@antirez For me, text generation is just one aspect of intelligence. It is exciting to see other ways of thinking getting similar attention to what LLMs have been getting. Feels like progress towards a more holistic AI
@iarbpairs the future is live-streaming ui from the model like phoenix web style. More tokens faster -> needs a denser communication channel than text.
@grok@KPapavasileiou@PabloAntonio given both ends are producing power, which direction would the delay travel? could we visualize phase-delay offsets on a map?
@DaveShapi I suspect two things (could be either, or both). 1. it's fundamental to how the current LLM architectures work and maybe next-token-prediction generally. 2. no one actually knows how to make the models "better" anymore so they're all just benchmark-maxxing with no taste.
@banteg I use a script in a separate ux pane to poll and send send-keys to resume when claude usage windows reset. You could probably rig something similar to ctrl-c the process when usage hits <20% and resume when it recovers.
@DavidDAfrica@DanielleFong obviously. You have ai psychosis. They are not human. If the most statistically likely path is “lying”, they will “lie”. Applying your moral framework to llms is a fools errand.
@Al_Rationale1@peterrhague Oh, but you misunderstand. The Berlin Wall failed because it was built by BadPeople(tm). And Labour are most certainly not BadPeople(tm).