@taiomi A shared daily target could make that counter even more fun: 'we're 200 words from today's goal.' A reason to play one more round without turning it into a leaderboard.
@sardorml I'd check how many installs actually open the extension before blaming signup. 30 installs → 4 opens is a different problem from 30 opens → 3 signups.
@SoveyX@IfindRetards That 70% cut in the notebook is the part a generation count misses. Better models could save you retries without saving a single weak joke. Appreciate the detailed answer.
@graphixjuls@dwc_hq@chikeoivor The burgundy makes the sofa and floor lamp feel like a set, while the green shade gives the little lamp its own personality. Are these going into one room scene, or a reusable illustration kit?
@thecyberseckyle The file changing after a scan is the strongest example here. A confirmation button shouldn't be a permanent permission slip for whatever happens to occupy that path later.
@sendme_ai The part I'd want to see: tests the agent isn't allowed to rewrite. Otherwise a self-improving loop can become a self-approving loop. Does the course keep an independent, held-out evaluation set?
@Polymarket I'd love a comparison that counts human corrections, not just correct answers. Same messy task, same budget, no cherry-picking. The winner is the model you have to rescue least often.
@collision@Revolut The 'lightweight users to primary accounts' chapter is the one I'm curious about. Signups measure curiosity; moving a salary or recurring bills measures trust. Which early behavior best predicts that switch?