yruzin

I think this is where harness makes a lot of sense. Use LLM to produce all possible attack angles/phrases and just stupidly filter them out on input.

Claude is Extremely slow, especially fast few days. Codex is so much faster in in my opinion with comparable quality. Anthropic is going to fix their problems, but for me as a user that depends for the service to work, it's not acceptable. ChatGPT 5.5 is already a better version than Opus 4.8, at lest in my experience. Firstly, thing get done today and I don't see quality to be worse. I has many situation where Opus was just going in circles, finding more and more issues with the code it wrote a few hours back, to me this is counterproductive.

I pretty much giving up on Claude. It's Extremely slow, especially fast few days. Computer use takes so long to do anything that sessions expire. I switched to Codex and it's a different story. I'm not saying that Anthropic is not going to fix their problems, but for me as a user that depends for the service to work, it's not acceptable. So, we switch back and forth between a few LLM's but Claude and Codex are the leaders and I can't afford using anything else if I know I can get better quality results. The cost is important but not as much as the quality of the solution. You might save a few bucks on a cheeper model but you will have to prompt it many times and time is more important to me, as well as the feeling that I'm getting the best result on the first try.

How did you measure that the agent used >90% of time re-reading the context? The idea is that the agent has the KV cache, that helps with inference, but I'm not sure if what you are saying is related to re-doing the inference or there some other internal mechanisms that re-reads the KV cache on every agentic loop start?

Sorry, my bad :)

Spoiler for the quiz below.

The defer delays the function call, but it does not delay evaluation of the call’s arguments. This means that both timestamps are therefore captured near the beginning of the request, even though the second line is printed when the handler returns. Also, time.Since can use the monotonic component stored in time.Time, so the duration is not distorted by wall-clock corrections.

start := time.Now() defer func() { log.Printf("request completed in %s", time.Since(start)) }()

The answer is A: the two timestamps are approximately the same.

Thanks for posting this! This is fun :)

Durability is not the problem, but because the state lives somewhere you structurally can't relocate. That constraint never shows up in a benchmark, it shows up in an enterprise procurement call.