Evolution of the Revolution | Transcription

Note: This transcription is split and grouped by topics. You can navigate through the Table of Contents on the left. It's interactive. All paragraphs are timed to the original video. Click on the time (e.g., 01:53) to jump to the specific portion of the video.

Why a new feed? (00:00)

Hello and welcome to the Cognitive Revolution. If you're hearing this, you've found your way to our new feed. So thank you for being here as we kick off year two of the show. Before returning to our regularly scheduled programming, I want to take just a minute to explain why we're launching a new feed and to share a bit about my high-level goals for this project.

As you know, I describe myself as an AI scout, and my mission, across everything that I'm doing, is to understand AI developments as well as I possibly can, so as to inform public discourse and hopefully contribute to better decision-making. This means analyzing AI at every level of abstraction, from high-level model capabilities and behavior, to market dynamics and societal impact, all the way down to mechanistic interpretability, datasets, and hardware. I also try to take every possible angle, from academic and industry researchers, to application developers, to power users, to policymakers.

This mission predates the podcast, which I originally thought of as a fun way to document and share some of what I'm learning. Taking inspiration from Tyler Cowen, I set out a year ago simply to have the AI conversation that I want to have, with really zero expectations that anyone would want to listen to such deeply in-the-weeds content. I certainly did not expect to make 100 episodes in the first year. But the Turpentine team made it so easy, and there are just so many fascinating topics in AI to explore right now, that I caught on a bit of a roll, and so here we are. I definitely want to thank the entire Turpentine team for setting me up for success. There is no way I would have been so productive without their support.

Today, making the podcast has become one of the few main ways that I'm spending my time. And while I never have been and still do not intend to become a full-time content creator, I do find this a natural and effective way to communicate. And along the way, I've realized that the Cognitive Revolution audience, while still fairly modest in number, is already extremely high value, such that if all I can do is help the current audience improve its understanding of AI developments as they happen, that in and of itself could be a meaningful contribution worthy of intense focus and energy, particularly since I really love doing it.

So, why change something that's working better than I expected? Above all, while this podcast is the closest thing I have to the platonic form of the AI-scouted work that I most want to be doing, I know from how unstrategically I started this project that it can't possibly be the best or final form. And I wanted to set up an independent platform to give myself a venue to experiment with new formats and different cadences that may be of even greater value to the audience in the future. In practice, for now, not much will change.

What to expect on the new feed? (02:40)

Over the next several weeks, I'll be posting original episodes to both the new Cognitive Revolution feed and the original feed, which we'll be rebranding as Turpentine AI. And I look forward to sharing that platform with others going forward. I've got a dozen shows already recorded and ready to go, so it's safe to say that I'll have plenty of interesting content for you on both feeds for the foreseeable future.

As for what to expect on the new Cognitive Revolution feed in particular, I hope to go even deeper into critical topics, ideally delivering even more alpha per episode and per minute. When I look back at the last year's worth of work, the correlation between effort and value is clear. And the Mamba episode from December stands out. I spent two full weeks reading in preparation for that one, and I edited it far more intensively as well. But overall, it seems to have been well worth it. I'd guess that episode probably strengthened our collective understanding of where AI is headed as much as 10 startup CEO interviews. Maybe even more. So I definitely want to do more of that sort of intense research and reporting where I can.

And at the same time, I'll try to cultivate and spotlight other AI scouting talent as well. As I've noted a few times, the AI exponential, as measured in terms of papers, products, and policy proposals, is getting beyond any one person's ability to keep up. I need to be more collaborative, if only to cover my blind spots better. And certainly, society in general needs more people going down any number of rabbit holes to understand developments as deeply as possible. I hope this new feed can help such people share their best work with a high-value audience without requiring them to spend a lot of time creating content or seeking attention.

My first attempt at this new format will be a deep dive into the flurry of Mamba-inspired state-space model research that's been published over just the last 90 days. That one's already recorded and will be published here soon.

Seeking collaboration (04:40)

If you or someone you know might be interested in working with me on another deep dive into a particular topic, I would love to connect. I'm specifically looking for a bio or biotech expert who can help explore the intersection of AI and biology. And I'd also love to work with someone who loves testing new AI products and would be interested in putting together a red team report on recently launched AI calling agents and other applications.

Seriously, if this is you or you have another idea for a deep dive, please DM me on Twitter or the social media platform of your choice.

Business model (05:05)

Business model wise, we will continue to have sponsors, but I will definitely not be chasing numbers. I've been extremely flattered and honored that philanthropists have expressed interest in supporting this work. But donor dollars are scarce. And I've also been pleased to learn that a medium-sized audience, if it's sufficiently high value, is economically viable in the AI space. So as long as there's no conflict between commercial sponsorship and editorial integrity, and thankfully to date that has not been an issue at all, I think the sponsorship model will continue to make sense.

Of course, I will be glad to have new listeners and continue to ask that you share the show with friends. But no matter how large the audience may grow, I will keep today's high-value audience in mind as the target profile. And if that limits growth in the future, that will be totally fine with me.

You might see a few ads online as we transition people over to the new feed. But if anything, the content is likely to become a bit more advanced and potentially less accessible. Our production quality will likely remain pretty basic. I will try to make sure that the audio is clear, and we will edit a bit more aggressively to maximize the value of your time spent listening, but I am really not in show business, and I don't want to get on a hedonic or budgetary treadmill by dramatically raising the bar on production.

Commitment to Deep Work and Serving the Public Interest (06:20)

I am definitely committed to staying in direct touch with emerging AI technology through a combination of hands-on building and testing. I want to spend more time doing deep work, and I believe this is truly critical to delivering value. If I ever start spending all day giving instant analysis on Twitter, I genuinely believe you should all stop listening to me.

For now, at least, I will continue to focus more on what is than what ought to be done about it. The smartest people I know have quite radical uncertainty about where AI is ultimately headed. So while my position may seem at times a bit both-sidesy, it is the result of sincerely ambivalent feelings.

In the long term, I will not shy away from advocacy, and I will not hesitate to criticize powerful people or companies if I feel it's appropriate. While I've experienced no editorial pressure to date whatsoever, I do expect that AI issues could become more politically contentious in the future. And part of my intention in setting up this independent feed is to ensure that I am always able to use whatever influence I have to serve the general public interest.

With that, I invite your feedback, suggestions, and requests. What has been the most valuable part of the Cognitive Revolution for you? What are you struggling to understand better? What do you feel are my blind spots, and what might I do to correct them?

Thank you again for listening, and stay tuned for lots more AI Scouting reports coming soon. For now, if you would take a moment to share the new feed, as always, we would very much appreciate it. And with that, I thank you all for being part of the Cognitive Revolution.