AGI is here (and I feel fine)

Robin Sloan -- novelist, programmer, and newsletter writer -- declares that AGI has arrived, arguing the 2020 paper "Language Models are Few-Shot Learners" (GPT-3) marked the threshold. Published as the first edition of Winter Garden, a six-edition pop-up newsletter that will self-destruct.

Summary

Sloan makes a "strategic but/and also sincere" case: AGI is here, the critical word is "General," and the open-ended capability of language models was "a genuine surprise, even to their custodians." He takes the extreme position that GPT-3 outperforming task-specific models on linguistic benchmarks was the AGI moment -- the paper should have been titled "This Thing Can Do It All?!"

He probes the reluctance to declare AGI: critics don't want to "concede defeat," industry participants are "nervous to claim the achievement because it feels like more should have changed," and there's a strategic incentive to keep the goal receding ("more money, more electricity, utopia just around the corner"). His unilateral declaration is a "strategic countermove" -- he congratulates the industry and then asks: "And...? What now?"

He acknowledges severe limitations -- the physical realm is closed, many symbolic tasks confound models, and Chollet's novel puzzles remain hard -- but invokes "Robin's Razor": if you find yourself parsing different flavors of generality, it's AGI.

The essay's second half argues big models are "more like Twitter than they are like jet engines" -- communication systems that are "open to meaningful interpretation & evaluation by anybody/everybody." Companies have "no monopoly on understanding them." The final section draws a parallel to the personal computer revolution: those visions were substantially realized (everyone owns a computer, everyone is online), yet "we appear not to live in utopia." The "and...?" of those inventions "roars in our ears." Now we face the same question with AGI.