Progress is a trend. That doesn’t prove we’re racing toward self-improving superintelligence that could kill us this decade.Current LLMs are still limited. Jumping from that to “superhuman systems that can hack anything and seize power” needs evidence, not private fear and a race story.
Popular Fable 5.1 quota hack: let Fable only plan and review, push all the coding to Opus/Sonnet subagents.
I think it's backwards. Subagents hand back summaries, and summaries are where the evidence disappears: which leads got ruled out, what the tests actually showed.
Split work horizontally (Fable plans, Opus builds the whole thing, Fable reviews), not vertically (Fable as foreman over a crew reporting one-liners).
Don't make Fable your general contractor. Make it the architect and the site supervisor.
https://t.co/6glGVkkC9C