Hacker News: augment_me

New comment by augment_me in "Sebastian Raschka's from-scratch video course on building reasoning models"

augment_me — Mon, 26 Oct 2026 21:30:10 +0000

People are going to be upset about this, but any learning material in this day and age is solely made for your own understanding when writing it. The LLMs can be better teachers and understand your level of need and engagement better than anyone else.

New comment by augment_me in "If AI displaces human workers, they will lose more than just pay"

augment_me — Mon, 26 Oct 2026 15:16:27 +0000

Like the previous commenter said, people have dependents, mortgages, children, partners, etc. Percentually many more are tied up economically with the capital/state than in 1940s Germany.

Given the current economic trajectory there will be a line to put on uniforms, I don't think I am more eligible for this task than many others who have more dependents.

New comment by augment_me in "If AI displaces human workers, they will lose more than just pay"

augment_me — Mon, 26 Oct 2026 05:39:43 +0000

This is how I meant it, but the replies above might have attributed this to moral values somehow

New comment by augment_me in "If AI displaces human workers, they will lose more than just pay"

augment_me — Mon, 26 Oct 2026 05:33:16 +0000

I find some irony in that this hyperbolically late-stage neoliberal take is being taken as something I want in all but 2 replies

New comment by augment_me in "If AI displaces human workers, they will lose more than just pay"

augment_me — Mon, 26 Oct 2026 03:55:31 +0000

I think the economically rational choice is re-education of the group into value-providing employment and mass extermination of those that refuse.

Otherwise you might end up with disruptive things like protests, blockage of value creating infrastructure, etc.

Another less dramatic option would be containment in prisons or similar where minimal costs can be incurred.

New comment by augment_me in "Amazon’s “digital pacifier” for toddlers"

augment_me — Mon, 26 Oct 2026 03:30:55 +0000

Neoliberalism is at a state where moral shaming has very little effect. If you shame someone for trying to build a career or better themselves, you are the problem to a bigger extent than the person doing the previously morally bad thing. People's shame in a neoliberalist society comes from personal failures, such as not working hard enough, eating too much, not looking good enough, etc.

New comment by augment_me in "R/open source bans all solo projects and projects using LLMs"

augment_me — Sun, 25 Oct 2026 15:02:45 +0000

After pondering this for a while, I actually agree with this from the principle of community and socials. I have been in 4-5 open source communities where I met a lot of friends, but since LLMs the communities are flooded with slop PRs, slop projects, and the human interaction in general has fallen dramatically.

In some sense, if everyone can build everything, people's attention becomes the currency. If you are able to hold together an audience by banning automated tools, you might had suboptimal software but you might keep the actual goal.

New comment by augment_me in "No signs of AI payoff outside the tech sector"

augment_me — Sat, 24 Oct 2026 23:34:31 +0000

Can you demonstrate your take on 2) happening in practice?

- You have tokens priced differently as evident by for example pricing on for example OpenRouter.

- You don't reach 100$ of Fable/Mythos output by running Deepseek for the same amount of money. How do free open source models fit in here? Do you have to run Nemotron for an infinite amount of steps to reach 1$ of Deepseek compute? If not, then the argument is not valid because the pie should be shrinking.

New comment by augment_me in "No signs of AI payoff outside the tech sector"

augment_me — Sat, 24 Oct 2026 23:20:51 +0000

This again assumes a total monopoly or a cartel which does not exist today

New comment by augment_me in "No signs of AI payoff outside the tech sector"

augment_me — Sat, 24 Oct 2026 23:18:35 +0000

But the AI companies have no access or knowledge of how much value you produce, they would have to calibrate against some average SWE salary and some average productivity which is task-dependent. There is no way for them to control this. Even if they did, they would lose out to other companies who do not do this and let you "keep" surplus value, it feels like this again assumes a no-competition market.

New comment by augment_me in "No signs of AI payoff outside the tech sector"

augment_me — Sat, 24 Oct 2026 23:11:21 +0000

I don't really see the bloated-code take here. With this reasoning, there should appear a non-bloated open-source tool that all knowledgeable, smart developers will migrate to in order to get things done faster, and the market will have to follow.

Also waste and token spending could very well be a short-term thing. A lot of teams had crazy AWS bills in the mid-2010s on lambda functions and workflows, then it got cut down and optimized.

New comment by augment_me in "No signs of AI payoff outside the tech sector"

augment_me — Sat, 24 Oct 2026 22:01:50 +0000

Not sure how productivity and rent capture are connected here.

1) If a firm spends $99k on compute to replace a $100k engineer and produces 5x the output, how is that not a 5x increase in aggregate labor productivity?

2) In your aspirin example, why wouldn't a competitor with an identical 1¢ machine sell it for 50¢ or 10¢ to take 100% of your business? You are either making the assumption that 1 company is owning the market and no models are useable beyond that one, or that all of the models that are in the front of the market have a cartel and fix the price, which is not true either looking at API costs.

New comment by augment_me in "GPU-unlock: FP/BF16 instruction issue rate on Nvidia 3000's cards"

augment_me — Sat, 24 Oct 2026 18:09:22 +0000

I assumed the comment above is referring to operand dtype and not accumulator dtype since they mentioned ops. In general even for nvfp4 you use fp32 accumulators because you are usually reducing over large Ks which mean large values.

New comment by augment_me in "GPU-unlock: FP/BF16 instruction issue rate on Nvidia 3000's cards"

augment_me — Sat, 24 Oct 2026 14:51:42 +0000

These number claim 240W locked power draw on 4070:

https://github.com/JustAResearcher/gpu-unlock-ada/releases/t...

New comment by augment_me in "GPU-unlock: FP/BF16 instruction issue rate on Nvidia 3000's cards"

augment_me — Sat, 24 Oct 2026 14:49:21 +0000

The repo has fp16 numbers.

Also this one for 4070s shows that its for fp8 as well:

https://github.com/JustAResearcher/gpu-unlock-ada/releases/t...

New comment by augment_me in "How is music possible"

augment_me — Fri, 23 Oct 2026 20:29:41 +0000

Isn't the easier explanation that music/sound/rhythms has a social function for much of human history, and since humans are social creatures, we find safety and meaning in such sounds?

New comment by augment_me in "Kimi K3.1 Pro and K3.1 Pro Highspeed"

augment_me — Fri, 23 Oct 2026 20:19:26 +0000

If we ignore the older releases (sonnet 5, etc that exist in that one) and the new models epoch has not benchmarked, we end up with the same order within the noise brackets as in ArtificialAnalysis.

I do trust epoch more than AA due to capital incentives, but in this case I don't think like there is a difference

New comment by augment_me in "Kimi K3.1 Pro and K3.1 Pro Highspeed"

augment_me — Fri, 23 Oct 2026 19:44:35 +0000

Do you have an alternative? Its a benchmark collection. When a benchmark gets saturated, they remove it from the index and add new ones.

I think lm-arena is a better concept that reflects the actual target users, but it only captures chat, and relies on users spending their time improving it.

New comment by augment_me in "What Is Gridifcation"

augment_me — Thu, 22 Oct 2026 19:06:47 +0000

Suno and the AI models are actually better at generating jazz and classical music due to the smoothness of the waveforms.

My classical music friends have unbeknowingly been listening to a lot of AI classical music

New comment by augment_me in "Claude Duet"

augment_me — Thu, 22 Oct 2026 18:30:12 +0000

I am doing a PhD, and I have a loop that is autogenerating papers by pulling down the newest developments in the field from Arxiv together with the accepted papers from certain conferences.

Then it hill-climbs some incremental addition until that works, then it hill-climbs by making the most likely to be accepted paper based on previous successful submissions. Then autosubmits. I don't even need to read the research.

Got 4 conferences lined up next year, and 10 papers in queue from 80 ideas. This is research now.