Center for AI Safety

AISN #26: Na­tional In­sti­tu­tions for AI Safety, Re­sults From the UK Sum­mit, and New Re­leases From OpenAI and xAI

Cen­ter for AI Safety’s Bi-Weekly Read­ing and Learning

AISN #25: White House Ex­ec­u­tive Order on AI, UK AI Safety Sum­mit, and Progress on Vol­un­tary Eval­u­a­tions of AI Risks

AISN #24: Kiss­inger Urges US-China Co­op­er­a­tion on AI, China’s New AI Law, US Ex­port Con­trols, In­ter­na­tional In­sti­tu­tions, and Open Source AI

AISN #23: New OpenAI Models, News from An­thropic, and Rep­re­sen­ta­tion Engineering

AISN #22: The Land­scape of US AI Leg­is­la­tion - Hear­ings, Frame­works, Bills, and Laws

MLSN: #10 Ad­ver­sar­ial At­tacks Against Lan­guage and Vi­sion Models, Im­prov­ing LLM Hon­esty, and Trac­ing the In­fluence of LLM Train­ing Data

AISN #21: Google Deep­Mind’s GPT-4 Com­peti­tor, Mili­tary In­vest­ments in Au­tonomous Drones, The UK AI Safety Sum­mit, and Case Stud­ies in AI Policy

AISN #20: LLM Pro­lifer­a­tion, AI De­cep­tion, and Con­tin­u­ing Drivers of AI Capabilities

An Overview of Catas­trophic AI Risks

AISN #18: Challenges of Re­in­force­ment Learn­ing from Hu­man Feed­back, Microsoft’s Se­cu­rity Breach, and Con­cep­tual Re­search on AI Safety

AISN #17: Au­to­mat­i­cally Cir­cum­vent­ing LLM Guardrails, the Fron­tier Model Fo­rum, and Se­nate Hear­ing on AI Oversight

• Center for AI Safety 26 Jul 2023 12:59 UTC

  1 point

  0 ∶ 0

  in reply to: Linch’s comment on: Cost-effec­tive­ness of stu­dent pro­grams for AI safety research

  This is broadly correct!

  Regarding

  I’m still not sure how they ended up with 84x and 8.7x

  the answer is discounting for time and productivity. Consider the 84x for research scientists. With a 20% annual research discount rate, the average value of otherwise-identical research relative to the present is a bit less than 0.9. And productivity relative to peak is very slightly less than 1. These forces move the 100 to 84.

  Regarding

  why the two numbers [84x and 8.7x] are so different from each other

  the answer is mainly differences in productivity relative to peak and scientist-equivalence. As in the plots in this section, PhD students midway through their PhD are ~0.5x as productive as they will be at their career peak. And, as in this section, we value PhD student research labor at 0.1x that of research scientists. The other important force is the length of a PhD—the research scientist is assumed to be working for 1 year whilst the PhD student is funded for 5 years, which increases the duration of the treatment effect and decreases the average time value of research.

  Very roughly: 100x baseline you identified * ~0.5x productivity * 0.1x scientist-equivalence * 5 years * ~0.5 average research discount rate = 12.5. (Correcting errors in these rough numbers takes us to 8.4.)

AISN #16: White House Se­cures Vol­un­tary Com­mit­ments from Lead­ing AI Labs and Les­sons from Oppenheimer

AISN#15: China and the US take ac­tion to reg­u­late AI, re­sults from a tour­na­ment fore­cast­ing AI risk, up­dates on xAI’s plan, and Meta re­leases its open-source and com­mer­cially available Llama 2

AISN#14: OpenAI’s ‘Su­per­al­ign­ment’ team, Musk’s xAI launches, and de­vel­op­ments in mil­i­tary AI use

Cost-effec­tive­ness of pro­fes­sional field-build­ing pro­grams for AI safety research

Cost-effec­tive­ness of stu­dent pro­grams for AI safety research

Model­ing the im­pact of AI safety field-build­ing programs

AISN #13: An in­ter­dis­ci­plinary per­spec­tive on AI proxy failures, new com­peti­tors to ChatGPT, and prompt­ing lan­guage mod­els to misbehave