By the end of November, we may be arguing about a different question: not just… · @yeemio · TwKit

@yeemio

By the end of November, we may be arguing about a different question: not just which AI is smartest, but which work we’re finally willing to hand over. I have two bets. First: GPT‑6.2 gives OpenAI a serious comeback in complex coding. I’m betting that Codex takes the lead over Claude Code in independent coding-agent tests while keeping a lower cost per task. The interesting part would be what that means for the difficult work: tracing an unfamiliar codebase, fixing a bug across several files, checking the result, and not leaving a second mess for you to clean up. Another model name in a menu wouldn’t change much. A model that makes you reconsider which tool gets your hardest tasks would. Second: Fable 5.5 makes a stronger case for AI as a research collaborator. Current models already contribute to real research. The next jump I’m betting on is a public, externally validated case where the agent drives several rounds of exploration: proposes an approach, tests it, sees why it failed, and chooses a better route with less step-by-step direction from people. The result needs to be reproducible. A convincing explanation alone won’t do. That would change the kinds of questions people bring to AI. Instead of arriving with a plan and asking for help executing it, you could arrive with a problem and ask it to help find the plan. Neither model’s launch is confirmed in the official sources I’ve checked. These are November bets, not inside information. I’ll come back on December 1, including if neither happens.

曝光 255 · 评论 6 · 点赞 5 · 书签 0