acb12

It would be great if they atleast continue to provide free plans for OSS contributors. Currently they do provide quite generous plans like Claude Max 20x but its just for 6 months.

How is it ensured that the AI agent will actually search for the given information vs ignoring the instruction as its always been the case with any skill or system prompt?

Maybe old software was also fast because the people building it were more likely to care about computers first and careers second. Not that they were better, just that fewer people were there because tech was the obvious high-paying path.

What are the minimal hardware requirements to run a reasonable good local model for real SWE work these days? And have a reasonable inference speed? Seems like the requirement are pretty high and the inference speed is not ideal.