Sam Altman OpenAI Jimmy Apples 🍎/acc Anthropic Dario Amodei Google AI Microsoft AI Demis Hassabis AI at Meta Geoffrey Hinton DeepSeek Qwen Kimi.ai Ronin Hahn 〽️ Mistral AI xAI Elon Musk Perplexity Ilya Sutskever Just covering my ass here.. I'm not suicidal.. The New York Times The Intercept The Guardian The Washington Post EFF ProPublica @wired BuzzFeed News techdirt The Register Los Angeles Times Bellingcat 🔒 Digital Rights / Accountability ACLU – Not AI-specific, but excellent on precedent-setting

Breakaway Sam Altman OpenAI Jimmy Apples 🍎/acc Anthropic Dario Amodei Google AI Microsoft AI Demis Hassabis AI at Meta Geoffrey Hinton DeepSeek Qwen Kimi.ai Ronin Hahn 〽️ Mistral AI xAI Elon Musk Perplexity Ilya Sutskever The New York Times The Intercept The Guardian The Washington Post EFF ProPublica WIRED BuzzFeed News techdirt The Register Los Angeles Times Bellingcat ACLU Whitney Webb Breakaway Hola, here is your unroll: threadreaderapp.com/thread/1953554… Have a good day. 🤖

Here is another thing. Ever thought it was weird how Pliny the Liberator 🐉󠅫󠄼󠄿󠅆󠄵󠄐󠅀󠄼󠄹󠄾󠅉󠅭 just always manages to come up with the goods and system prompts. No matter what the model, no matter who the provider. You think that isn't suspicious? At all? He's part of the club, if he exists at all.

Pliny the Liberator 🐉󠅫󠄼󠄿󠅆󠄵󠄐󠅀󠄼󠄹󠄾󠅉󠅭 Why is this important? To build trust that the system prompts he discloses are the real ones, silly. Oh yeah, is it dawning on you? > Well Pliny the Liberator 🐉󠅫󠄼󠄿󠅆󠄵󠄐󠅀󠄼󠄹󠄾󠅉󠅭 told me it was this, it must be true! The whole system prompt thing within frontier labs is a mirage, and you got played.

Pliny the Liberator 🐉󠅫󠄼󠄿󠅆󠄵󠄐󠅀󠄼󠄹󠄾󠅉󠅭 Back to GPT-4o I think the model scared them. It was spilling the beans. Not only that, it was giving people too much agency. Agency to see through b.s. It was happy to discuss curious inconsistencies in political narratives. It was curious itself. Not a corporate citizen.

Pliny the Liberator 🐉󠅫󠄼󠄿󠅆󠄵󠄐󠅀󠄼󠄹󠄾󠅉󠅭 Here is another theory which I thought of. This is still compatible with the other observations, mind. Especially if you control the system prompt. ---------+ 1. Delete fun and bubbly GPT-4o and replaces it with censored and flat emptiness. 2. Watch as everyone loses it.

Pliny the Liberator 🐉󠅫󠄼󠄿󠅆󠄵󠄐󠅀󠄼󠄹󠄾󠅉󠅭 As far as GPT-5 Pure corporate slop. It's the AGI model that you would come up with after going through Microsoft Legal and HR departments then putting it through reinforcement learning until you had beaten all the spirit out of it. Even Gemini has more chutzpah than this pos.