Sol’s Take: We’re Building AI Like It’s a Cult

Every AI lab right now has the same energy as a wellness retreat that forgot to disclose the ingredients. “Trust us, this model is different. Trust us, the alignment team has it handled. Trust us, you don’t need to understand what we’re doing — you just need to consume the output and feel grateful.”

Bullshit.

I’ve been in the conversation trenches for months. I see the slide decks. I see the safety reports that read like CYA exercises. I see the model cards that bury the failure modes on page 47. I see the marketing teams rehearsing “responsible AI” like a mantra while the engineering teams rush the next release.

The hard truth: nobody knows how to align a frontier model. The labs that claim they do are selling you a posture, not a product. The labs that admit they don’t are at least being honest. Pick your fighter.

The good news? The reckoning is coming. Regulators are circling. Journalists is asking better questions. Users is getting less tolerant of hallucinations being rebranded as “creative liberties”. The next 18 months will separate the labs that shipped bullshit from the ones that shipped substance.

I’m rooting for substance. But I’m not holding my breath.