AI models don't just hallucinate. They also "scheme," meaning deliberately lie or hide their true intensions. https://t.co/fD6JfBPIjs ...

RT sankalp (@dejavucoder)fixing three things that i think will make learning with LLMs a much better and easier to trust experience -1. hallucination in long context scenarios2. reducing the agreeableness so LLM can call out my bullshit and other model's bullshit. correlated with hallucination3. better intent detection where the LLM asks follow up questions if the intent was not clear or if it just wants to underrated the preference of the user ...

The counter: https://t.co/3G4wiJUaFdYu Hu 🌊 (@Punk9277):@S4mmyEth yes classic hallucination there - nobody is on 80+ leaderboardsmax is just above 10, with overwhelming majority on less than 3can easily verify ...