Practical, beginner-sized projects in technical AI safety, biosecurity, governance, policy and philosophy. Each one lists the time it takes, what you need and where the idea comes from.
Searches every project, mentored research project, library resource and glossary term. Try flattery, pandemics, chips, consciousness.
108 projects · 38 with step-by-step guides · 228 mentored research projects
Choose your background to see three projects to start with.
Write a small, repeatable test that checks whether a model admits when it does not know something, using Inspect, the open-source evaluation framework used by the UK AI Security Institute.
Make a tiny calculator (a spreadsheet, notebook or web page) that tells anyone how much a portable air purifier would clean the air in their room.
Download Epoch AI’s free dataset of notable AI models, chart training compute over time in Python, estimate how fast it grows, and count how many models cross the EU AI Act’s compute threshold.
Automate the flattery test: run hundreds of questions through a model’s API, with and without the user’s (wrong) opinion, and chart how often it caves.
Use the rule of thumb “training compute ≈ 6 × parameters × training tokens” to estimate the compute behind a model, then compare your answer with Epoch AI’s database.
The MIT AI Risk Repository organises over a thousand AI risks into a shared taxonomy. Use it to classify 20 recent news stories, then look at what the pattern tells you.
The MIT AI Risk Repository organises over a thousand AI risks into a shared taxonomy. Use it to classify 20 recent news stories, then look at what the pattern tells you.
At the 2024 AI Seoul Summit, frontier AI companies signed voluntary safety commitments, including publishing a safety framework. Pick five signatories and check what they actually published.
How did ancient and medieval societies protect themselves from plague? Research one historical practice (quarantine, isolation, trade rules) and draw a lesson for today.
UK public bodies publish records about the algorithmic tools they use under the Algorithmic Transparency Recording Standard. Read five records and assess what is disclosed and what is missing.
Write a short, jargon-free explainer of one AI safety idea (reward hacking, sycophancy, compute governance, DNA screening) for someone outside tech.
Read two frontier AI companies’ published safety frameworks and compare them side by side using METR’s list of the elements these policies usually share.
Write a short, jargon-free explainer of one AI safety idea (reward hacking, sycophancy, compute governance, DNA screening) for someone outside tech.
Companies make synthetic DNA to order. Screening those orders and the customers who place them is one of the main defences against engineered pandemics. Explain how it works and where the gaps are.
Why do so many governance proposals focus on chips? Read one key paper and write a jargon-free explainer for someone outside tech.
Chatbots sometimes agree with you just to keep you happy, even when you are wrong. Researchers call this sycophancy. Test for it yourself with five questions and two ways of asking.
Write a short, jargon-free explainer of one AI safety idea (reward hacking, sycophancy, compute governance, DNA screening) for someone outside tech.
A short video, a carousel, an illustrated explainer or a podcast segment: make one piece of honest, accurate AI safety content for an audience you understand.
Each theme has guided projects, larger project ideas, mentored research and reading.
Something small enough to finish in an afternoon or a weekend. For example, test whether a chatbot changes its answer to agree with you, explain three real cases of an AI gaming its goal, or compare the safety promises AI companies have made. This site has 38 guided projects like these, each with step-by-step instructions.
No. Many projects only need a free chatbot account or a web browser. If you do code, there are Python projects too, such as small model evaluations and interpretability experiments. The project matcher asks how you like to work and suggests projects to match.
Yes. The 5 themes here are technical AI safety, biosecurity, frontier AI governance, AI policy and AI safety × philosophy. There are projects for people from biology, medicine, law, politics, economics, philosophy and the humanities, as well as computer science and maths.
Guided projects take 1 to 3 hours or a weekend. Larger project ideas take a few weeks to a few months. Every project page says how long it takes, what you need and what you will have made at the end.
Yes. There is no sign-up, no cookies and no tracking. Most projects cost nothing to do. A few need a small budget for AI API calls or GPU time, and the project page says so.
Write up what you did and share it, for example on LessWrong, the EA Forum or a blog. Then take a free course such as BlueDot Impact's, apply for mentored research such as SPAR (228 SPAR projects are listed here), or join a local AI safety group. The Next steps page lists these.
AI systems are becoming more capable quickly. AI safety is the work of making sure they behave as intended and don’t cause serious harm.
It includes testing and interpreting models, reducing the chance that AI helps cause a pandemic, and shaping the rules and institutions around AI. People come to it from computer science, biology, law, policy and philosophy.
Find a project