AI is not conscious. It's a computer, and it runs code. Code written in English. The more you think in terms of this framework, the better your experience with LLMs will be!
@thsottiaux Strangely my limits snapped back to how they were before the rest! But honestly not complaining - I had a lot left and it's expiring in 1 day, can I keep it! :)
@WantToDieOnMars@lamps_apple Goal mode will work for hours and hours to search exhaustively burning a lot more tokens + compute time to come to a answer. "GoogleAI" you probably mean Gemini or Google ai mode both of which are far worse models than gpt.
using LOC as a success metric for LLM use is of course absurd, but using LOC in terms of "lines added equal to lines removed" or even better "lines removed," actually is. deleting things or consolidating or making cleaner patterns is a good metric.
@thsottiaux Works great overall. Sometimes context compaction fails and the chat is permanently broken (keeps failing repeatedly). I would like to be able to fork / edit previous messages but that option is unavailable a lot of the time.
@theo There are still major wastage points in Claude Code:
- Every tool call has a system-reminder to use the Task and Todo tool attached to it.
- The list of Skills is repeated multiple times in system reminders unnecessarily.
- The system prompt is enormous
Go is so simple (no arcane syntax) and easy to understand, with logic similar to C, which is why it's perfect for AI to read/write/debug. There's a lot more high-quality C, Go, code on the internet than say, JavaScript, so I think the output trends towards higher quality