AGI Is Here (Robin Sloan)
Robin Sloan -- novelist and programmer -- declares AGI has arrived, pins the threshold on GPT-3's 2020 "few-shot learner" paper, and argues the reluctance to acknowledge it is psychological and strategic. The real question isn't "is it AGI?" but "what now?" -- a question the personal computer revolution also left unanswered.
Key Quotes
"Generality has been achieved."
The thesis, stated plainly. Sloan isn't hedging. His argument rests on the word "General" -- not "human-level," not "conscious," not "economically valuable." Just: can it do a vast range of things? Yes.
The 2020 paper "Language Models are Few-Shot Learners" -- which introduced GPT-3 -- "marks the AGI moment." A more direct title: "This Thing Can Do It All?!"
This is Sloan's most provocative move. Not GPT-4, not Claude, not reasoning models -- GPT-3 in 2020. He loves the idea "that the epochal threshold was passed years ago... without noticing the sign." It's a shrewd reframe: if you're waiting for a dramatic threshold, you already missed it.
"Robin's Razor: if you find yourself parsing different flavors of generality, it's AGI."
A useful heuristic. Once you're distinguishing "this kind of generality" from "that kind of generality," you've already conceded the premise. The parsing is the tell.
Big models are "more like Twitter than they are like jet engines" -- communication systems, not engineered artifacts with a single correct interpretation.
This is Sloan's strongest structural argument. Jet engines have designers who understand them definitively. Twitter had a company that repeatedly failed "to make the platform sensible to new users." LLMs are in the second category. The implication: "a regular user can have an experience with these systems, an idea about them, that is not available to even the deepest, dankest, insider-iest insider." Companies have no monopoly on interpretation.
"Everybody really can call upon AGI, or something like it: a wildly general computer program." The personal computer parallel: those visions were realized, "yet we appear not to live in utopia." The "and...? of those inventions roars in our ears."
The essay's closing move. Sloan isn't making a prediction -- he's noting that we've been here before. Transformative technology arrives, achieves its stated goals, and doesn't produce utopia. The "what now?" question is the same one.
Key Themes
The generality threshold was crossed quietly. Sloan's strongest claim: AGI wasn't a dramatic breakthrough but a gradual recognition that a 2020 paper had already demonstrated the thing. The retrospective "we already did it" framing is more rhetorically powerful than "we're almost there."
Strategic anti-millenarianism. Sloan identifies the "ever-receding goal" as a feature of industry rhetoric, not a bug. Declaring AGI is a "strategic countermove" that forces the question: if this is it, what are you actually building toward? It tears away "the veil of any-minute-now millenarianism to reveal deployed technology."
Democratized interpretation. The Twitter analogy is the essay's most original contribution. If LLMs are communication systems rather than engineered products, then everyone gets to have an opinion, and "insider" understanding is deep but narrow -- possibly blinkered. This is a defense of the essayist's right to opine, but it's also a genuine structural claim about the technology.
The personal computer parallel. Sloan reads 1970s-80s PC history as a rhyme with the current moment. Same culture, same geography, same mix of utopian promise and genuine achievement. The parallel doesn't predict outcomes but frames the question: transformative technology arrives, delivers on its promises, and the world doesn't become paradise. What now?
Critical Analysis
What's valuable. Sloan's "just declare it and move on" stance is a useful corrective to the endless definitional quicksand that swallows most AGI discourse. His argument that the word "General" is the one that matters -- not "human-level" or "conscious" -- is clean and hard to refute without entering the parsing game his Razor preempts. The Twitter analogy is genuinely insightful: it explains why company pronouncements about their own models aren't authoritative in the way Boeing's pronouncements about jet engines are.
What's oversimplified. Sloan underweights the Chollet objection he acknowledges. If AGI means "general in the way a human is general" -- able to tackle genuinely novel problems from first principles -- then current models demonstrably aren't that. Sloan's response ("the big models have become REALLY VERY WILDLY GENERAL" since Chollet's 2019 paper) is enthusiasm dressed as argument. He's defining away the objection rather than answering it.
What's missing. No engagement with the economic dimension. If AGI is here and it's as general as Sloan claims, why aren't the productivity numbers showing it? The essay's personal-computer parallel points toward an answer (utopia doesn't follow automatically) but doesn't develop it. There's also an unexamined tension: Sloan wants both "AGI is here" and "these systems have severe limitations." The essay doesn't reconcile how both can be true without making "AGI" a hollow category.
The essayist's privilege. Sloan is a novelist and programmer who's been working with these technologies since 2016. He's writing in a six-edition pop-up newsletter that will self-destruct. This gives him the freedom to be provocative without consequences -- he can declare AGI and walk away. That's a feature (essayists should provoke) and a limitation (the declaration doesn't have to survive contact with implementation). Contrast with Hassan who has to build trustworthy systems on top of these models, or Flake who has to reason rigorously about their mathematical properties.
Where it fits. Sloan's essay pairs usefully with A Non-Anthropomorphized View of LLMs -- Flake would call Sloan's "generality" framing more measured than most AGI discourse but still too loose. Why We Fear AI provides the missing economic dimension: if this is AGI and we're not in utopia, that's because capitalism, not technology, determines distribution. The Future of Everything is Lies I Guess catalogs the harms that make "I feel fine" seem cavalier. Creative Firewall shares Sloan's instinct that interpretation belongs to users, not companies, but grounds it in creative practice rather than declarative argument. Zheng Dong Wang's 2025 Letter offers a compute-thesis perspective that's more cautious about declaring thresholds crossed.
See Also
- A Non-Anthropomorphized View of LLMs -- Flake's mathematical rigor as counterpoint to Sloan's essayistic declaration
- Why We Fear AI -- The economic dimension Sloan omits
- The Future of Everything is Lies I Guess -- Why "I feel fine" might be wrong
- Creative Firewall -- User interpretation over company authority, from a creative practice angle
- Zheng Dong Wang's 2025 Letter -- The compute thesis: more sober about thresholds
- The Metamorphosis of Prime Intellect -- One answer to Sloan's "what now?" question, written in 1994: a perfectly general AI that grants every wish, and the resulting world is dystopia. Sloan asks if we'd notice the threshold; Williams shows what's on the other side of it.
- The Mundanity of Excellence -- Excellence as qualitatively different choices, not more effort -- relevant to the "what now?" question
- Talking to Transformers -- Practical engagement with LLMs that doesn't require declaring AGI
Sources: AGI is here (and I feel fine) Last updated: 2026-05-15