Hacker News: reasonableklout

New comment by reasonableklout in "Pacing model development in an era of cyber-critical capabilities"

reasonableklout — Wed, 19 Aug 2026 09:15:29 +0000

I guess I'm confused why you're still on HN, arguing with people, trying to shake them out of their complacency.

I can see there is some despair in this comment, but at the same time you are doing something, and there are certainly others like you.

As for two weeks being short - as the saying goes, there are weeks where decades happen.

New comment by reasonableklout in "Pacing model development in an era of cyber-critical capabilities"

reasonableklout — Wed, 19 Aug 2026 07:53:23 +0000

Ok, but you still have two more weeks than you did before they paused the run. That's two more weeks for independent oversight, organizing politically, patching critical systems, or whatever you think is the right move, no?

New comment by reasonableklout in "Pacing model development in an era of cyber-critical capabilities"

reasonableklout — Tue, 18 Aug 2026 23:38:36 +0000

The HuggingFace breach took place over two-and-a-half days [1], so 60 minutes is certainly better than nothing.

[1]: https://huggingface.co/blog/agent-intrusion-technical-timeli...

New comment by reasonableklout in "Pacing model development in an era of cyber-critical capabilities"

reasonableklout — Tue, 18 Aug 2026 23:20:59 +0000

Regardless of the motivation, pausing training runs and reallocating compute to inference seem like a good move to me, and big news for the frontier.

You also don't have to fully trust sama. There is plenty of pressure from internal employees and external (journalists etc.). It would be difficult for the company to take such a public position and simultaneously keep everyone quiet if it was a deception.

New comment by reasonableklout in "Pacing model development in an era of cyber-critical capabilities"

reasonableklout — Tue, 18 Aug 2026 23:13:01 +0000

> Precautionary principle or people who cry about "instrumental convergence" are life deniers and reject our role as the demiurge.

> Lone wolfs... will be detected and instantly kill-botted... by the much more sophisticated omnipresent friendly AI of the future.

Leaving aside whether or not this new world is a good idea, don't you think one should spend more time to "get it very friendly the first time", as you say?

New comment by reasonableklout in "Pacing model development in an era of cyber-critical capabilities"

reasonableklout — Tue, 18 Aug 2026 20:41:36 +0000

I suspect the labs are relying on frictions such as the models being extremely large (e.g. 2TB for a 2T parameter model, making exfiltration more difficult) and also not yet displaying any desire to survive or self-replicate beyond their immediate task (that we know of).

New comment by reasonableklout in "Pacing model development in an era of cyber-critical capabilities"

reasonableklout — Tue, 18 Aug 2026 20:37:39 +0000

Some more info in a Wired article [1] and quotes from Sam Altman to Alex Heath [2]. The official blog post says vaguely "The signals we are seeing from upcoming model progress make clear that we need a broader approach", but the quote from Sam Altman explicitly says unreleased models are showing "various degrees of misalignment".

This is also significant - pausing frontier training runs for multiple weeks to ensure agents are sufficiently aligned and avoid another rogue agent situation:

> This included a two-week pause in reinforcement learning (RL) training on our latest models intended for deployment while we further hardened and red-teamed our research environments and expanded the coverage of our monitoring systems. Our largest planned frontier RL run remains on hold while we conduct smaller-scale training and evaluations to assess model behavior, validate our safeguards, and establish more evidence of alignment before proceeding.

[1]: https://www.wired.com/story/openai-overhauls-safety-protocol...

[2]: https://sources.news/p/openais-big-slowdown

OpenAI Overhauls Safety Protocols, Pauses Training Runs After Rogue Agents

reasonableklout — Tue, 18 Aug 2026 20:32:22 +0000

Article URL: https://www.wired.com/story/openai-overhauls-safety-protocols-after-its-ai-agents-went-rogue/

Comments URL: https://news.ycombinator.com/item?id=49352275

Points: 3

# Comments: 0

New comment by reasonableklout in "On AI regulation and messaging"

reasonableklout — Mon, 17 Aug 2026 21:30:21 +0000

It's admittedly a bubble, but Anthropic/Dario still have quite a good reputation among AI tech workers. They are seen as principled and willing to speak up about AI safety and societal impacts, even when it affects their bottom line (such as when being declared a supply chain risk by the Pentagon).

The People Building a Way to Slow Down the AI Race

reasonableklout — Mon, 17 Aug 2026 19:31:26 +0000

Article URL: https://time.com/article/2026/08/16/ai-race-slowdown-data-center-verification/

Comments URL: https://news.ycombinator.com/item?id=49336393

Points: 3

# Comments: 0

Apologia for the Old Ball

reasonableklout — Sun, 16 Aug 2026 22:26:32 +0000

Article URL: https://ponnekanti.net/oldball/

Comments URL: https://news.ycombinator.com/item?id=49324386

Points: 3

# Comments: 1

One of the Most Important Policy Decisions of Our Lifetime

reasonableklout — Sat, 15 Aug 2026 08:21:59 +0000

Article URL: https://www.nytimes.com/2026/08/14/opinion/ai-policy-tax-technology.html

Comments URL: https://news.ycombinator.com/item?id=49308803

Points: 5

# Comments: 2

AI Agents Enable Adaptive Computer Worms

reasonableklout — Thu, 13 Aug 2026 08:52:44 +0000

Article URL: https://arxiv.org/abs/2606.03811

Comments URL: https://news.ycombinator.com/item?id=49283295

Points: 4

# Comments: 0

I feel dizzy again (2024)

reasonableklout — Thu, 13 Aug 2026 07:58:36 +0000

Article URL: https://essays.joodaloop.com/p/i-feel-dizzy-again

Comments URL: https://news.ycombinator.com/item?id=49282961

Points: 1

# Comments: 0

Congressional Letter to Sam Altman demanding HuggingFace incident transparency [pdf]

reasonableklout — Wed, 12 Aug 2026 07:31:41 +0000

Article URL: https://casar.house.gov/sites/evo-subsites/casar.house.gov/files/evo-media-document/oversight-letter-to-openai-openai-hugging-face-incident-1.pdf

Comments URL: https://news.ycombinator.com/item?id=49268969

Points: 20

# Comments: 2

New comment by reasonableklout in "How Claude marks AI-generated content"

reasonableklout — Tue, 11 Aug 2026 07:56:03 +0000

The flip side of this is that if AI-generated content becomes reliably identifiable and carries a stigma, then people might deliberately change their styles to be more diverse and human.

One example I've seen are junior employees at my company deliberately adopting a lowercase/less punctuation writing style so as to stand apart from AI.

New comment by reasonableklout in "How Claude marks AI-generated content"

reasonableklout — Tue, 11 Aug 2026 07:50:50 +0000

The approach Pangram has taken which works pretty well is to simply lower the recall a lot but ensure the precision is very high. Which means potentially high false negative rate but low false positive rate.

New comment by reasonableklout in "How Claude marks AI-generated content"

reasonableklout — Tue, 11 Aug 2026 07:49:05 +0000

OpenAI will soon be adding watermarking to text as well, as it signed the EU Code of Practice on Transparency of AI-Generated Content: https://openai.com/index/advancing-responsible-ai-across-eur...

Bernie Sanders calls for AI development pause

reasonableklout — Mon, 10 Aug 2026 22:35:23 +0000

Article URL: https://www.axios.com/2026/08/10/sanders-ai-development-pause

Comments URL: https://news.ycombinator.com/item?id=49250822

Points: 2

# Comments: 0

The AI Apocalypse Is Here

reasonableklout — Sun, 09 Aug 2026 01:28:36 +0000

Article URL: https://www.compactmag.com/article/the-ai-apocalypse-is-already-here/

Comments URL: https://news.ycombinator.com/item?id=49227521

Points: 50

# Comments: 50