If you don't vote, I don't want to hear you ever complain about any of the filth, crime, or decline in LA again. Get off your butt and VOTE, or enjoy what you get.
Introducing MAI-Code-1-Flash
Microsoft's latest small coding model.
51.2% on SWE-Bench Pro.
Rolling out now to @GitHub Copilot Free, Pro, Pro+ And Max users.
https://t.co/vQTuCTQTnF
At some point, my forked opencode repo will no longer be able to track all the changes happening in the original space. Codex failed to rebase cleanly, but my forked version of opencode somehow managed to do it. Happy to see that. Lets see how long I can keep up :)
The best way I have found to benchmark if the model is good or not - have then rebase on the top of branch that is 100s of commits ahead and 100s of commit behind the remote branch.
This is the reason why breaking down the project into byte size chunks and tasks is so important. Given that AI is able to output code at such a faster pace, any project out there imo, should focus on laying the foundation slowly and steadily. The same way you would for non-AI assisted ones. I have been there and feel the anxiety.