No one hates msft more than me, but there’s no way you think this one data point is useful to explain anything at all

Isn’t “Anthropic won’t fix it” a little sensational for barely month old issue with little activity(two upvotes, one of which is me), in backlog of 5k+? Agree that it’s a real issue that need fixing however

There's around a dozen issues about this, the oldest one is from a bit under a year ago (though the author marked it closed... prematurely I'd say):

https://github.com/anthropics/claude-code/issues/4172

or this one from December, still open:

https://github.com/anthropics/claude-code/issues/9258

Some more fun facts about this:

- it is silent, not announced anywhere, and calls rm on your data

- setting the days to zero doesn't disable it, it immediately rm's ALL your chats

- even setting it to _something_ big doesn't prevent data loss if you launch Claude Code in some modalities (certain subagents, etc) that don't load this config key but perform deletion anyway (with the 30 day default)

They claim that they can automate development, so their backlog should really be ~0. /hj

  Isn’t “Anthropic won’t fix it” a little sensational 

It's editorializing and against HN guidelines either way.

"fix it" here is reverting their bad change

I just noticed that it’s a config option. Weird that it’s so short though, I can understand why it may be needed for users who spawn hundreds of sessions a day

The link sharing platform is full of people interested in sharing links, that makes sense, similar to how a mall would be full of people interested in shopping

The article is about how people decide to share links, nie whether they share links at all.

What definition of “often” includes a 0.21% rate?

The term "often" wouldn't relate to the overall rate, but to the frequency. According to my back-of-the-envelope math, it happened about once every 7 minutes.

Try to affect human behavior at scale in any way and you pray for an effect size like that

I created a watcher for this problem, to watch my PRs for unfinished scope and have a fresh Claude review

Uses tmux and gh https://github.com/Kyu/claude-pr-watch

I dont uninstall apps that annoy me with notifications, but I do disable them. Most of my notifications these days are news or texts. So be it

Or create "blind" benchmarks.

10 groups of 3 researchers, all have their own benchmarks that they do not share (testing it without the authors knowing is a different problem, maybe they only run the benchmarks when the gen-pop has access to the models).

that's 10 different tests. Aggregate pass rates

“Exam Question: {prompt}” was enough to get me the right answer on whatever model you get with logged-out ChatGPT.

Neither prompt was enough for llama3.3 or gpt-oss-120b

Pydantic + FastAPI are my two favorite python shops right now, they’re always dropping fun new projcts

The PhD can't read minds, the quality if the request from a moron would be worse than the quality of the request from someone with avg intelligence. And the output would probably noticeably differ accordingly