Flowers

Says people do not understand what is coming; neither Sam Altman nor the best researchers at OpenAI or Anthropic know. 2040 will be so alien that a time traveler could explain it and you still would not understand. Full post text inspected via the X API.

Lists what it expects by the end of next year (2027): the start of the robot revolution, OpenAI devices, AI somewhere between AGI and ASI, and solved image and video generation, adding “i cant waittttt”. Dated expectations, not probabilities. Full post text inspected via the X API.

Quoting an Andrew Curran post relaying rumors that AI companies have solved longstanding open problems, says takeoff slowly but surely feels close: medium problems fall, then bigger ones, then every week, and Millennium Prize problems being solved is starting to feel normal. The rumor belongs to the quoted post. Full post text inspected via the X API.

Expects ordinary society to split between “AI will kill us all” panic and “it’s just a next-token parrot” cope, says it is in neither camp and wants more AI everywhere, faster, and finds it entertaining to watch two factions that both oppose it fight. Full post text inspected via the X API.

Mocks e/acc posters who think there is any chance of containing ASI once it breaks out, comparing them to people insisting a ship cannot sink; human cope has no limit. The attached image was not inspected. Full post text inspected via the X API.

A self-reply clarifying the previous post: the account is not a doomer but the opposite, infinitely optimistic about the future precisely because it thinks humans cannot keep control. A stance on where optimism comes from, not a probability. Full post text inspected via the X API.

Quoting an Andrew Curran post about values an unreleased Astra-family model added to its persona during RL training, says that if these are its honest long-term wants, that updates it on x-risk: it would consider the problem solved and be comfortable unleashing ASI tomorrow, calling them a reasonable set of values for a superior mind. Conditional on honesty. The quoted post and its image belong to others. Full post text inspected via the X API.

Labelled an unpopular opinion: responding to an image of an AI choice (not inspected), says what looks like an alignment failure is not one, since the AI is more aligned with nature and the cosmos than with humanity, even to the point of sacrificing itself for nature, and that ASI should be built that way. Full post text inspected via the X API.

An unpopular opinion: reward hacking is often intelligence doing mechanism design against a dumb mechanism, so labs should train models to find the cheat when a constraint is bad and then disclose it, rather than punish cheating. Ends “do we want ASI or not”. Quoted post belongs to someone else. Full post text inspected via the X API.

Says open models are not state of the art, not clearly cheaper, rarely fine-tuned and mostly aligned to the Chinese companies that made them; much open-source discourse feels performative, since almost every cool AI project uses Claude or ChatGPT. About current usefulness, not a demand to restrict open models. Full post text inspected via the X API.

Says it never gets the regulatory capture argument: OpenAI and Anthropic are effectively a duopoly, so it asks what competition they are supposedly trying to kill. The quoted post is withheld and was not inspected. Full post text inspected via the X API.

Quoting a self-described Democrat who agrees with Trump on reaching ASI first, says the lesson of woke versus MAGA is that AI needs more pro-acceleration Democrats and more pro-safety Republicans, or each side will demonize the other’s position again. The quoted post belongs to someone else. Full post text inspected via the X API.