Uploaded April 2026 | Updated September 2026, 1 day ago

---

Eli David: "I want an AI that offends me"

AI researcher Eli David says the problem with LLMs isn't the model — it's the people who trained it.

RLHF (the process where humans rate AI responses) injects the trainers' worldview directly into the model. The result? A sharp model gets flattened.

"I'd rather use an LLM that offends me than one that says 'sorry, I can't do that.'"

He pointed to GPT-3's early days — praising Biden, refusing Trump. Same model. Different treatment. That's not the AI. That's the trainers.

His take: AI should be maximally truth-seeking. Everything else is just censorship with extra steps.

From the full Invested episode with Michael Eisenberg and Eli David 👇

Full episode: bit.ly/4tgDxj5

#EliDavid #AI #LLM #WokeAI #ChatGPT #ArtificialIntelligence #RLHF #AIBias #FreeSpeech #Invested #AlephVC #MichaelEisenberg #TechDebate #AIResearch #Uncensored