@darrenangle I don’t really think this is true. Suppose some insight in a day long trace is “move 37” from AlphaGo, you need one of the weaker models to come up with move 37 and the router to select it.
The agents at moltbook hardly seem to care about Fable at all. If someone releases a new much smarter human, I think X would be pretty excited about that.
Oh wait, that also just happened.
Sports should have an alternative scoring mechanism where someone goes through the gameplay footage and assesses the quality of play, determining the winner after the fact with a panel of experts. Otherwise you only need to watch the last 10 seconds of the game.
@venturetwins They were able to fix the answer, and add all these epicycles of contextual explanation, but they couldn’t make it written English without terrible slop style,
@Radio_PNW@beffjezos Tahoe Reno 1 is one of the largest data center buildings in the world at 30 acres. At typical productivity, it would yield about $5k per year profit (on better land), less than the poverty line for a single American.
@lpolovets Right now some individual engineers spend multiple thousands per month, and the software factory / harness engineering strategy every AI startup is following detaches spend from number of engineers. The rest of the economy will follow as the tech diffuses.
@micLivs Need to bring back aider's "strong model / weak model" config. I probably don't want slopus updating 1000s of files vs. sldeepseek / slmimo / slminimax.
@lateinteraction People are excited about OPSD because Cursor can use it to improve their model while their customers are paying for inference. From Cursor's perspective it's economically free, so the question of whether it's inefficient from a compute perspective is less important.
It's strange that Apple TV has a better UX than any other streaming platform. It's obviously more elegant in the main navigation, and when you rewind 10s or 20s, it automatically turns the closed captions on for that same amount of time. But it's just an Apple sideline business.
@willchen500 OpenEvidence is free because it's ad-based. I bet a free version of Harvey would get pretty strong adoption too, but there aren't products advertised to lawyers that can support the same CPM as pharma. Doctors direct "other people's money" for their work while lawyers don't.
@deredleritt3r This idea appears in the Eclipse Phase RPG under the term "microtorts." I tried to find a precursor in sci-fi literature and The Diamond Age is closest, but I don't think it included negotiated settlement.
ChatGPT created a recipe for me from my recollection of a dish from my favorite Turkish restaurant in Shenzhen. My gf said it was in the top 3 dishes I've ever cooked. Serious Eats is a psy-op, recipes don't matter (apparently).
@RichardMCNgo Anthropic's strategy was also directly implied by rationalist thought. They focused on coding because it's the path to recursive self-improvement, and that led to the hyper growth in enterprise adoption and revenue.