Sinidir

Wow. I'm literally reconsidering having any mails at proton anymore. This is beyond abysmal attitude.

Wow. Luna 6 is an insane value bargain at this point. If Anthropic doesn't finally come out with their own small low cost model instead of still having haiku 4.5 they'll be history soon.

Yeah i noticed this heavily. Working on a branch i critique and give a simpler design. Claude implements and then in a huge comments reference the outdated worse design. Zero value to that. Even having a "how to write comments" section in claude.md doesn't seem to help much with that.

Yes these things happen as part of RL Training. Same way that you can see the "But wait ..." phrases in thinking traces. They get rewarded.

Article is seriously wrong, because it makes a huge mistake in the last part. You can't simply look at the produced tokens and that is your cost. In agentic coding there are lots of turns meaning you not only pay for the output tokens you also pay for all the input tokens sent each time (even if a lot cheaper, like 10x when cached). So this calculation does not accurately represent the api cost at all.

Second thing is you can starkly upgrade the token generation locally if you use agent teams. Single conversations are memory bandwidth bound and don't fully make use of your compute. If you can batch tokens from multiple agents you can easily 5x token generation.

Harness: a piece of equipment with straps and belts, used to control or hold in place a person, animal, or object.

So yes the generel meaning applies to test setup and running and also to the agent cli which is the harness for the model.

The K/V Cache is just an optimization. But yeah you would expect the attention for the model producing "Ok im doing X" and you asking "Why did you do X?" be similar. So i don't see a reason why introspection would be impossible. In fact trying to adapt a test skill where the agent would write a new test instead of adapting a new one i asked it why and it gave the reasoning it used. We then adapted the skill to specifically reject that reasoning and then it worked and the agent adapted the existing test instead.

Does anyone know. How would that relate to simply wrapping claude code as a subprocess?

In the example given in the article i think the correct behavior would have been to infer the type backwards from the return type of the function. Is that not why mypy actually errors here?

Its not really hard to tell.

Pipe to dev/null. Fastest database i have ever used.

It is literally true. You don't need to run a type checker.

mypy is also written in a style conducive to speed ups when compiling with mypyc

Funny i know about Proust from The Sopranos.