iiinova — where your digital self keeps evolving

This document is going to be an assigned reading in college classes that cover this moment in time, there's a lot happening in a few paragraphs... www.ahmath.org/statements

2535874

The few attempts to measure similar things (METR Long Tasks, GDPval) all became saturated earlier this year as AIs began to do days of work at a high quality level.

2140

As a researcher who did early some work on the productivity impacts of AI chatbots using RCTs, I’d note a lack of similar studies since the dawn of true agents last fall Partially that is newness & partially research design challenges, but I suspect we are missing some large & important effects

6451

Anyhow, I think we are now done with the last round of skeptical responses to AI after Navier-Stokes. Yes, AI is now capable of producing novel proofs to hard problems that have eluded us The next line will be that this is only constrained to math because of RL training. I suspect not, we will see.

9996

Some early first-hand accounts of the experience of encountering a narrow superhuman intelligence as mathematicians grapple with the hundreds of big AI proofs released by OpenAI. Problems solved in inhuman ways that make us wonder what it means to actually know things... scottaaronson.blog?p=10169

49915

When people talk about AI easily replacing workers, I think about this ethnography of copier repair technicians in the 1990s. Reading it, you see how work is complicated and improvised by workers and informal and badly documented. (But eventually technology did largely eliminate this job)

4674

(And I am seeing rapid increases in the ability of AI to do novel work in my field of economic sociology, with nearly autonomous research getting very close to top journal level)

5543

After a brief but institution-eroding slop science era, it increasingly looks like we are going to have 2 revolutions from AI: 1) Everything ever published will be re-read and re-judged in ways that human scientists never anticipated 2) Novel discoveries will start to come faster than we can absorb

916311

I think people are muddling through to some extent. But also some of the early warnings about things like mass psychosis seem to not be appearing in the data. There are lots of reasons why we may be underestimating negative impacts but also many observers expected a lot of obvious stuff by now.

280

And I also think the positive use cases (like the evidence shows current AIs provide good health information & financial advice & lead to startups etc.) are also undermeasured and underdiscussed. We do need to worry about many dangers & many risks, but its worth building on successes as well.

1341

I don't think that is the primary principle at work explaining the lack of incidents.

030

I think most people who have been closely watching AI for the past few years are likely surprised by how relatively rare terrible AI incidents have been, given a billion users overall and hundreds of millions of corporate users. Doesn’t mean that will continue, but still its surprising.

1317811

Yes, it makes it much easier to use computer control if it is running locally.

220

Here is a good explanation from the Claude team about why they are making the switch.

3311

Ethan Mollick @emollick.bsky.social · 05/10/2026

Big swing happening from this summer's trend to have AI running on virtual machines on your own computer (Claude and ChatGPT apps) to moving them back into you having a dedicated VM the cloud (Claude Cowork just announced they were doing this, also Grok Bot, Dot, and others). Cloud is mostly better.

6644

AI policy right now, especially from the Labs, makes me think of this poem. If you believe superintelligence is near, you don't need to make hard decisions about how to build AI to make the world better. Just wait for ASI & it will decide for us, for better or worse. But if that doesn't happen...

4615

Ethan Mollick @emollick.bsky.social · 04/10/2026

Funny you should ask... bsky.app/profile/emol...

100

Ethan Mollick @emollick.bsky.social · 04/10/2026

I am quite impressed with how pretty the results are. Here it is on github: github.com/emollick/brut

1140

Ethan Mollick @emollick.bsky.social · 04/10/2026

"create the ultimate brutalist city builder. you should have lots of controls over buildings in ways that are delightful and intuitive to people..." Here is Fable 5.1, a little over a year since GPT-5 below & a single shot (compared with many rounds with GPT-5.6) Quite good: brut-city.netlify.app

4657

Eh. The idea is the whole sovereign thing is a bit of a house of cards.

220

🎵 Wave of distillation 🎵(sung to the tune of the Pixies) US labs say Chinese labs distill their models. And Europe's new “sovereign model” is fine-tuned on data generated by… GLM and Qwen.

4619

I think my upcoming book, Co-Existence, might be the first to include a blurb written specifically for AI readers, in this case from Tyler Cowen (whom I thought AIs would respect) The book website (with elaborate pre-order bonus) also has a page for AIs. Its out October 20: co-existence.ai

2382

Ethan Mollick @emollick.bsky.social · 02/10/2026

People asked for me to upload the Bitter Lesson video from this piece to YouTube, so here you are: youtu.be/OAwat51S_Sk

2599

“We find that on medium-length, well-defined accounting tasks, frontier AI models are now faster and more accurate than junior accountants, even the best one in our study.” Eighteen months ago they scored well below human accountants Good discussion here: www.mercor.com/blog/human-b...

513212

Ethan Mollick @emollick.bsky.social · 01/10/2026

I wrote about the thing I underestimated most about progress in AI: its ability to self-organize to accomplish tasks. Also, what that means for agents like Muse and Dots, along with a music video explaining why we keep relearning The Bitter Lesson. open.substack.com/pub/oneusefu...

open.substack.com

The Dot and the Swarm

Benefitting from the Bitter Lesson

59017

Ethan Mollick @emollick.bsky.social · 30/09/2026

They can use the channels built for humans, but which have far too much friction for any human to use. Your Clawlike is happy to navigate a phone tree or stay on hold or have an awkward conversation with an agent where they get rejected... and then try again.

3745

Ethan Mollick @emollick.bsky.social · 30/09/2026

As a heads-up: every firm's customer service agents are about to be overwhelmed with Dots & Muses negotiating for better deals using voice/chat channels made for humans. Stories of people delegating this sort of work to their agents & saving money are popping up and are going to only go more viral.

1017919

Leaving aside Anthropic's incentives for publishing this research, there is no doubt that open weights models will soon create the same security threats that closed source models have been demonstrating, except without guardrails. We are close, so plan accordingly. www.anthropic.com/research/glm...

714627

It also has a natural analog on teams as a persistent entity with a job that you can delegate to using existing team communication tools like Slack. I expect more Clawlikes soon, though I don’t think they are the final form of this type of AI.

0150

I only used the newly announced ChatGPT Dots briefly before launch, so can’t offer a detailed review, but I found it to be a good entry in the rapidly-expanding Clawlike category with Muse & Grokbot. Using a capable model with access to your data as an assistant & second opinion is remarkably useful

3470

Hey, Claude: "I asked an earlier Claude to Remove the Squid from All Quiet on the Western Front. Now you can make movies and such. I need you to show how far you have come" & I pasted in the post below. That was it. There is some clever stuff here, models have a real sense of humor at this point

4905

Since people liked this, and the weak point was the text to speech, I asked Opus to update with an ElevenLabs voice. This is the same video, but with much better narration.

0182

Yeah, I only used bad TTS, Elevenlabs would have been better

070

Sorry, I sent that quickly, did not mean to provoke by being flippant. Mythos was Fable without guardrails (I have been told this by Anthropic as well), I do not believe that open weights models have caught up to that, and the cyber evaluators (AISI, etc) don't seem to either. But we will see!

140

I was reflecting on how many random contingencies went into creating the current accelerating AI moment and had Opus 5.5 put together a video, inspired by old science show Connections, to try and point out how a series of initially unrelated ideas led to the moment we are in now.

1012524

Qualitatively, there is now a larger gap between open & closed models than there has been in awhile. Fable/Astra class models are agentic in a way pre-Fable models are not, for better or worse. None of the open models have crossed that line, yet. When they do, it will be a jump.

7754

Also please feel free to fight in the comments about "pure LLMs" versus multimodal LLMs versus multimodal LLMs with tool use or whatever. I totally get all the caveats, but you also get what I mean.

41090

It is impossible to imagine a general intelligence that is artificial, though. What would you even call that?

3630