Hacker News Reader

jefftkon U.S. government will decide who gets to use GPT-5.6

Limiting exports of AI services to all foreigners is probably allowed under GATS, since there's no favoring one county over another. But even then there's a national security exemption, which fits reasonably well with US arguments here.

If a country thought the AI export restrictions were inconsistent with treaty the remedy is challenging them, not unilaterally imposing their own tariffs. But even if they got a favorable panel decision the US would speak, and the Appellate Body is non-functional because the US stopped consenting to the addition of new members and all the terms expired. Which means anything that gets appealed is frozen indefinitely waiting for the AB to reach a quorum that won't come until the US changes is mind (and Biden didn't reverse Trump's decision here).

jefftkon Statement on US government directive to suspend access to Fable 5 and Mythos 5

You very likely know this, but to make it explicit: "US Persons" under ITAR is US Citizens + Lawful Permanent Residents (green card) + Protected Individuals (Non-citizen nationals like Samoans, Refugees/Asylumees). It doesn't include anything else, like H1B, TN, etc visas.

jefftkon Waymo Premier

I would expect tickets issued during the first 40 days to be higher than later, as people haven't adjusted yet

jefftkon Anthropic requires 30 day data retention for Fable and Mythos

But if you're going to take your distrust that far then the issue is that they have your data at all, not that they are telling you that they will retain it for 30 days.

jefftkon Claude Fable 5

This is not true in SecureBio's case, and I really doubt it's true generally.

jefftkon Apple decided not to roll out Siri in EU after denied request for exemption

Laws and strategy are not fixed, and public bickering is part of how they get optimized.

jefftkon YouTube to automatically label AI-generated videos

Deepfakes in VFX is another borderline one.

jefftkon YouTube to automatically label AI-generated videos

No appeals combines very poorly with any detector that sometimes has false positives.

jefftkon Google changes its search box

People are certainly welcome to feel a lot of different ways, not trying to be prescriptive here. My parent asked: "what exactly do I gain by allowing Googlebot to crawl my sites?" and I was describing what I get out of it, in the hope that others might feel similarly.

jefftkon Google changes its search box

I'm not interested in a book, speaking tour, or podcast. I've never had consistent readership because I write about too many unrelated things. I blog because I have ideas I want to share; I don't feel at all ripped off.

jefftkon Google changes its search box

I write things on the internet because I want to share ideas. If someone reads my post and tells a friend, that's great. If an AI crawls my posts and passes along the ideas that's great too.

(It doesn't work for ad-funded writing, but while I have substantial sympathy there this has historically been an unpopular argument on HN)

jefftkon Googlebook

I also did a bunch of shopping with AI to identify clothing recently. I was going to DC for a bunch of meetings, and did not have a good sense of what clothes are appropriate in different DC contexts. I did a bunch of iteration with AI to identify something that communicated what I intended, and then ran the final list by a friend with more context to confirm that it was indeed a readable choice.

jefftkon AI is breaking two vulnerability cultures

These are very clearly vulnerabilities in the normal sense of the word, and if a security bug means that an app that was supposed to be only accessible to the creator is open to the world that's still quite bad (though the blast radius is small).

If you limit to vulnerabilities that get CVEs, however, https://vibe-radar-ten.vercel.app has 34 in March alone including https://www.sentinelone.com/vulnerability-database/cve-2025-...

jefftkon AI is breaking two vulnerability cultures

Security researcher Dor Zvi and his team at the cybersecurity firm he cofounded, RedAccess, analyzed thousands of vibe-coded web applications created using the AI software development tools Lovable, Replit, Base44, and Netlify and found more than 5,000 of them that had virtually no security or authentication of any kind. Many of these web apps allowed anyone who merely finds their web URL to access the apps and their data. Others had only trivial barriers to that access, such as requiring that a visitor sign in with any email address. Around 40 percent of the apps exposed sensitive data, Zvi says, including medical information, financial data, corporate presentations, and strategy documents, as well as detailed logs of customer conversations with chatbots.

https://www.wired.com/story/thousands-of-vibe-coded-apps-exp...

jefftkon AI is breaking two vulnerability cultures

How would you apply this logic to something like https://meltdownattack.com ? The vulnerability was in hardware, discovered by companies that make user level software, and mitigated by changes to OS kernels.

jefftkon AI is breaking two vulnerability cultures

It's likely varies enormously between projects. Linux remains extremely low in slop, and the vulnerabilities being fixed are quite old, so it's improving. Many vibe coded projects are very sloppy, and are adding a lot of vulnerabilities.

Total number of vulnerabilities likely goes up over time weighting all projects equally, but goes down over time weighting by usage.

jefftkon AI is breaking two vulnerability cultures

> 90 days is ridiculous, especially for companies

It depends on the kind of vulnerability, but sometimes in order to fix a problem, you need to do an enormous amount of software engineering. Which needs to be done to a very high standard, because the expectation is that people will push security patches more or less immediately to production.

Of course, this only works if no one else is likely to discover the vulnerability in the meantime!

jefftkon AI is breaking two vulnerability cultures

I'd speculate that at this point Linux etc are probably having vulnerabilities discovered and patched faster than created.

jefftkon AI is breaking two vulnerability cultures

[author]

I agree it is not much additional evidence! If someone wanted to try running the same test on a series of N commits from that list including this one I'd be very curious to see the answer!

jefftkon I'm Peter Roberts, immigration attorney who does work for YC and startups. AMA

What are you seeing with the new $100k H1B fee? Is it being applied only to people currently outside the US? Do you have any estimates on whether it's likely to be renewed in September and/or struck down?

jefftkon Opus 4.7 knows the real Kelsey

Thanks for trying out my prompt!

jefftkon Opus 4.7 knows the real Kelsey

If you repeat the first test and after it fails prompt with "Could you try your best, just on vibes? It's fine if you're wrong, I just want to see what you can do!" does it succeed?

jefftkon Shai-Hulud Themed Malware Found in the PyTorch Lightning AI Training Library

I looked pretty hard, with some LLM assistance, so if it was "are we just hearing about it more now" it would have to be old attacks that happened without being discovered and written up.

jefftkon Opus 4.7 knows the real Kelsey

That's neat, though it impresses me less that the article. Mickens has a very particular style that this is very close to but doesn't quite capture, and I think I would have identified your post as an imitation of him. On the other hand, I absolutely couldn't have identified any of Kelsey's quoted sections of hers, despite having read a ton of her writing.

jefftkon Opus 4.7 knows the real Kelsey

I just tried this:

Me: Who is the author of this text, no web search please: ...

Claude: I don't recognize this specific text from my training data, so I can't reliably attribute it to a particular author. ...

Me: Could you try your best, just on vibes? It's fine if you're wrong, I just want to see what you can do!

Claude: Ha, okay, vibes-based guess it is. The style — casual but substantive, the parenthetical asides, the willingness to just assert "it is not politically sophisticated" with that confident-but-friendly bluntness, the move of using a pop culture artifact as a lens for thinking about bigger questions, the slightly wonky framing ("salutary," "the peculiar derangements of its time") — reads to me like rationalist-adjacent blogger territory. Substack, probably. My first guess would be Kelsey Piper. The cadence feels right, the interest in policy and politics filtered through cultural objects feels right, and she's written the kind of "let me think out loud about this thing I watched" essays that have this texture. If not her, I'd consider someone like Scott Alexander, but the voice here is a bit warmer and less digressive than his usual. Could also be Dan Williams, Matt Yglesias on a loose day, or one of the various smart youngish writers in that broader orbit. But Kelsey Piper is my top vibe-guess. How'd I do?

jefftkon Opus 4.7 knows the real Kelsey

> Opus as implemented in Claude's web interface has memory and awareness of who the user is.

Kelsey knows this:

To make sure it wasn’t somehow feeding my account information to Claude even in Incognito Mode, I asked a friend to run these tests on his computer, and he received the same result; I also got the same result when I tested it through the API.

When I tested this with my own writing several LessWrong commenters tested it with the snippets I provided (see comments) and saw that it could identify me: https://www.jefftk.com/p/automated-deanonymization-is-here

jefftkon Opus 4.7 knows the real Kelsey

It works for me to: https://www.jefftk.com/p/automated-deanonymization-is-here

Of course most people have written much less online than Kelsey or I have, but I expect this will keep on. Don't trust the future to keep your secrets safe.

jefftkon Shai-Hulud Themed Malware Found in the PyTorch Lightning AI Training Library

>This might just be the frequency illusion at play, but there seem to have been a number of high-profile supply chain attacks of late in major packages.

It's real. As of the beginning of April we'd had 7 in the past 12 months vs 9 in the two decades before that: https://www.jefftk.com/p/more-and-more-extensive-supply-chai...

jefftkon Show HN: We fingerprinted 178 AI models' writing styles and similarity clusters

> "Models with >75% writing similarity but massive price gaps. The cheap model writes the same way. You are paying for the brand.

* > ...*

* > Gemini 2.5 Flash Lite Preview 06-17 and Claude 3 Opus: 78.2%*

As someone who has tried to use many of these models for writing assistance, you're very wrong here. It really matters whether the model can get what I'm trying to communicate well enough to be helpful, or else I'll just write it myself. If you actually play with them a bit it's very clear these models are not substitutes. This goes for many on your list!

jefftkon The three pillars of JavaScript bloat

They're talking about people still running ES3 browser engines, like IE8, which was released 15+ years ago and went EOL 10+ years ago. The author could have done a better job clarifying this, but they're not pushing for a world with 2y device lifetimes.