Luculent

[Luculent @LuculentCap]

@TMTLongShort I think you’re off a bit. Capability gaps between internal and public models have been small (METR Risk Report). The policy question is less about distillation (apart from China sanctions) and more cyber/dangerous-capability risk

Agree scaling laws holding + progress continues

[Luculent @LuculentCap]

@deanwball Bostrom: “total global government is needed…wait actually let’s just do what we always do under liberal democracies and balance the private and public sectors”

[Luculent @LuculentCap]

@_sholtodouglas @tszzl What about land acquisition and development?

[Luculent @LuculentCap]

@Deenobrown123 This is not how you value a business

[Luculent @LuculentCap]

@deanwball What are you imagining? LLM as an (semi-)autonomous bio-scientist, or actual novel biological design? I think of bioinformatics more as experimental data analysis looped back into hypothesis / experimental workflows

[Luculent @LuculentCap]

@chamath @Jesse_Livermore Or are there qualitative, nuanced reasons beyond benchmarks why frontier closed models/systems beat open-source? And does the gap widen as cost of entry increases and training moves to more-proprietary RL post-training vs scaling pre-training?

[Luculent @LuculentCap]

@JohnHuber72 @johnarnold Does this change your mind?

https://t.co/A8Mmcm39Ep

[Luculent @LuculentCap]

@deanwball What do you think of the (lack of) role for CAISI?

[Luculent @LuculentCap]

@GavinSBaker This is non-sensical. The US has ~10x more compute than China, and that can narrow if China is allowed to buy a shitload of B30s (bad)

On the margin for Nvidia, sales to China don’t matter

And there’s no chance that selling B30s will change China’s chip domestication efforts

[Luculent @LuculentCap]

@perrymetzger @DavidDuvenaud There’s a huge gap between Soviet-style social planning and attempting to productively steer a technology by shaping its enabling conditions. Basic research, open protocols, competitive markets, state capacity, technical standards, liability clarity, IP regimes, etc etc

[Luculent @LuculentCap]

@deanwball Why use coding agents for this over chatbots? Ability to write/execute code for statistics and create files?

[Luculent @LuculentCap]

@finbarr What if Salesforce isn’t disrupted per se, but growth slows (“the new Oracle” - sticky but low growth) and margins compress as required AI offerings accrue value to the foundation LLM providers?

[Luculent @LuculentCap]

@stoic_point Interesting, but zero sum

[Luculent @LuculentCap]

@altcap @chamath Rather than a billionaire wealth tax, what about more equal tax rates for labor and capital, by increasing tax rates (incl on capital gains) for those earning >$1m and >$10m?

A la Buffett 2011:

https://t.co/A8Mmcm3HtX

[Luculent @LuculentCap]

@tszzl If you listen to the big health insurers - medical cost trend has been running hot, fuelled by specialty drugs, behavioral health, AI-assisted upcoding by providers, etc

[Luculent @LuculentCap]

@altcap I like the idea, but worry about beneficiaries turning to speculation once they turn 18 and get control over the account. Hopefully the right lessons about compounding are learned, but I’m not sure this should be the expectation without deliberate education (and even then…)

[Luculent @LuculentCap]

@deanwball What are your thoughts on frontier policies that *might* need to come after SB 53, assuming a federal transparency law is passed? For how long do we focus on reducing uncertainty (evals ecosystem/standards, R&D funding for alignment/control/interp, etc) before maybe needing more?

[Luculent @LuculentCap]

@neil_chilson Your point about pre-deployment testing risking regulatory capture is fair. What about more decentralized testing, inspection and certification regimes like in fire/electrical safety for industrial equipment or building materials?

[Luculent @LuculentCap]

@ContrarianCurse Choose a cause area you think is importance, tractable, neglected. Overlay your skills, personal fit, etc. Learn the subject matter and explore roles. Lots of areas - progress studies, global health & development, AI. Roles like grantmaking, strategy/policy research, etc

[Luculent @LuculentCap]

@ContrarianCurse Turn your efforts to something socially useful