree with this argument though, people still want to go to websites.

That doesn’t matter very much for the free tier market though. The game is to have something useable that is actually used, and that can be offered at sustainable margins. Google has significant advantages in those regards.

Still, to get results from that you need someone who can create a decent finetune. That might not be realistic, but it very well could be realistic to have someone optimize some prompts and curate a knowledge base. Point is, its a possibility but “best” depends on the nuances of reality.

Ironically, or sadly (or perhaps a mixture of these+), this statement was absolutely informed (hopefully not inspired) by an LLM.

Best option by what metric? For which enterprises. I say this having worked at an “enterprise” where this was not a good option. For (lack of) talent/expertise, budget, infrastructure, and actual value relative to the eventual bottom line.

So the second option then.

That’s because a significant portion of the work and spending going on at frontier labs is generating new, more curated data, in select domains like software engineering and now biology[1].

[1] https://www.synbiobeta.com/read/anthropic-is-hiring-biologis...

Hopefully none of those hardened criminals know that a circular saw is the fastest way by far to cut through posts!

That sounds exhausting. Do things because they’re right or because you want to. Rules are fake (being a little dramatic there, but the point stands - rules create plenty of downside as well, while allowing both people to do the wrong thing with an excuse, and preventing people from doing the right thing without arbitrary consequences that are unrelated to actual outcomes).

Maybe because people still want to eat food. The culture of trying to re-engineer life is not it.

> It's a spiritual belief.

Yes. Thats part of human reality. Denying it is counterproductive at best.

Sci-fi is not reality.

No. That is not how AI works. It’s built by people, contrary to some of the more woo-ey hype. There’s no “agency”, or personal desires, or person to desire anything.

Thats some silly intellectualism, people are conscious and machines aren’t. We don’t need precise definitions or scientific rigor to know things.

LLMs are not conscious.

If you do any sort of homelabbing, it’s wonderful for making access from different machines easy and portable. Especially since you can leave it on and (unless you set an exit node) it will only route traffic meant for your tailnet through your tailnet, and use the internet otherwise.

When I wrote that I had in mind the idea of “mixed race” people. In the US it is very common for someone to identify as half black and half white, or one quarter Irish and three quarters Mexican, and so on. So, nuanced but not fuzzy, in that sense at least. But that is closer to ethnicity than race on second thought.

Yet, the models are objects built by humans. Any and all responsibility inevitably falls on the humans that built something with those flaws, and/or allowed them to exercise said flaws.

That’s very anthropomorphized language. It’s still a program operating under the constraints of the programmer. So agreed, don’t blame the AI, but it’s not clear at all that it’s even possible to “blame” an AI.

Well yeah, thats kind of what race is. It’s something you’re born with, and it’s not fuzzy. It can be nuanced, thats different. How it’s regarded is a separate issue, and that has generally improved over time.

Edit: I might be conflating race with ethnicity.