so, AR? how is AI relevant here?

also, since when is CLI outdated?

still, far more effective than "NEVER FUCKING GUESS"

in my experience, "undocumented tricks" break as often as documented features

like when they removed "clear context and execute plan" option after releasing 1M opus because "context window is not a problem anymore"

I so miss that clear context and execute plan mode! Now i have to keep clearing it manually again.

fyi you can re-enable it with `{ "showClearContextOnPlanAccept": true }` in ${CLAUDE_CONFIG_DIR:-$HOME/.claude}/settings.json

also find `"disableAutoMode": "disable"` useful, since I'm typically switching between yolo and plan

Thanks!!

> like when they removed "clear context and execute plan" option after releasing 1M opus because "context window is not a problem anymore"

I then show me banner that I can clean up "xxx.x k" of context by typing clear right after it started on the plan.

why would anthropic do that?

The argument is that the sweet deal we're getting now, which involves Anthropic practically giving away a product while they lose lots of money, might not be sustainable for them.

> wrong 10% of the time

that doesn't sound nearly as bad as you think it does; I don't see how ethics are relevant here either, unless oil is also somehow a scam

regardless, one must be delusional to deny the fact that it's useful tech

"but they're evil" is not an argument

> deny the fact that it's useful tech

Is this denial in the room with us now?

I'd say "plagiarism machine" implies that

you think it should be deterministic?

If LLMs were configured to be deterministic, instead of the current "small amount of randomness to provide the illusion of a thinking machine", they would be much more useful in exchange for being less exciting to laypeople.

In other words, scam.

you know you can set the temperature to zero and get exactly that, right?

now I know what to call it, thanks

so, hashline?

was wondering myself, just tried comparing to petnames crate -- gets you about 2 tokens per word on average

not that anyone should ever care; typos in random-looking ids are very real but already covered by human readable ids

besides, this is for a specific tokenizer

I think their point was that the project is complex, with the implicit assumption that the complexity is to a large degree inherent.

Even if it's mostly accidental, and the code is overengineered slop (which it is), the system being able to decompose a problem and deliver something is impressive in terms of stability: it wasn't sucked into rewriting everything from scratch every time it would run into issues, it didn't have infinite subagent recursion with a one-agent-per-line type workflow, etc.