Hacker News: jrflo

New comment by jrflo in "Strengthening our defenses against distillation attacks"

jrflo — Tue, 13 Oct 2026 04:56:09 +0000

Copyright is meant to protect against reproducing a work, and something that is transformative should be fair use. If it was free on the internet, they don’t need permission anyway

New comment by jrflo in "AI models can provably hide arbitrary capabilities"

jrflo — Tue, 13 Oct 2026 04:33:38 +0000

There would not be the same incentive, closed models wouldn’t want to steal from / Trojan agent swarm their paying customers

New comment by jrflo in "Strengthening our defenses against distillation attacks"

jrflo — Tue, 13 Oct 2026 04:28:03 +0000

They did not steal, they used free information on the internet. And, they invested billions to turn that raw information into something useful, it did not have the same inherent value before being transformed into a language model.

A better analogy would be if I read an open scientific paper and used the result to design some device. I put up the money and effort to turn that work into something useful, but someone did the ground work and it was released for everyone to see.

New comment by jrflo in "Strengthening our defenses against distillation attacks"

jrflo — Tue, 13 Oct 2026 04:11:19 +0000

Idk, as an engineer who makes physical products it sucks having an overseas company take the design you spent months on, copy it, and resell at a cheaper price. All that money and effort into R&D and someone steals the answer at the end. I understand their incentives but it’s a shitty practice.

New comment by jrflo in "Strengthening our defenses against distillation attacks"

jrflo — Tue, 13 Oct 2026 04:05:06 +0000

Who were these hundreds of organizations with powerful AIs?

New comment by jrflo in "Strengthening our defenses against distillation attacks"

jrflo — Tue, 13 Oct 2026 04:03:39 +0000

Yes, it is reductive. HN is unreasonable when it comes to defending distillation, the billions labs pour into transforming raw information into a model is transformative work that they should reasonably try to protect. Folks just want to have free stuff and jump through hoops to defend that view.

New comment by jrflo in "Strengthening our defenses against distillation attacks"

jrflo — Tue, 13 Oct 2026 03:58:33 +0000

1) It probably scales with model capability, distilled models are inherently less capable and therefore less able to commit a crime

2) If the open providers are truly just distilling bigger models, the distillers have no reason to do the crazy levels of training and evaluation that would result in a crime being committed

They are definitely spinning this for PR and their own self interest but open models are genuinely more unsafe since it’s a Pandora’s box situation- if you have a very capable hacking model there are plenty of bad actors with access to <$100k in compute resources that could raise hell with a jailbroken model. At least with Anthropic they can restrict access to bad actors.

New comment by jrflo in "Just say no to OpenAI and Anthropic"

jrflo — Mon, 12 Oct 2026 19:20:26 +0000

But you're still trusting the company to actually follow those practices, which is what I'm getting at. The author assumes that OpenAI and Anthropic won't follow their stated practices, then assumes Tinfoil will follow theirs.

New comment by jrflo in "Just say no to OpenAI and Anthropic"

jrflo — Mon, 12 Oct 2026 18:55:19 +0000

If you don't trust OpenAI and Anthropic to follow their own policies, how do you decide what companies to trust? Vibes?

New comment by jrflo in "Research Firm Calls Anthropic's IPO 'Unprecedented Test of Investor Gullibity'"

jrflo — Mon, 12 Oct 2026 15:21:28 +0000

People keep talking about the risk of open source models, but if you're outside the HN/SV bubble, no one is using them. Anthropic's revenue is ~$60B supposedly, OpenRouter's revenue is $140M. I know they're not the only place to get open source models from, but they're the most popular and Anthropic is bringing in 400x more revenue than them.

New comment by jrflo in "An open-source, clean-room reimplementation of Microsoft Word in pure Rust"

jrflo — Mon, 12 Oct 2026 03:07:33 +0000

While they may be a buggy mess, I think it's really interesting that we're starting to see this unfold in real time. People have been forecasting these kind of changes for the last few years, and it seems like it's actually happening now. Really curious to see where this goes, if everyone has their own custom fork of some common codebase for specific features/needs they have across these base applications.

New comment by jrflo in "An open-source, clean-room reimplementation of Microsoft Word in pure Rust"

jrflo — Mon, 12 Oct 2026 03:04:28 +0000

I see what you're saying, but we're just getting to the start of it being worthwhile to do something like this. Once these projects have people testing them and proposing changes/improvements (AI driven), I don't think it would be long until they are on par with, if not better than the originals.

The only catch will be if attention coalesces around an individual project instead of hundreds of people vibecoding their own versions.

New comment by jrflo in "The gap in shared understanding of LLM capability is widening"

jrflo — Mon, 12 Oct 2026 02:58:30 +0000

Interesting idea, but I'd have to disagree. MENSA is basically a group of benchmaxxed people, they are super good at a set of testing but that doesn't mean they have the ambition or other social skills to make a difference in the world. When you align all those factors, you find the greatest contributors to science and engineering. What happens if you completely eliminate the need for someone to be inherently intelligent? Suddenly, the number of important difference makers in the world increases by an order of magnitude.

I agree with the sentiment though, intelligence is not all you need. We're not going to "solve science" through RLVR, people are still needed in the loop until we get superhuman construction robots and superhuman communicators.

New comment by jrflo in "The gap in shared understanding of LLM capability is widening"

jrflo — Mon, 12 Oct 2026 02:48:16 +0000

AI really struggles with "collision with reality". Computers suck at modeling the real world in a detailed enough way to make scientific progress, aside from small niches like drug discovery where you have a very well scoped problem. If any part of the try something -> test it -> look at the results loop requires hitting the real world, AI is going to struggle.

On the other hand, the areas where it doesn't need to hit the real world, progress is absolutely flying. I would not be surprised if AI is superhuman in math and physics in a year, considering we went from homework helper to IMO Gold to Millennium problem solving in a couple years time. I think we'll start to see more meaningful scientific progress in maybe 2-3 years when scientists have access to an oracle that can formulate theories and direct real-world research better than a human can.

New comment by jrflo in "The gap in shared understanding of LLM capability is widening"

jrflo — Mon, 12 Oct 2026 00:45:29 +0000

I don't think 99% of people have any application for AI as it is today short of personal assistants. But, if we can have recursively improving systems for that 1% of people that are responsible for scientific and technological innovations, that's going to be transformative for everyone.

New comment by jrflo in "The gap in shared understanding of LLM capability is widening"

jrflo — Mon, 12 Oct 2026 00:42:50 +0000

I think that's a bit reductive, there are decades of sci-fi novels about how transformative artificial intelligence would be, so I feel like I don't need to explain where the trajectory could go. The applications and "collision with the real world" aren't there yet, but you can't pretend that a machine that's smarter than any living human is on the same level as Amazon drone deliveries.

New comment by jrflo in "Terence Tao: Math 2.0 [pdf]"

jrflo — Sun, 11 Oct 2026 16:23:06 +0000

There's a difference between "no one understands it" and "no one has tested and validated it's safety". You don't need to fully understand the mechanics of something to test it and validate it. Plenty of things in the real world work this way already.

New comment by jrflo in "Scandinavia's first farmers slaughtered hunter-gatherer population, DNA suggests"

jrflo — Sun, 11 Oct 2026 14:01:33 +0000

Why is “nature” being intact so important? Humans are part of nature just like every other living thing, even today.

New comment by jrflo in "OpenAI withdraws three mathematical results"

jrflo — Thu, 08 Oct 2026 16:54:50 +0000

You only need the problem statement to be correct in Lean, no matter what route it goes down is correct as long as the original formalization of the problem is correct, which is substantially easier to check. Not sure how many people have checked that so far, but I'm guessing the math community would be quick to point out if something basic like that was missed for something like a new bound on RH.

Nothing wrong with being skeptical, but I see no reason to be skeptical as of yet.

New comment by jrflo in "OpenAI withdraws three mathematical results"

jrflo — Thu, 08 Oct 2026 15:59:33 +0000

None of the papers withdrawn were formalized in Lean, only about half the papers in the repo are formalized. I don't think we yet have an example of what you're suggesting actually happening.