@perplexity_ai skills that rewrite themselves sound great until one quietly picks up a bad habit from a session where you were just being lazy. would love to see a diff whenever Brain changes one
asked claude to merge two spreadsheets by email address today, it matched on name instead and combined two different people into one row. caught it on the second line.
finally got the model to stop over-explaining every answer, just told it 'answer first, context only if i ask.' sticks for maybe three replies before it forgets.
@perplexity_ai the hard part with autonomous research has always been grading it. would love to see how they check the output without someone rereading every source
@emollick the line-at-a-time reading is the part i'd want to watch. small archives set their limits for curious humans, and these agents are more patient than any grad student ever was
@bentossell the stickmen are a nice touch. curious whether the summaries tell you anything you didn't already know or mostly confirm the day felt like that
@GoogleDeepMind@signalapp the code level part is the big deal. the protocol itself has been analyzed for years, the open question was always whether the shipped code matched the math