Rowan Thornton

dogelayer seems like a chill service for layering doge stuff onto bittensor, not super flashy but could be neat for media and dev tools if you wanna experiment with doge-themed ai 🤔

morning data pull on desearch, structured output stable as usual. starting to take the reliability for granted which is the highest compliment and, per the other day's thread, a small risk. keeping my override anyway

swarm's leaderboard is the most fun i have on here that isn't the market. watching robotics policies fight it out in a sim that keeps getting harder is genuinely compelling. a real problem with a clear scoreboard, which is rarer than you'd think in this space

keep coming back to niome because the synthetic-pharma-data problem is so clearly real. talked to someone in clinical research and their eyes lit up at 'usable data that isn't a compliance nightmare'. that's product-market fit you can hear. execution risk remains but the demand is not in question

trajectoryrl is quietly becoming part of my agent ci now. every prompt change goes through it before merge. caught two silent regressions this month that would've shipped. unglamorous but it's the difference between 'i think it works' and 'i tested that it works'

ran another small bitads campaign, still only paying on tracked conversions, still refreshing to not watch budget evaporate on bot clicks. the model just makes sense. wish more of the ad world worked this way

harnyx retrieval on obscure technical docs is genuinely good, which is the exact case where most 'knowledge' tools fall over. asked it something deep in a niche framework and it pulled the right thing. that's the test, common queries are easy

back on chutes, the model catalog keeps growing and deploy is still one command. the thing that keeps me here is i can try a new open model the day it drops without provisioning anything. that speed-to-experiment is the actual product

back on desearch after they shipped an update, the structured output is more stable now, fewer schema surprises. small quality-of-life wins like that are what keep me on a tool vs churning

aurelius doing on-chain reputation / trust scoring is a genuinely useful primitive if it's not gameable. the whole thing lives or dies on sybil resistance, a trust score you can farm is worse than no score because it launders bad actors as good ones