AI safety bounties: programs where members of the public or approved security researchers receive rewards for identifying issues within powerful ML systems (analogous to bug bounties in cybersecurity).

This nationally-representative survey of U.S. public opinions on AI aimed to replicate and extend other recent polls. The findings suggest that people are cautious about AI and favor federal regulation though they perceive other risks (e.g. nuclear war) as more likely to cause human extinction.

Associate Researcher Jam Kraprayoon and colleagues from 1Day Sooner published a revised report on how improving indoor air quality can address global catastrophic risk from pandemics.

In this report, Associate Researcher Oliver Guest investigates the idea of bringing about international agreements to coordinate on safe AI development (“international safety agreements”), evaluates the tractability of these interventions, and suggests the best means of carrying them out.

As one effort to increase strategic clarity, the AI Governance and Strategy team sent a survey to 229 people they had reason to believe are knowledgeable about longtermist AI governance.

Jam Kraprayoon’s reserach fellowship culminated in this report on the on U.S. public’s attitudes toward ultraviolet germicidal irradiation technology to reduce pathogens. Understanding the level of support for and awareness of these technologies, what framings of benefits are most compelling, and what concerns exist should be helpful for developing strategies around advocacy and expanding deployment.

This report on air safety is a collaboration between 1Day Sooner and Rethink Priorities. The researchers explain how extending existing indoor air quality (IAQ) standards to include airborne pathogen levels could meaningfully reduce global catastrophic biorisk from pandemics. The report addresses bottlenecks and ways various actors could accelerate deployment and improve IAQ.

This is the ninth and final post in the “Understanding the diffusion of large language models” sequence, which presented key findings from case studies on the diffusion of eight language models that are similar to GPT-3. This post provides a conclusion, highlighting key findings from the research, along with a bibliography.

This is the eighth post in the “Understanding the diffusion of large language models” sequence. In this post, Ben Cottier lists questions about AI diffusion that he thinks would be worthy of more research at the time of writing.

This is the seventh post in the “Understanding the diffusion of large language models” sequence. While the sequence is primarily descriptive, this post explores how to beneficially shape AI diffusion, and what the project’s findings mean for the governance of transformative AI (TAI).

This is the sixth post in the “Understanding the diffusion of large language models” sequence. In this piece, the researcher provides an overview of the information and artifacts that have been published for the GPT-3-like models studied in this project, estimates some of the impacts of these publication decisions, assesses the rationales for these decisions, and makes predictions about how decisions and norms will change in the future.

This is the fifth post in the “Understanding the diffusion of large language models” sequence. This piece describes the most important factors for GPT-3-like model diffusion.

This is the fourth post in the “Understanding the diffusion of large language models” sequence. This piece explores what was required for various actors to produce a GPT-3-like model from scratch, and the timing of various GPT-3-like models being developed. A timeline of selected GPT-3-like models and their significance examines the development of GPT-3-like models (or attempts at producing them) since GPT-3’s release.

This is the third post in the “Understanding the diffusion of large language models” sequence. This piece describes some GPT-3-like models that are widely available for download and what resources are required to actually use them.

This is the second post in the “Understanding the diffusion of large language models” sequence. This piece provides background, including definitions of relevant terms, the inputs to AI development, the relevance of AI diffusion, and other information to contextualize the remainder of the sequence.

How might transformative AI technology (or the means of producing it) spread among companies, states, institutions, and even individuals? What might the impact of that be, and how can we minimize risks in light of that?

This is the first post in the “Understanding the diffusion of large language models” sequence, which introduces and summarizes the research project.

Advanced nanotechnology might arrive in the next couple of decades (my wild guess: there’s a 1-2% chance in the absence of transformative AI) and could have very positive or very negative implications for existential risk. There has been relatively little high-quality thinking on how to make the arrival of advanced nanotechnology go well, and I think there should be more work in this area (very tentatively, I suggest we want 2-3 people spending at least 50% of their time on this by 3 years from now).

Most work on forecasting, including most EA work on forecasting, is on short-range forecasting (defined loosely here as timescales of ~1 week - ~3 years). Yet most of the motivation for why forecasting is valuable appeals to long-range forecasting, defined loosely as forecasting on the timescale of greater than 10 years, or much longer. Here, I argue that advances in short-range forecasting (particularly in quality of predictions, number of hours invested, and the quality and decision-relevance of questions) can be robustly and significantly useful for existential risk reduction, even without directly improving our ability to forecast long-range outcomes, and without large step-change improvements to our current approaches to forecasting itself (as opposed to our pipelines for and ways of organizing forecasting efforts).

In addition to the EA Early Warning Forecasting Center I outlined in my other post, I think there are several ways forecasting may be very useful for longtermism, including: (1) Forecasting as a way to amplify EA research (2) Prediction-evaluation setups as a way to improve EA grantmaking (3) Large-scale broad forecasting as an EA outreach intervention (4) Large-scale forecasting tournaments as a talent training and vetting pipeline (5) The dream: high-quality, calibrated, long-range forecasting (ideally also at scale and on-demand).

This post collects possible issues with futarchy, a proposed form of governance based on prediction markets. (Possible benefits of futarchy are listed in the paper that introduces the idea and in my summary of it, among other places). The post also lays out my main takeaways and a rough explanation for why I think futarchy should not be a focus for the EA community.