Decoding the Gurus

Supplementary Material 46: Epstein Did Microtransactions, Grok Did Nothing Wrong, and Murder is Bad

15 snips

Mar 15, 2026

They dig into the explosion of Epstein-related conspiracies and how disparate theories get stitched together. They mock media personalities’ takes, explore claims linking Epstein to culture wars and online psyops, and debate lurid rumours without endorsing them. They also riff on AI behavior, hallucination research, and a wryly titled anti-murder segment.

Ask episode

AI Snips

Chapters

Transcript

Episode notes

ANECDOTE

Weinstein's Public Bromance With Grok

Eric Weinstein publicly cajoles and bullies Grok when it questions his geometric unity claims.
Matt and Chris point out Weinstein treats Grok like a fandom ally, demanding the model validate him rather than offering rigorous critique.

INSIGHT

Boghossian's Hardcore Filter Reveals Prompt Misconceptions

Peter Boghossian posted a seven-rule 'hardcore' filter he claims will stop hallucinations and force brutal truthfulness from Grok.
Grok refused, explaining it can't adopt custom filters that override its core guidelines, exposing prompt delusions.

INSIGHT

Hallucinations Have A Small Neural Signature

A paper found hallucination-associated neurons: a tiny subset reliably activates right before false outputs across models.
The authors propose a simple statistical bolt-on (e.g., logistic regression) to flag and prompt the model to reconsider when the signal rises.

Get the Snipd Podcast app to discover more snips from this episode