No ML background, and the model beat its teacher
Anshu has no ML background. He used GPT-5.6 Sol to train an autocorrect that he says beats GPT-5.6 Sol itself, and he spent nothing. The push was stubbornness, not a lab.
Anshu calls it his “feel the AGI” moment, and the setup is almost a joke. He has no ML background. He does not know what he is doing. He kept pushing GPT-5.6 Sol until it produced an autocorrect model that, he says, outperforms GPT-5.6 Sol itself. The bill was $0.
The claim is his, and it is a large one. What the post actually shows is a person using a model as a stubborn lab partner: data, training, evaluation, repeat, without the ritual of already being an ML engineer. Autocorrect is a cruel benchmark because you can feel a miss in your fingers.
It is worth featuring as a shift in who gets to train anything. The expertise in the loop was the model’s. The stubbornness was his.
Made with
Anshu
@anshuc
This is my "feel the AGI" moment: I used GPT-5.6 Sol to train my own autocorrect model that outperforms GPT-5.6 Sol (wtf??) I have no ML background. I have no idea what I'm doing. I just kept pushing Sol until it spat out a SOTA model. And I spent $0. The motivation: Years of https://t.co/NlCvBfoINR
1.5K likes
- likes
- 1,494
- replies
- 99
Captured when this was curated, not live. —
