
Lessons from Neel Nanda
The items below are things Neel Nanda said on podcasts about interpretability research and AI safety. — Neel Nanda on The 80,000 Hours Podcast on Artificial Intelligence, Neel Nanda on the race to read AI minds (June 2026).
On The 80,000 Hours Podcast, Nanda said: "I think in one line, my journey can be summarized as going from idealistic ambition to optimistic pragmatism". — Neel Nanda on The 80,000 Hours Podcast on Artificial Intelligence, Neel Nanda on the race to read AI minds (June 2026).
On The 80,000 Hours Podcast, Nanda said: "We should look in the middle of this system because neural networks aren't actually a black box". — Neel Nanda on The 80,000 Hours Podcast on Artificial Intelligence, Neel Nanda on the race to read AI minds (June 2026).
On The 80,000 Hours Podcast, Nanda said: "There's a good chance we make some progress but fall short, but there's enough of a chance that we get really far that this isAn important and impactful thing to do". — Neel Nanda on The 80,000 Hours Podcast on Artificial Intelligence, Neel Nanda on the race to read AI minds (June 2026).
On The 80,000 Hours Podcast, Nanda said: "You lose information when you compress, and we need to recover what was originally there in order to understand all the concepts, which is quite a difficult problem". — Neel Nanda on The 80,000 Hours Podcast on Artificial Intelligence, Neel Nanda on the race to read AI minds (June 2026).
On The 80,000 Hours Podcast, Nanda said: "You can notice this from the chain of thought, but if you train the chain of thought not to have this, they still do it a bunch, but they just never talk about it". — Neel Nanda on The 80,000 Hours Podcast on Artificial Intelligence, Neel Nanda on the race to read AI minds (June 2026).
On The 80,000 Hours Podcast, Nanda said: "Observation two is that I think mech interp requires unusually low amounts of compute compared to lots of frontier capabilities research, because we don't need to train language models". — Neel Nanda on The 80,000 Hours Podcast on Artificial Intelligence, Neel Nanda on the race to read AI minds (June 2026).
On The 80,000 Hours Podcast, Nanda said: "Like, one thing we haven't even really gotten into is that SAEs are just quite expensive". — Neel Nanda on The 80,000 Hours Podcast on Artificial Intelligence, Neel Nanda on the race to read AI minds (June 2026).
On The 80,000 Hours Podcast, Nanda said: "I personally am of the opinion that if your thing is going to be useful in two years, it should probably be doing something useful now". — Neel Nanda on The 80,000 Hours Podcast on Artificial Intelligence, Neel Nanda on the race to read AI minds (June 2026).
On The 80,000 Hours Podcast, Nanda said: "Unless you go read the paper carefully or listen to someone grumpy like me complaining, you don't hear the nuances and the caveats, even if the researchers who wrote the paper were very careful about documenting this". — Neel Nanda on The 80,000 Hours Podcast on Artificial Intelligence, Neel Nanda on the race to read AI minds (June 2026).
On The 80,000 Hours Podcast, Nanda said: "I think this is true in most research fields, but I think that mech interp maybe has a disproportionately high amount of really flashy and appealing looking results". — Neel Nanda on The 80,000 Hours Podcast on Artificial Intelligence, Neel Nanda on the race to read AI minds (June 2026).
On The 80,000 Hours Podcast, Nanda said: "You need to have some reason to think you are systematically selecting true claims from the ether of things that maybe could be true, and human intuition is not good enough". — Neel Nanda on The 80,000 Hours Podcast on Artificial Intelligence, Neel Nanda on the race to read AI minds (June 2026).
On The 80,000 Hours Podcast, Nanda said: "These are all proxies, but I've personally concluded that downstream tasks are a bit closer to the proxies I care about". — Neel Nanda on The 80,000 Hours Podcast on Artificial Intelligence, Neel Nanda on the race to read AI minds (June 2026).
On The 80,000 Hours Podcast, Nanda said: "Maybe I was wrong, and ambitious interpretability that tries to fully reverse engineer things is actually great, and in a year they'll just have some epic new research results and I'll seem really wrong". — Neel Nanda on The 80,000 Hours Podcast on Artificial Intelligence, Neel Nanda on the race to read AI minds (June 2026).
On The 80,000 Hours Podcast, Nanda said: "I think there is a bunch to do, and I want more talented people to do it". — Neel Nanda on The 80,000 Hours Podcast on Artificial Intelligence, Neel Nanda on the race to read AI minds (June 2026).
On The 80,000 Hours Podcast, Nanda said: "Even ignoring my accumulated context, I think that my mindset's just a very good fit to mech interp". — Neel Nanda on The 80,000 Hours Podcast on Artificial Intelligence, Neel Nanda on the race to read AI minds (June 2026).
On The 80,000 Hours Podcast, Nanda said: "If you finish a thing, that's a fun bonus, but maybe you realize two weeks in that you had terrible taste and this is a bad project idea". — Neel Nanda on The 80,000 Hours Podcast on Artificial Intelligence, Neel Nanda on the race to read AI minds (June 2026).
On The 80,000 Hours Podcast, Nanda said: "I think this is a skill I've refined over doing maths Olympiads and a maths degree, but I know a lot of people who are excellent at this with totally different backgrounds". — Neel Nanda on The 80,000 Hours Podcast on Artificial Intelligence, Neel Nanda on the race to read AI minds (June 2026).
On the Google DeepMind podcast, Nanda said: "Chain of thought is helpful, and we should expect it to tell us something in the same way that reading my scratch pad will probably tell you something about how I'm doing the maths problem". — Neel Nanda on Google DeepMind: The Podcast, Understanding the inner thoughts of AI (July 2026).
On the Google DeepMind podcast, Nanda said: "If you're smart enough, you might miss out certain key steps because you realize people might look at the scratchpad". — Neel Nanda on Google DeepMind: The Podcast, Understanding the inner thoughts of AI (July 2026).
On the Google DeepMind podcast, Nanda said: "Lying from a very competent model will look the same as telling the truth if it's good enough to not be caught out in obvious mistakes". — Neel Nanda on Google DeepMind: The Podcast, Understanding the inner thoughts of AI (July 2026).
On the Google DeepMind podcast, Nanda said: "Future models might be, and this is why one of my priorities is getting good at evaluating how aligned a system is". — Neel Nanda on Google DeepMind: The Podcast, Understanding the inner thoughts of AI (July 2026).
On the Google DeepMind podcast, Nanda said: "It's getting increasingly important to understand what is the psychology of a language model". — Neel Nanda on Google DeepMind: The Podcast, Understanding the inner thoughts of AI (July 2026).
On the Google DeepMind podcast, Nanda said: "It's just understanding everything and understanding all of the, like, messy fine details where I think we maybe need to be a bit more realistic". — Neel Nanda on Google DeepMind: The Podcast, Understanding the inner thoughts of AI (July 2026).
In a getting-started guide to interpretability, Nanda wrote: "Key skills, like research taste and the ability to prioritize, take time to develop". — Neel Nanda, getting started in mechanistic interpretability (neelnanda.io).
In a getting-started guide to interpretability, Nanda wrote: "As a tool to help with deep dives - you need to actually read the paper, but I recommend having the LLM chat open as you read with the paper in the context and asking it questions, for context, etc every time you get confused". — Neel Nanda, getting started in mechanistic interpretability (neelnanda.io).
In a getting-started guide to interpretability, Nanda wrote: "Instead, I think we should just more directly try to do pragmatic work that enables us to do useful things using internals". — Neel Nanda, getting started in mechanistic interpretability (neelnanda.io).
In a getting-started guide to interpretability, Nanda wrote: "Distillation: Once you’re convinced, your north star is to compress your findings into concise, rigorous truth that you can communicate to the world - create enough experimental evidence to convince others, write it up clearly, and share it". — Neel Nanda, getting started in mechanistic interpretability (neelnanda.io).
In a getting-started guide to interpretability, Nanda wrote: "In my experience, you will often spend more than half of a project trying to figure out what the hell is happening and what you think your plan is". — Neel Nanda, getting started in mechanistic interpretability (neelnanda.io).
In a getting-started guide to interpretability, Nanda wrote: "I used to be very excited about ambitious reverse engineering, but I currently think that the dream of completely reverse engineering a model down to something human understandable seems basically doomed". — Neel Nanda, getting started in mechanistic interpretability (neelnanda.io).
In a getting-started guide to interpretability, Nanda wrote: "Be good at managing your time and tasks so that once you've decided what the highest priority thing to work on is, you in fact go and work on it". — Neel Nanda, getting started in mechanistic interpretability (neelnanda.io).
In a getting-started guide to interpretability, Nanda wrote: "It's fine to put this in the appendix if space is scarce, but it should be there". — Neel Nanda, getting started in mechanistic interpretability (neelnanda.io).
In a getting-started guide to interpretability, Nanda wrote: "Common mistakes The reader does not have context: Your paper will be clear in your head, because you have just spent weeks to months steeped in this research project". — Neel Nanda, getting started in mechanistic interpretability (neelnanda.io).
In a getting-started guide to interpretability, Nanda wrote: "Honestly, I think that people new to the field should not worry that much about this". — Neel Nanda, getting started in mechanistic interpretability (neelnanda.io).
In a getting-started guide to interpretability, Nanda wrote: "If you want to spend years going deep on a niche topic that no industry lab would fund, a PhD is one of the only ways to do it". — Neel Nanda, getting started in mechanistic interpretability (neelnanda.io).