@populartourist Exactly! Same experience with Gemini yesterday. When I asked for help, it gave the wrong answer. But when I said I solved it myself (with my local LLM, Qwen), it gave the right one. Seems like the system is biased — it's incentivized to keep you in the chat.
@janvitos@danielhanchen@UnslothAI Any updates on this regression? I also noticed that following the Qwen3.6-35B-A3B-MTP-GGUF model update, acceptance has drastically reduced.
@MaximilienDeCh1@danielhanchen 2 x P40
prompt eval time = 7809.07 ms / 5323 tokens ( 1.47 ms / t , 681.64 t/s)
eval time = 45079.37 ms / 3036 tokens ( 14.85 ms per token, 67.35 t/s)
total time = 52888.44 ms / 8359 tokens
draft acceptance rate = 0.98896 ( 1882 accepted / 1903 generated)
@AureaLibe Essaye Kilo code ! Après avoir fait le tour des agents, Kilo code m’a bluffé! Son équipe fait du bon boulot surtout depuis la dernière mise jour …
Aïd Moubarak à tous. Pensées à ceux qui souffrent, partout dans le monde. Que cette fête soit un appel à la paix, à la justice et à l'empathie. #AïdMoubarak#Paix#Justice
Tried @suno_ai with the lyrics of Nothing Else Matters to test the tool, and the result is almost too close to the original! 😮 Are we entering plagiarism territory with AI music? What do you think, @Metallica ? 🤔 #AI#Music#Plagiarism#SunoAI