74 Comments

They are a bit full of themselves, aren't they? Claiming that "we, and only we, truly understand what's coming." Very insular of them. But also, they clearly don't really understand how the rest of the world works. Integration and deployment of new technology is always slower than it hypothetically could be. These Promethean myths of sudden transformation rarely happen. New infrastructure needs to be created, best practices need to be developed, oh, and businesses and governments needs to actually adopt the technology. It's sort of sadly ironic when the tech bros are so disconnected from the rest of humanity like this.

Expand full comment

Humility clearly isn't your strength either. What is it about the possibility of them being right that is so threatening to you that makes you feel the need to attack and degrade them for sharing their best understanding of what is going on with the world?

If things end up going the way they say and they hadn't shared it with the world then I have no doubt that you would attack them for their arrogance in keeping it to themselves and not providing society with the opportunity to weigh in or influence what happens. Am I right?

Expand full comment

Chris, you seem to believe that questioning tech leaders' prophecies is somehow an attack on them personally. But you're demanding a strange form of intellectual submission - that we must all nod along with whatever predictions emerge from Silicon Valley, no matter how grandiose. I reject that premise entirely. My expertise in studying technological diffusion and organizational change gives me both the right and responsibility to critically examine these claims. No one, not even brilliant technologists, deserves unquestioning acceptance of their vision for reshaping society. The fact that you interpret professional skepticism as a personal attack suggests you've confused critical analysis with disrespect. Technical genius doesn't grant immunity from reasoned critique, nor does it automatically confer expertise in economics, sociology, or human behavior. If you find that threatening, perhaps examine why you feel such a powerful need to defend billionaire tech leaders from scholarly criticism.

That you went straight for an ad hominem attack about my humility speaks volumes about the substance of your position. If the only flaw you can find in my reasoning is that I'm insufficiently deferential to Silicon Valley's self-appointed prophets, then you've inadvertently proven my point. Epistemic humility is a virtue in scientific discourse, but it's not a requirement for challenging delusional claims from powerful actors. In fact, history suggests that confronting grandiose predictions from unaccountable power brokers requires precisely the opposite - the willingness to stand firm against pressure to simply accept their vision of the future.

Expand full comment

In defense of Chris, there’s a difference between professional skepticism and saying they are full of themselves, insular, and don’t understand how the rest of the world works… those are indeed attacks, and your failure to see that indicates that you also may be a bit full of yourself, insular, and unaware of how the rest of the world works. Hence why Chris probably questioned your humility.

If you truly lack humility you may read this and think to yourself that other people simply lack the intelligence that you have and are misunderstanding you and there’s no need for you to course correct. Your long defensive response suggests you will.

Hopefully you read this and it gives you enough pause to go back and read your other statements and see that sometimes perceptions are mirror-like and the things we criticize in others are often the things we struggle with the most ourselves. I’m sure you’re already familiar with that, but your responses seemed a little out of touch.

Expand full comment

If you had information on a technology that was sure to change the world, and no one else had that information, then are you "full of yourself" in claiming you have special knowledge of how the future will go? I'd argue no.

Being full of yourself is having the same info as everyone else and saying "I alone understand where this goes"

Expand full comment

This presumes that they have exclusive information. They do not. The vast majority of AI information is public domain, and as Mira Murati said, the time between lab work and public disclosure is very short.

They are in an echo chamber, and very much full of themselves.

Expand full comment

There's is no certainty of neither your claim ("they do not have exclusive information"), or their claim ("we have exclusive information"). I know it's on brand for you to be tough, but I dislike the fake certainty. "There's a high likelihood they do not have exclusive information" seems like a more leveled and real version of your argument.

Expand full comment

well said.

don't forget altman's claim that poverty will end and the cost of intelligence and energy will fall to zero, lol.

Expand full comment

Altman is many things. Historian, economist, and anthropologist are not among them.

Expand full comment

Insightful as always, thanks Ethan. Blown away by the 🦦 videos! Thanks also for endnote caveats on o3's benchmark achievements - very helpful.

Expand full comment

Your statement, "Again, very little (human) work was involved." I think it understates the work of the human being. What a human mind has to do to understand using AI, creating a prompt and validating the output is the result of years of education and experience and "work".

Expand full comment

My sense is the openAI o3 announcement was a bid by Sam Altman for openAI to stay relevant, or rather to stay first tier, as it was looking like google might pull ahead and turn openAI into something like a new Netscape. And that bid turned out to be fabulously successful.

o3 is of course interesting, but it is at best a proof of concept. I know costs come down, but there is not a single year of regular scaling to get individual queries down from $10,000 a go to $1 a go. With even $1 meaning monthly subscription costs in the hundreds of dollars rather than the $20 right now for regular ChatGPT.

I suspect o3 will be able to support agentic work, but again, agent work requires many many queries hidden from the end user to accomplish tasks. So multiply thousands of dollars per query by many many queries to get you “book me a holiday in Florida next month”. I suppose I’m thinking of things from the consumer point of view or the SME point of view, whereas even expensive systems can have specialized uses that are very impactful (cutting edge scientific research, financial analysis, weapon design). But then we are talking 1960s or 1970s computer use as an analogue. Whereas everyone thinks we are soon to be deep in the PC era in that analogy.

Expand full comment

Yes, isn't it interesting how GPT 5.0 is no longer mentioned. That was supposed to be the release everyone was waiting for. Then they went to o1, because they must have hit a wall with what they hope 5.0 would be. Now there are iterating on the o series.

Expand full comment

Keep in mind that it sounds like o3 is the same as o1, except that they spent even more time having it fine-tuned w/ reinforcement learning on helpful chains of thought. But both are very likely fine-tuned versions of GPT-4o. They've said that o3 costs the same to run as o1 does. They just really upped the max thinking time to make progress on the ARC benchmark, which is something it has extra trouble with, perhaps partly due to how it is trying to understand visual grid logic problems via arrays of text.

The vast majority of stuff isn't going to cost anywhere near $10k per query. Some problems may actually be cheaper with o3 than o1, since it may have been exposed to more kinds of chains of thought, meaning it doesn't need to experiment as much with those problem classes to get the right answer.

Expand full comment

A critical aspect of writing a report is the weeding-out of content that is superfluous, non-additive, or wrong-in-context ... at base, akin to Dr. Seuss's "calculatus eliminatus." In my experience, significant personal learnings come from exposure to such content, including decisions driving the framing of an argument. I wonder how much the skill and desire to wander among (and ponder) the source material will be lost.

Expand full comment

Agree - the act of research is one of exploration, making connections, deciding what to take or leave, reframing your perspective. It's not that AI couldn't necessary do (or appear to do) these things - but what is the point if a human researcher no longer engages in this intellectual exercise?

In my experience AI has helped me synthesize information and make interesting connections more quickly, but I am still on the hook for deciding how to apply and use that information in a meaningful way. Which often requires slow human thinking and discussions with other humans.

Expand full comment

As long as we are allowed to function and thrive after disagreeing with AI we should be alright.

More concerning is whether people will maintain sufficient awareness to know when it's necessary to disagree with it.

Expand full comment

I love the thoughtfulness around narrow agents, and think that is the mental leap most of us can start to take beyond "simple Q&A with a bot" where most of the initial usage tends to go. I like to think of agents as a "new in career employee" sitting next to me that I can kick off on a project to get something done, and narrow agents folds into that model.

As an aside, your video has a proposed answer to "How would an otter carry their things?" - I had always thought it would be tight on their back, but your video seems to imply that a loose backpack slung off to the side. I'd never thought of a four legged businessanimal being a "single shoulder carry"...now I need to reconsider many of my assumptions :-)

Expand full comment

I think an insight into the direction of travel for such technology can be seen in the first otter video at 00:22, when the otter carries out some gratuitous product placement for the soft drink "Dirkidua" (?!)

Expand full comment

So I guess we're moving from whether we're "bringing the AI to the table" (as you say in your book) to the point where the AI agents will decide whether they need to bring the human to the table. There are many exciting possibilities in there. That otter video is very impressive, by the way.

P.S. I was half-expecting you'd mention the delays with Grok 3 and Claude 3.5 Opus. I can't wait for the new version of Opus! It will probably have lots of AGI-ish qualities to it.

Expand full comment

why talk about those future hypothetical reasoning models when we already have google gemini?

https://lmarena.ai/?leaderboard

Expand full comment

I see your point, and have loved Gemini 2.0 Experimental Advanced 1206 since they released it (I use it all the time!)… but Claude is still the best writer around in my opinion (the folks at the writing app Lex recommend Claude 3.5 Sonnet for writing, and I agree). Claude 3.5 Sonnet is surprisingly low on that leaderboard, but I'm thinking it's because of reasoning and math tasks (4o is so much of a worse writer at the moment than 3.5 Sonnet). I expect wonderful skills from 3.5 Opus.

Expand full comment

Thanks for sharing your thoughts there. Before this recent Gemini model I was using Claude 3.5 as my main model so I concur that it is amazing and even 3.5 haiku is pretty good as well.

Just like Claude I do have to fine-tune the latest Gemini model a lot to really make it a useful AI advisor for me that doesn't just do all of my work and makes me think so there's room for improvement on both sides I believe.

I wonder if this Gemini resoning model release as well as the '03 model announcement will speed up the schedule for 3.5 opus.

Expand full comment

I hope so! 3.5 Opus was expected for "late 2024."

Gemini 2.0 Advanced really blows my mind. Prompt adherence, depth, quality of advice, help with all sorts of writing tasks. Its writing does sometimes lack the amazing qualities (the "make it fresh" quality that is an old adage in writing) often found in 3.5's writing. And 3.5 is very good as a writing coach.

Looking forward to all these new models—including Gemini 2.0 Advanced once it includes those multimodal qualities from the demos.

Expand full comment

The Otter exercise comparisons are fascinating. Your article does a good job of explaining the tests and the context. I'm struggling to find who is doing that envisioning beyond futurists and artists. Must we sit in reactivity or have our heads in the sand? We need more articles exploring ideas and fewer stalemate debates. Thanks for writing and sharing!

Expand full comment

The Deep Research feature of Gemini is terrifying for those of us who assign High School level research papers. In under 3 minutes, students can have a plausible rough draft with citations. This one has flown under my radar. And if the trajectory continues, what will these models be capable of in another 18-24 months?

Expand full comment

I imagine in-class essays might be good to do more. Is this feasible at the HS level? (And keep the copies as writing samples!)

Expand full comment

The written “essay” as a concept and as the main vehicle of assessment needs to be deprioritised. Oracy, vivas, socratic in person dialogues, if equally valued could be transformative for society.

Expand full comment

Ethan, a helpful post. Could you point your readers to a few folks that you think are doing a decent job trying to articulate what a world awash in AI looks like?

Expand full comment

This is a great question. After consuming extensive AI-related content over the years, I believe no one explains this better than Professor Ethan Mollick. He rationally articulates the possibilities without falling into hype. It's challenging to define clearly since a transformation like this has never happened before. Like Mollick, I think the real discoveries will come from people on the front lines of various industries, using these tools and sharing their insights. As Mollick says, "HR is Now R&D." Together, we're likely to uncover and build—step by step—what a world immersed in AI truly looks like.

Expand full comment

"As AIs get smarter, they become more effective agents, another ill-defined term (see a pattern?) that generally means an AI given the ability to act autonomously towards achieving a set of goals."

Who will be liable for the Agent's actions?

Would an insurance company write liability insurance for a corporation who is using autonomous AI?

This could even have an impact on AI developers who are careless with their toys.

Expand full comment

The last two posts where you've shown what the ai video creations can be have been absolutely jaw-dropping! I'd be terrified as an artist looking at those

Expand full comment

I’m immediately reminded of Miyazaki‘s remorse and disgust seeing AI-generated animation for the first time. When a grotesque video is presented to him by animators at Studio Ghibli; he states, “I feel like we are nearing the end of times. We humans are losing faith in ourselves…”

Expand full comment

Thank you for the thought-provoking article, the flood of AI is imminent. These advancements in AI will dramatically change industries, but societal adaptation will take time.

https://www.univ-msila.dz/site/gtu-ar/

Expand full comment

Reading your excellent and informative posts this morning, and watching the simultaneous destruction from the LA fires, illustrates the disconect between the remarkable potential and advances of AI and the real-world consequences of organisational and political ineptitud. Professor Jeremy N White. Bayes Business School. London

Expand full comment