yorwba

It's not clear from the description whether he submitted his valid New Zealand visa. No doubt Anthropic don't make it clear that they aren't asking for ID to verify someone's name, they're asking to verify that they're dealing with the kind of person they're in principle willing to do business with.

It's a custom design for the Globe theater. Of course they aren't going to let you download their unique branding, that would defeat the point.

In this image: https://m.itsnicethat.com/original_images/typeland-shakespea...

Those wouldn't be connecting from Russian IPs, would they? And typically they would also have some kind of document from the country they're staying in (residence permit, refugee registration, whatever) to indicate that they're in one of Anthropic's supported regions.

My read on the post is that it's trying to get clicks by stirring fear in Claude users in China, who are similarly not allowed to use the service and worry that their accounts might get disabled.

That's always how empires are subjugated. An initial invasion force demonstrating their fighting prowess gets some locals to join them, and with that bigger army they can get other big armies to do the same and so on, and if everything works out, they end up with the by far most powerful army around and conquer everything their logistics can reasonably reach.

That's how China got overrun by steppe nomads twice, for example. The unusual thing about the Spanish conquistadores is mostly how little of an initial force they needed to set off an avalanche. Even Genghis Khan had to slowly work to unify the Mongol tribes before he could switch to rapid expansion.

It would make more sense to have the decision model check for each possible reason whether it applies in the current case and then have the resulting consequences determined by a bog-standard if-then-else program.

If you want to suggest a new travel destination for Matt Lakeman, you'll have to use the comment functionality on his site instead of here.

China has no such policy. Article 41 (3) of the Labor Contract Law https://english.court.gov.cn/2015-08/17/c_761484_6.htm authorizes layoffs if "The enterprise changes its line of production, introduces a major technological updating or adjusts its business method, and, after modification of the labor contracts, still needs to reduce its personnel." There's been a prominent case recently http://english.scio.gov.cn/m/chinavoices/2026-04/30/content_... where a company tried to claim a change in the "objective conditions taken as the basis for conclusion of the contract" under Article 40 (3) instead, which the court disagreed with. This has led to a bunch of clickbait reporting in the West, claiming that workers in China are somehow protected from automation, but all it means is that they will be laid off instead of fired.

In 2025, METR published a study that found AI slowing down developers by 19%... https://metr.org/blog/2025-07-10-early-2025-ai-experienced-o...

Not infinite, just enough to make more than one model. OpenAI and Anthropic both have four different size categories...

No. In a large enough state space, you can of course easily find a state that has never existed before. For example, if you pick a 1024-bit integer uniformly at random, it will certainly be different from any previous 1024-bit integer. However, it's much more difficult to find a new integer that cannot be assembled out of bits taken from previous integers. Composition from a handful of base atoms can cover very large state spaces, and unbounded composition can cover infinite state spaces. Which is why only being able to combine things that already exist is not much of a limitation.

To demonstrate:

"Are you saying that a" https://news.ycombinator.com/item?id=49642931

"complex stateful system" https://news.ycombinator.com/item?id=14881439

"cannot generate a" https://news.ycombinator.com/item?id=45956162

"state that is not already" https://news.ycombinator.com/item?id=43013637

"recorded in text" https://news.ycombinator.com/item?id=44731577

"somewhere?" https://news.ycombinator.com/item?id=50391679

A novel state, obtainable by combining six existing ones.

All novel insights expressible as text can be assembled from snippets of existing text. The challenge is "only" that you need to find the right snippets. Hence being a stochastic parrot does not represent an upper bound on capabilities, but a suggestion that if you want your parrot to give you a specific answer, you should arrange for a context where it's statistically likely to be correct.

They don't have to choose, though, since they can release multiple models with different tradeoffs.

Rearranging blocks at the top to make them taller shifts up the center of gravity for the entire structure, which can destabilize lower levels. So copying part of a more stable tower into a less stable one won't necessarily be an improvement.

AFAIK the currently biggest US open-weight model is Inkling, which they include in the comparison, so at least their claim is not limited to a size category. Still they score 0.42 percentage points higher than the second-best model on their chosen benchmark, so it's hardly a meaningful lead.

He's one of six named people in the article. They're providing different perspectives, not trying to find the one authoritative source who has all the answers.

According to your ref, it entered the front page at a "Top" rank of 8 at 8:10:00Z, when it had 2 upvotes. Just as I told you. (Did you confuse the "New" rank for the front page?)

Maybe the midterms caused you to refresh the front page more often so that you noticed this post before it had been there long enough to get upvoted a lot. (The minimum to get on the front page is 2 upvotes.)

Throughout the paper, there are expandable sections labeled "Show the full prompt". If you read those, you'll see that they include a description of the inputs and outputs and the model architecture.

If you trust the token generator and just worry about tampering in transit, signing the tokens after the fact is a fine solution.

This work is about the case where you trust nobody and therefore require even the token generator to do extra bookkeeping to prove that it's not taking any shortcuts.

The cookie banner says they store "a single in-house cookie." That's the technically unnecessary data.

Reading an article takes more time than commenting, which takes more time than voting. So the typical life cycle of a HN post is that people vote on it without reading the article, comment without reading the article, vote on the comments without reading the article, and respond to comments without reading the article, until after some time people who've actually read the article show up in the comments with some clarifications, saving everyone else the bother of having to read it themselves.

If you want to escape the hype, reading posts with no upvotes and no comments on https://news.ycombinator.com/newest is an option, though somewhat hit-or-miss.

"action 3 left" is documented in the input, so you can find the corresponding weights in column 3 of the output weight matrix. If that's past what you expect from humans or conventional algorithms, your expectations are too low.

I'm curious how you created AI images using Claude despite Claude not supporting this: https://support.claude.com/en/articles/9002504-can-claude-pr...

It's just irrelevant for big-O complexity what happens for instances below a constant threshold, so the algorithm uses the simplest approach and moves on. If you have an exponential-time algorithm that's faster than brute force for small instances, you can use that and get a faster overall runtime at the same big-O complexity. If you can lower the switchover threshold, ditto. It's a bit weird to dismiss the subexponential algorithm just because it's leaving an obvious optimization on the table... it can only get faster from here.

Maybe the SEC should instead block funds from distorting the market by making large predictable transactions, then.

In cases where Astra succeeded, it succeeded using standard techniques also used by humans when they read and write neural network weights. It's a rare skill, mostly used for fun (e.g. https://blog.janestreet.com/can-you-reverse-engineer-our-neu... ) and basic research rather than profit, but Astra is big enough to have learned many rare skills.

At least this 1984 report on "Aspects of Soviet Secondary Education: School Performance and Teacher Accountability" https://www.wilsoncenter.org/sites/default/files/media/docum... claims that a more ambitious curriculum did not lead to more repeaters ("grade repeating has declined over the years"), but instead "most teachers have chosen the easier solution of adapting their grading practices to the results expected of them." The students who didn't get algebra were apparently filtered out later: "According to the few statistics available on entrance exam results, it seems that somewhere between 20 and 30% of all students fail tests in their chosen field of specialization."

About $22B of it, yes.

People pay money for the solar panels, though. It's not obvious to me that giving away expensive LLMs for free is meaningfully helping grow the domestic economy, as opposed to foreign inference providers who get their GPUs from Nvidia.