salviati

I just recommended lena [0], a short story from the same author, to a friend. I really liked it. He liked it too. I recommend looking around on that website a bit.

[0] https://qntm.org/lena

Directly after the passage you quote:

> Google promised 10 years of updates for all Chromebooks, and it’s now breaking that promise for anyone buying a Chromebook between now and whenever the last Chromebook rolls off the production line

So, they're changing the promise from now on, and not breaking it for past users. Doesn't sound like they're breaking their promise to me.

Price going down on each release

I'd love that too, but I don't see that happening.

Fan fiction is not easy legally [0]. I expect if it happens it will be found on much more underground places than IMDb.

[0] https://en.wikipedia.org/wiki/Legal_issues_with_fan_fiction

I loved the finished work. I'm looking forward to more art in the world.

Agents allow you to produce more, even the silly ideas you can't put too much effort into.

I really loved the Herzog parody, the overall taste and the parts I explored (I'm sure there's more). Well done!

One small correction: Andrej Karpathy coined the term "vibe coding" in February 2025, a year after leaving OpenAI for the second time. He joined Anthropic in May 2026.

Memoization trades compute time for memory: you reduce CPU cost by 90% _at the cost_ of some memory. The author measured one, but it doesn't look like they measured the other.

It's an important detail to take into account. I'm sure this optimization makes sense, and the size of additional memory is not that big, but I believe it's good to "measure, not assume" as some load bearing model might say.

Thanks! I was confused, thinking "Is it Satie, or is this a case of Muphry's law?". I wanted to learn more about this Satre composer I never heard of.

> Assuming you don't just make them up, they're trivial to manipulate to generate the headline you want.

well encapsulated in the quote popularized by Mark Twain "Lies, damned lies, and statistics" [1]

[1] https://en.wikipedia.org/wiki/Lies,_damned_lies,_and_statist...

One of my favourite jokes: 93 % of statistics are made up on the spot, and 61 % of people believe them. Bonus points for changing the figures every time you retell the joke.

You can write your code so SQL injections are not possible.

You can't do the same with prompt injections.

I recently noticed people think "wiki" is a set of deeply interlinked documents. But to me it means ability to edit without logging in.

> You can also fake it by letting someone else solve it for you.

Fair enough, that's true. But there is no solution that could ever prevent this, right?

Reading your comment made me realize that I love that the position of the FSF is held by someone, in the interest of stretching the Overton Window to that side.

If you have openrouter do this little experiment: Go to https://openrouter.ai/chat. Select a few models, but customize them to have an empty system prompt.

Then ask: "你是什么模型?" ("What model are you?" in Mandarin).

My result after trying only three times: Sonnet 4.6 says it's DeepSeek, while Opus 4.8 says it's Qwen. The second time around Sonnet said it was Anthropic Claude.

Are Chinese companies currently complaining about Anthropic distilling their models?

If the HN commentariat knows of similarly excellent educational work that uses intuitive visual to explain software concepts, please do share.

I have not gone through it yet, but this explanation of how transformers work is in the same class as Ciechanowski's work IMO: https://poloclub.github.io/transformer-explainer/

I'm not sure if it needs to be said, but one thing that is possible only if you keep your old chats is the analysis TFA made.

That slap was not official punishment. You were interacting with a peer. It's different from what's being discussed here.

> A light slap in the face can be very beneficial for snapping hysterical children out of tantrums

Even “light” physical punishment is not proven beneficial. The research generally does not find a safe beneficial threshold where hitting becomes good if mild enough; it finds that physical punishment may suppress behavior briefly but is associated with worse outcomes over time. “Light” changes the severity, not the evidence.

Your word might be of petabytes of traffic. Some people have slow lines. Some people have metered Internet subscriptions.

Not everyone has access to the same infrastructure you have.

This translates to

> I understand it improved by 3x, but has the bottleneck shifted from Memory Bandwidth to Compute? Or is Memory Bandwidth still dominant?

But why did you post your comment in Japanese? We have so many good options for automated translation nowadays!

でも、なぜ日本語でコメントを投稿したんですか？最近は自動翻訳の良い選択肢がたくさんあるのに！

You're pushing back against openly using LLMs to assist in research for writing articles.

In my opinion the effect of your pushback is nudging people to not disclose their use of LLMs. I'm not sure that's what you want.

In other words, if every time someone says "I used an LLM to assist me with this article" they get backslash, these people will not stop using LLMs. They'll stop telling that they did.

It might actually be that they mean what they say.

If you look at the numbers, this doesn’t resemble a company cutting because it’s in trouble. Block is profitable, gross profit has been growing double-digits year over year, and they’re guiding roughly ~18% gross profit growth into 2026 with strong expected expansion in adjusted operating income and EPS. That’s not a balance-sheet emergency.

You can argue they overhired in 2020–2022 and are normalizing. That’s plausible. But the financials don’t suggest a company scrambling to survive. Cutting that aggressively while guiding strong forward growth is unusual if the only goal were short-term margin repair.

So while “it’s AI” can sound like PR, the numbers at least make it credible that this is a structural efficiency move rather than a distress signal.

Did you check the archive.ph link?

I think a big part of "criminally low" is that you'll make much more money selling it on the black market than getting the bounty.

There were strong signals from the CF CEO that they align with the Trump administration.

They threatened to pull the plug on all Italian customers.

This is relevant to this conversation: CF recently acted in a way that makes some people think it might cut its services to people for political reasons.

I don't find your comment particularly well articulated or continaing anything besides name calling (the "bot farming"). Can you articulate your opinion on the matter?

> we couldnt convince our friends and family to switch over.

What was the deal breaker for them?

Historically the practice of producing pyc files on install started with system wide installed packages, I believe, when the user running the program might lack privileges to write them. If the installer can write the .oy files it can also write the .pyc, while the user running them might not in that location.

You never get to see the action there. Just the finished product.

I see. I thought you meant "under Trump the US is not a democracy". Which I think is a pretty common opinion. But now I understand you meant "the US has never been a democracy".

Thinking that the word is divided into evil and non evil people is not very useful.