Ethan Mollick
@emollick
Professor @Wharton studying AI.
New book, Co-Existence, coming October 20. Preorder here:
Substack:
591 Following 374.3K Followers
I think most people, when talking about open weights models, don’t deeply believe in the vision of AGI/ASI that lab insiders believe in. Like they don’t expect AI to really present grave semi-autonomous biosecurity or other similar risks. Whether it is right or not, no one knows,
Show more
Agreed: GPT-5.6 Pro (which is only available via the chatbot) is the smartest model out there. GPT-5.6 Ultra (the best model in Codex) is not near as powerful on hard tasks.
For really hard problems: GPT-5.6 Sol Pro>Fable 5 Ultracode > GPT-5.6 Sol Ultra. (Yes, this is confusing)
Show more
I still believe that everyone is too fixated on the state of play in AI right now (which companies in which countries are winning, how to manage costs, etc.) and not focused enough on the continued steepness of the capability curve for AI
At high capabilities, a lot changes fast
Show more
A lot of swift conclusions are being drawn about Kimi K3 based on fairly saturated benchmarks and ELOs, rather than actually testing it on very hard problems. The AI frontier has already moved so far that a good model that is a still months behind looks like the future to many.
Show more
I kept asking Claude Fable to make the game "more AAA" over and over again. The results are... interesting.
In Claude's view, this meant upgrading graphics, boss fights, mechanics adding custom sounds and soundtracks until it reached the limits of WebGL.
Show more
Fable: "you have Unity and access to MCP. I want you to build a game that is a unique twist on a FPS. You want the player to say "wow" & "so clever" and to enjoy the core gameplay loop"
WebGL:
* It had no assets so the graphics are procedurally generated
Show more
Happy 4th of July everyone.
I came here as a kid at 9 years old, not knowing a single word of english, lived on food stamps in section 8 housing. My parents cleaned houses while my sister and I collected cans for recycling. Our clothing was all donated and our furniture was things we found on the curb (here I am in front of our "tv set", the broken one being used as a stand for the little working b/w one).
But this country gave us an opportunity--to learn, to work hard and earn something for it. "Grateful" does not begin to cover it. I owe this country and its people everything I have.
Thank you America, happy 250 to the idea.
Show more
This is true… but maybe less important than the fact that people don’t try ambitious things with these systems. Many models are excellent as a Google replacement, for homework “help,” etc. It is someone’s agentic use of frontier AI on long-horizon real problems that is impactful
Show more
We are now in a position where a tiny proportion of the population uses Fable or soon GPT-5.6, while everyone else's experience of AI is 8-30b-model level - Google's AI Overviews, Meta AI, ChatGPT free tier, maybe MS Copilot at best. People outside of tech must be completely baffled how this is supposed to take their job, and annoyed that hundreds of billions are being poured into it.
Show more
While it is obviously true that not having verifiable domains makes training models in those spaces difficult... it is also true that models are also getting much better at non-verifiable domains. The frontier is jagged, but less so than I'd have expected from verifiability alone
Show more
The talk about Mythos and cybersecurity was not, in fact, hype. (As anyone using Fable to do autonomous work has probably recognized)
AI appears to be finding software vulnerabilities at scale.
In June 2026, 21 notable organizations disclosed ~1,500 high- and critical-severity CVEs, over 3.5× the previous monthly record set before Claude Mythos Preview's release.
Show more
AI implementation advice on my X feed is divided between those who "feel the exponential" and those whose (unconscious?) mental model of AI is that this is about as good as it is going to get, so it is time to build around the limitations & cost structures of today's capabilities
Show more
Fable: "you have Unity and access to MCP. I want you to build a game that is a unique twist on a FPS. You want the player to say "wow" & "so clever" and to enjoy the core gameplay loop"
WebGL:
* It had no assets so the graphics are procedurally generated
Show more
Continual learning is probably the biggest barrier to explosive AI adoption (& may have big implications for recursive self-improvement as well)
As long as you deal with amnesiac models that require humans to do the learning for them, adoption will be gated by human processes.
Show more
Introducing EBR-bench, our new benchmark to measure on-the-fly learning.
AI repeatedly plays a challenging board game called Earthborne Rangers and tries to learn from its mistakes. So far: no signs of improvement.
Show more
Fable in Claude Code is capable of really amazing things, including for non-coders, but the interface is not really designed for managing 5+ hour long autonomous tasks. Really hard to observe what is happening and intervene in real time, you often have to wait until the outputs.
Show more
I took the new AA-Briefcase scores from @ArtificialAnlys (basically having the AI do multi-week consulting gigs with a lot of complexity) and graphed the frontier curve for open and closed models:
1) Surprise, rapid gains!
2) The open weights gap is clear
Show more
The ability of Codex (and Code) to solve problems on my various Windows machines has saved me so much effort. Just one of the most annoying uses of time, and an example of a clear small win.
Show more
A real problem with feeling the acceleration viscerally is that current models are really good and it is hard to feel the vibe difference on most individual tasks with new models, even as AIs continue to increase in ability by large amounts (which they actually are doing).
Show more
I had Opus 4.8 in Claude Code write a sophisticated, if minor, academic paper from a archive of hundreds of de-identified research files from years ago
I had to use GPT-5.5 Pro as a reviewer, it spotted one major error & some minor points. Opus corrected
Show more
It is cliché at this point, but most people don't realize how capable the current generation of AI systems in their harnesses really are
(And, as opposed to previous times where non-lawyers or non-mathematicians were making these comments about law & math, now it is the experts)
Show more
I recently put together a 50-state legal research workflow in Codex. This is the kind of work that a team of associates used to do in a week, at a cost of ~$150K-$300K. I can now have research of similar quality done in Codex in 2 hours for a fairly minimal cost (if paid via API).
Show more
As more people come to recognize the tells of AI, which mostly happens as you start to work with AI a lot, the scales are going to fall from their eyes and they are going to realize what some of us already see: how much of this site (and blog posts, articles, papers) are AI now.
Show more
