For most of my life, I’ve been obsessed with how to build critical infrastructure that changes the world. That compass led me to spend years modeling renewable energy systems at MIT, power grids and demand clusters at Oxford.
YC companies, before they're in the YC directory 👇
1. @usetokenless = a router that fans your prompt out to a group of models and bills you for the one that was right
2. @superradcompany = secured agentic computing, creators of microsandbox
3. @princepsdev = insurance for compute, underwriting GPU downtime and SLA breaches
4. @hbr_pbc = the data curation layer robotics teams are missing
5. @blaisedotnew = an AI system that builds your enterprise's internal software (two ex-lovable founders)
6. @securewithinner = detonates every npm and pypi package in a sandbox before your coding agent installs it
7. CueBench (@neelapolitan) RL environments for the things frontier models fail at 80% of the time
8. Kandor (@devansh_1704) catches undercoding and fights denials for specialty clinics, autonomously
9. Veltha (@edin_citaku) an AI claims adjuster that cites the regulation
10. Shepherd (@stochasticsloth) one memory for research, hardware, firmware, software and ops
11. E.env (@rricefan) market-derived RL environments that never saturate
Thank you @JoinFAI and @poastmilque for hosting a wonderful evening - @miramurati on open weight models and @mkratsios47 on the golden age of American innovation. Hard to think of a more important conversation than who gets to build the future of AI.
June’s @bBuilderSF meetup was all about running AI locally
@0xSero joined us to walk through which open models to use, how to choose the right hardware, and the tradeoffs across cost, privacy, and performance ↓
Timestamps:
0:00 - An example of local AI doing real work
0:39 - Why cloud AI keeps getting more expensive
4:10 - The best models might disappear
6:43 - The case for running AI locally
11:25 - Choosing the right local AI setup
14:12 - How to benchmark local models
22:03 - The best local AI hardware options
26:18 - Are powerful models becoming weapons?
29:40 - How model compression actually works
35:48 - The future of small local models
EARLY: @PRINCEPSdev - 7 followers - Compute
building the price of compute risk: a live benchmark for gpu downtime, failure, and model-output liability, priced from telemetry instead of loss history. actuarial tables for silicon.
founder: @anya_trofimova7 (ex @PalantirTech)