@thdxr Agents are multi-turn. If a prompt is 99% accurate, after 20 steps that’s 99%^20. Therefore, multi-turn agents that don’t lose track are a harder challenge for models. Claw relies a lot on multi-turn agents.
@AceMcWicked I’ve seen it from a health angle too - the social benefit of our favourite social lubricant (in moderation) offsets the health costs of the drinking.
It’s not to say we couldn’t find other ways to facilitate socialisation; but first we need to get off our phones.
@GergelyOrosz I’m reminded of someone asking “but what if I say ‘make it maintainable’?” trying to do the right thing, but not understanding why that won’t work.
@asmartbear Por que no los dos? The former is better for judging the health of your acquisition systems, the latter is better for judging the health of the company, benchmarking against peers, and reporting to potential investors.
@ChShersh@neetcode1 Is this the best or only way to determine whether an engineer is worth hiring? The problem’s not leetcode, is using leetcode in recruitment.
@mtxpost@leothrix Go find the post by the Bun creator asking why LLM engineers can’t make them 1000x faster by applying lessons from video games. I can’t tell if it’s rage bait or epic hubris.