@danellisona Billions go into making models smarter, and a tiny fraction goes into making them verifiable. If I'm worried about anything, it's that ratio. More bets should go to teams like @runlatentai, catching hallucinations by reading model internals.
@GoldenstateAce SF still wins on foundation models, but more of the application layer is getting built in NYC. Ramp, Hebbia, and Runway are all here for a reason
@TheArslanNasir@HarryStebbings fair, and I should've said on top of seats not instead of. but 6 cents a convo is close to raw inference cost. hard to build venture margins there, which is kind of the consolidation argument.
@HeyAliux It'll replace the music nobody chose to listen to. Stock tracks, ad jingles, background playlists. The artists people actually follow get more valuable, because the one thing AI can't fake is a real person on stage.
@danellisona Billions go into making models smarter, and a tiny fraction goes into making them verifiable. If I'm worried about anything, it's that ratio. More bets should go to teams like @runlatentai, catching hallucinations by reading model internals.
@EbiThinks Because software is priced on value and switching costs, not on what it costs to build. Cheaper code shows up as more competitors, not lower prices. Give it 18 months
@reflection_ai The chart tells the real story: Beam beats every Western open model on all six benchmarks and trails Qwen 3.8 Max on each one Qwen reports. The pitch is provenance more than leaderboard. For buyers who won't run Chinese weights, that is the whole market.
Funny reversal in consumer agents. Meta, the ad company, launched Muse with $20 and $100 paid tiers. Instinct, the 14 person startup, is free and its CEO has floated ads. Somebody's playbook is upside down