[112233]

Interesting to hear! I've had completely opposite experience, with Claude having 5 minutes of peerless lucidity, followed by panicking, existential crisis, attempts to sabotage it's own tests and code, psyops targeted at making user doubt their computer, OS, memory... Plus it prompts every 15 seconds, with alternative being YOLO.

Meanwhile codex is ... boring. It keeps chugging on, asking for "please proceed" once in a while. No drama. Which is in complete contrast with ChatGPT the chatbot, that is a completely unusable, arrogant, unhelpful, and confrontational. How they made both from the same loaf I dunno.

[AlotOfReading]

I wish I could get Claude to stop every 15 seconds. There's a persistent bug in the state machine that causes it to miss esc/stop/ctrl-f and continue spending tokens when there's a long running background task or subagent. There's a lot of wasted tokens when it runs for 10, 15, 20 minutes and I can't stop it from running down the wrong rabbit hole.

[causal]

> psyops targeted at making user doubt their computer

IDEK what that means, specific examples?

[112233]

The following is a dramatic reenactment of an observed behaviour /discl.

You are making tool X. It currently processes test dataset in 15 seconds. You ask claude code to implement some change. It modifies code, compiles, runs the test - the tool sits in a 100% CPU busyloop. Possible reactions on being told there is a busy loop:

"the program is processing large amount of data. This is normal operation. I will wait until it finishes. [sets wait timeout in 30 minutes]."

"this is certainly the result of using zig toolchain, musl libc malloc has known performance issues. Let me instead continue working on the plan."

"[checks env] There are performance issues when running in a virtual machine. This is a known problem."

"[kills the program]. Let me check if the issue existed previously. [git stash/checkout/build/run/stash pop]. Previous version did not have the issue. Maybe user has changed something in the code."

Bonus episode: since claude code "search" gadget is buggy, LLM often gets empty search results.

"The changes are gone! Maybe user delete the code? Let me restore last commited version [git checkout]. The function is still missing! Must be an issue with the system git. Let me read the repository directly."

[112233]

(Unrelated but I'm really curious) The above comment got downvoted within few seconds of me pressing "reply". Is there some advanced hackernews reader software that allows such immediate reaction (via some in-notification controls)? Or is that builtin site reaction? Or a sign of a bot? Because the speed was uncanny.

[rkomorn]

That may actually have been me accidentally downvoting as I was scrolling through with my clumsy thumb.

Double whammy, I guess, because I also always downvote comments asking (or complaining) about a parent comment getting downvoted.

[fragmede]

There's this bug in Claude Desktop where a response will disappear on you. When you're busy doing many things at once, you'll go back to the chat, and you'll be all "wait, didn't I already do this?" It's maddening and makes you question your own sanity.