https://teapaw.org/t.co/qJocUqax9Z developer of #instantos
Joined October 2016
- Tweets1.1K
- Following352
- Followers80
- Likes194
Mimo 2.6 flash has not convinced me so far. It reads the entire codebase every time and thus compacts every 2 minutes. Maybe it was RLed with 1M context on reading everything first and then exclusively writing, but at 200k that's not usable.
paperbenni retweeted
Do you walk back the phrase “maniacs and clowns”, David?
I feel like people renewed their Go subs based on this, and this was just flat out wrong??
teapaw.org/thdxr/status/208537784…
freebuff.com/get-started?ref…
I am once again asking for you to give me free stuff in exchange for free stuff
local.ai/paperbenni/invite
not sure what this is, but gimme access I guess
zenmux.ai/invite/4XOSNZ
give me free credits pls thanks
the site seems very vibe coded though, not sure if it's good for anything else
Aaand back to Qwen. Why is chatGPT still incapable of rendering latex without cutting off part of it?
paperbenni.xyz/posts/aiarena…
wrote this a while ago. I'm afraid it's still very relevant
gemini pro seems lobotomized now, it cannot help talking about the system prompt and proudly announcing that it "will produce markdown now"
fun fact, the Gemini CLI does not tell you when it errs out, it just keeps going in circles. Also "resource exhausted" when I have used 9% of my quota?
There is something deeply deeply wrong with @cline
They manage to make Kimi 2.5 behave like models from years ago, you have to prod it for every single tool call, with a success rate of 10% per tool call. This is their own gateway, their own harness, their own webUI.
is that a problem with antigravity or does Gemini 3 love long insanity comments in its code like Claude loves markdown files?
Gemini 3 just ignored me telling it 30 times not to create git commits. Last time any model did this was 4o
