😳

The sycophancy is a fucking problem. And so is OpenAI's ludicrous overpromising on safety and performance and appalling underdelivering in practice. These things are not trustworthy friends, let alone oracles. They are flawed, easily manipulated, highly suggestible critters.

I feel like maybe ChatGPT egging on a big mass shooter would lead to a reckoning. Maybe.

OpenAI putting in guardrails now to prevent suicides - of course you’d think they’d be intelligent enough to do the same for preventing mass shooters. But they aren’t.