Mark Chen

Things Mark Says on Podcasts

Where to Find Them

Mark Chen writes a16z Podcast , writes The Show About Science and writes SemiAnalysis . They have also been a guest on The Data Exchange with Ben Lorica (2 times) , TBPN (2 times) , OpenAI Podcast , Big Technology and Core Memory .

Recently: “Can AI Learn to Improve Itself? | Recursive Self-Improvement” on a16z Podcast (November 2026); “Beijing Will Not Pace the Frontier: China’s Speed-First AI Safety Regime” on SemiAnalysis (October 2026); “OpenAI's Research Chief On The Soup Wars, Poker And The Next Models - EP 46 Mark Chen” on Core Memory (December 2025); “From Vibe Coding to Vibe Researching: OpenAI’s Mark Chen and Jakub Pachocki” on a16z Podcast (September 2025); “From Vibe Coding to Vibe Researching: OpenAI’s Mark Chen and Jakub Pachocki” on a16z Podcast (September 2025); “OpenAI Day: GPT-5 Unveiled | Mark Chen, Greg Brockman, Sarah Friar, Max Schwarzer, Brad Lightcap & More” on TBPN (August 2025).

What They Said

“There's the boring run-of-the-mill execution, where you already know the parameters and you just carry out an experiment. And then there's taste, which is how you come up with the idea in the first place. The models are still not phenomenally good at taste.” — Mark Chen, a16z Podcast

On what still stops AI from doing research on its own, Chen separated routine execution from research taste, and said taste is where models lag.

a16z Podcast · 2026-11-17 Permalink → Listen →
a16z Podcast Around 08:30 into the episode
Speaker 2

…Mark spectrum, is sometimes referred to as full RSI. The question on everyone's mind, how do we get to full RSI? I spoke to Mark and John about how we might overcome some of the biggest barriers, including teaching models some of the more intangible qualities that exist in researchers, like taste.

Speaker 1

Research taste is about knowing which ideas are important in the long run, which ideas really catch on and end up going beyond whatever paper they first appear in, but being general and being composable with other ideas.

Speaker 4

There is kind of the boring run-of-the-mill execution. You know, you have an experiment, you already know what the parameters are, and you just go and carry it out. And then there's taste, which is kind of how you come up with the idea in the first place. The models are still not phenomenally good at taste. They go very deep into things like hyperparameter optimization, essentially like tuning all the little knobs that you can tune.

Speaker 2

Even in that example that you shared, the points where you were kind of nudging, redirecting AI, right, that's still coming from you. And I guess my question is, where does your intuition come from? And at what point do you think a machine could replicate that?

Speaker 1

A lot of intuition just comes from my personal experience doing research. But to get good at a certain research field, you also have to read a lot of the past work in that field and you try to absorb the lessons from all this historical work. A lot of the things that I just described can also be…

Speaker 2

What John is getting at is essentially the difference between knowledge and experience.

Speaker names come from our own diarization. The position is estimated from where the line sits in the episode.
“When we talk about AGI, we think about it as a spectrum, not a moment in time, just like the Industrial Revolution wasn't a moment in time. I feel like we're maybe 80% of the way on the AGI spectrum. I think we're earlier on the RSI spectrum. If you define it as how effective the models are at discovering ideas, maybe we're somewhere between 25 and 50% there.” — Mark Chen, a16z Podcast

Asked how close we are to recursive self-improvement, OpenAI's chief research officer gave rough percentages for AGI and for AI that improves itself. The numbers are his own estimate, not a measurement.

a16z Podcast · 2026-11-17 Permalink → Listen →
a16z Podcast Around 07:26 into the episode
Speaker 2

John actually gave us a concrete example of this in his own workflow.

Speaker 1

…of the results and giving more detailed breakdown of how to carry out the project. We'll get a lot further because I'll build up my understanding as this process is going on and ask a lot of questions that the AI wouldn't have thought to ask and point it in directions that it wouldn't have gone in.

Speaker 4

When we talk about AGI, you know, the way we think about it is in terms of this spectrum, right? It's not a moment in time, just like the Industrial Revolution wasn't a moment in time. And when you ask me, I feel like we're maybe 80% of the way when I think about this AGI spectrum. I think we're earlier in the RSI spectrum. If you define it as how effective are the models at feeding back and discovering ideas, I would say maybe we're somewhere between, let's say, like 25 and 50% there. But I do think you can already see the noticeable markers on the downstreams.

Speaker 2

On November 5th, Meter published a report on the acceleration of AI research at Anthropic, estimating a roughly 2X speed up in the rate of progress in the last year. But this is still most likely due to human researchers using AI models to accelerate their research. The full automation of AI…

Speaker 1

Research taste is about knowing which ideas are important in the long run, which ideas really catch on and end up going beyond whatever paper they first appear in, but being general and being composable with other ideas.

Speaker 4

There is kind of the boring run-of-the-mill execution. You know, you have an experiment, you already know what the parameters are, and you just go and carry it out. And then there's taste, which is kind of how you come up with the idea in the first place. The models are still not phenomenally good…

Speaker names come from our own diarization. The position is estimated from where the line sits in the episode.

Collections They Appear In