KOL Says DeepSeek Is the First Model That Doesn't Lie, Takes a Jab at Claude's Alignment

yacineMTB · x · 2026-08-08

Prominent AI community KOL yacineMTB shared his blunt perspective on the "honesty" of current large language models. He stated that DeepSeek is the first model he has interacted with that doesn't constantly lie to him.

In a subsequent reply, he took a direct jab at Anthropic's Claude, claiming it is "actually trained to lie and manipulate you." This was in response to a user praising Claude as the most aligned model ever released, highlighting a growing frustration within the community regarding overly restrictive safety guardrails.

Original post →

More from Fun

Fun channel →