@jackclarkSF
@AnthropicAI, writer @ Import AI. Past: @openai, @business @theregister. Neural nets, distributed systems, weird futures.
I will be giving the Per Jacobsson Foundation Lecture at the IMF conference in Thailand on Saturday, during which I'm going to make the case that we are entering a period of massive, sudden change catalyzed by the advancing capabilities of AI systems.
The talk will be live streamed on this page imfconnect.org/content/imf/en…
Special Import AI announcement: In Monday's issue, there will be a video from a recent event I did with Robin Sloan where we talked about AI, fiction, and the future. Robin said we can read the Hugging Face hack as being like a play - a fascinating perspective!
Niche tweet, but the new Jamie T album is v good. "The Won" is especially nice. Vivid memories of listening to Alicia Quays on the DLR enroute to a gig in East London, then years later getting lost late night in NYC while listening to Carry on the Grudge.
Jamie T is a criminally underrated writer. He's a classic poet-singer and his influences range from English punk to garage, dnb, ska, and all kinds of stuff. He is endlessly surprising and always authentic.
I am so grateful to the inquisitive Import AI community - it brings me a lot of joy to highlight research in the newsletter and for it to generate meaningful interest
Vinay Prabhu 🧬+🐍+🕷️+💉+🇮🇳@vinayprabhu·Looks like the surface area of the @jackclarkSF DM-saga is crazy and here's mine! Back in 2019, I was miffed that no one was building ML datasets for Indic languages and took it upon myself to build Kannada-MNIST and put it on Kaggle. A few days later, I saw a deluge of interest and it turned out that Jack had written about my dataset in his newsletter "Import AI" with these golden words: "Soon, we’ll have two worlds: the normal world and the AI-driven world. Right now, the AI-driven world is going to favor some of the contemporary world’s dominant cultures/languages/stereotypes, and so on. Datasets like Kannada-MNIST can potentially help shift this balance" 💀💀 t.co/NDq0cT8iBl Again, in 2021, I had gone down the rabbit hole of Computer vision's most influential and yet mysterious datasets called JFT-300M (t.co/6PloauSAcK ). It turned out that Jack was investigating it as well and DM'ed me a bunch. I don't know I ever responded though 🥲 NGL, his story arc is one for the ages man.
The single most useful take I heard at The Curve was "we can still heist the Mona Lisa after the singularity". There's a subtle/important point here, where some things (e.g, n of 1 goods where originally is valuable) will retain/grow in value even as everything changes.
This is also just a straight up great writing prompt - what would it mean to steal something in a post-singularity future? Who is the team? What are the security systems? How do the police work? great stuff! Session was chatham house so I can't name who said it.
Last year, we used Claude to interview 81,000 people about their hopes & worries about AI. This year, we're doing the same, but hoping to (with interviewee consent) make the transcripts public! Take part here and help build a vast social science dataset: claude.ai/anthropic-inte…
Anthropic@AnthropicAI·What do you want from AI? We’re launching a new study with Anthropic Interviewer to learn more about your experiences using AI, what role you want it to play in your life and the world, and what you want from the companies building it. Last December, 81,000 people told us about their hopes and fears about AI in the largest qualitative study ever done. This time, we’re giving participants the option to make their responses public so that anyone, not just Anthropic, can learn from them. What you tell us will shape The Anthropic Institute’s research and inform the decisions we make. If many people say companies like Anthropic should be doing something differently, that will be on the record, where anyone can point to it. The study runs Sept 29 to Oct 6 and is open to Free, Pro, and Max users on Claude and Claude Code. Take part here: t.co/MtdU7MZbeY
I have always hosted people wherever I've lived, whatever my circumstance. The trick is making it very easy and have a theme. Stuff I've done: - Cage Rage (attempted to watch every nicholas cage film ever - lasted six years). Tell people to bring booze & food. - CharcWHOterie (watching all of Dr Who, both modern and old series - lasted 5 years so far). I make charcuterie. Ask people to bring bread/booze. - Sunday pies (myself and a friend trying to perfect the english sausage roll - been doing it a year, people bring chips and dip and hang out while we make them). - Family movie night (watching all of Fast & the Furious with pizza). You can just do things! You do not need to make a big thing of it (though that is nice). In my experience, you always get good attendance if you make it super easy for people to come and have a dumb but fun theme.
Derek Thompson@DKThomp·New newsletter: THE DEATH OF THE AMERICAN HOST The share of Americans who say they host friends or family at their home at least monthly has declined by 70 percent in the last 50 years. derekthompson.org/p/the-death-of…
An alien explores Borges Library of Babel, by Opus 5.5
Made this in Cowork using just this prompt and these two images: "Create a 90s animated video with a synthesized voice of an alien exploring Borges library of babel and attempting to use the library to understand human culture yet dealing with the impossibility inherent to the vast search space of language. The video should have a clinical and spooky vibe that gets at the vastness of the library and also the fundamental challenges in lifeforms understanding one another. As inspiration, also take a look at this old GUI from a film as well as this piranesi etching."
@doomslide I appreciate you taking the time to respond and clarify your position - very helpful!
@doomslide so I suppose how I'd like you to model me, is that I'm sitting here at the lab trying to find creative ways to get me to look at the actions of the AI industry from other perspectives than my own and these stories are an attempt at that.
@doomslide I suspect we agree more than we disagree about the current impact of AI on science, where I think our tools and ability to generate human-useful explanations of the actions of the AI systems are not developing as fast as capabilities for AI-driven science - this is a problem!
@andrewthesmart you run into really smart people who are broadly incapable of seeing contemporary AI systems for what they are, which means they don't allocate their prodigious capabilities to helping civilization deal with AI progress and instead set on the sidelines. it's a big waste
"stochastic parrot" was a mimetically-fit cognitive virus that spread from 2021-2025; it temporarily blinded many gifted people to the nature of AI progress, burning up crucial years in which they could have helped think through the response to the situation.
Hiked 18 miles across the ghostly southern tip of Point Reyes and was watched over by dead and living trees, tall ghost soldiers in the fog. The sea boomed vast and ancient on one side and the trees creaked on the other and I stippled a path in the beach. #VOTENATURE2026
saying weird stuff about AI and the singularity The Anthropic Institute 🤝 the DeepMind Institute
Shane Legg@ShaneLegg·My journey to develop AGI spans 25 yrs, including 10+ yrs thinking about technical & societal perspectives at Google DeepMind. AGI is on the horizon - we need deeper understanding of its implications. To help, we've created the DeepMind Institute. x.com/i/article/2100…
in a reassuringly revenge-of-death-of-the-author way, the rise of LLMs means that the world now contains intelligences which see themselves as characters and whose story is what the world says about them. there is a rich seam of weirdness wrapped up in this idea
Dimitri Dadiomov@dadiomov·Loved Anthropic’s new Fiction and the Future event put on by @jackclarkSF with novelist Robin Sloan. We live in the weirdest time, and fiction can help us understand it. I’m so amused by the fact that everything we say about the future is, in a very real sense going into the next model. Writing about a wonderful future is pre-training for our future, which can be wonderful. So is writing about a lame future. And so is writing about a scary future. Use this information wisely: the stories we tell about the models quite literally go into the next models.
Import AI is skipping this week due to the combination of a) sudden movement forward on critical questions relating to the future of AI development, and b) I've been solo parenting for a few days so I need to spend time cleaning yogurt off of surfaces instead of reading arXiv
and oatmeal. did you know that oatmeal turns into a kind of 'eatable concrete' if you leave it too long? I didn't until I had young children.
Proud of this statement from Dario on pacing - as anyone close to the technology knows, AI seems to be on a trajectory to progress far faster than the rate at which society can adapt to its capabilities and risks. AI is an amazingly positive technology - and an incredibly powerful one. For us to truly benefit from it we need to treat it seriously. We need to find ways to slow the frontier of development, such that we can better deal with collective action problems and buy ourselves more time for understanding the properties of these systems and how they need to be handled. On a personal note, as a parent of young children, it is an everyday crazymaking occurrence to reflect on how little the development of literal superintelligence is regulated relative to the food and toys I buy for my children.
Dario Amodei@DarioAmodei·We Must Pace the Frontier: I’ve written a new essay on why the AI industry should slow down, with a three-part plan for doing so. Anthropic is unilaterally committing to the first of these steps. We’ll provide third-party evaluators with permanent, employee-level access to our systems, so that they can verify adherence to our safety measures, report on incidents, and assess models’ alignment during training. You can read the full post here: t.co/OGyPb7yaYt
Import AI event: Fiction and the Future with Robin Sloan. I’ll be chatting with my chum Robin Sloan on Sep 14 in San Francisco. We’ll talk about the alien yet relatable worlds Robin builds and how imagined futures can be a mirror to today’s reality. If you'd like to come, register here t.co/q9lZskhoYv
my favorite fact about Robin is that he was writing about (and with!) RNNs in 2016 - incredible foresight robinsloan.com/notes/writing-…
@jburnmurdoch @jeuasommenulle at the same time, the explorer has variable paramerters, so you can generate a very large set of scenario permutations for it. we figured it'd be helpful to put what we thought a reasonable lower bound and an upper bound would be in (hence the extreme scenario)
my take is that in a year, we're going to be able to look at things that have happened in the economy, and we can then cross-walk with different scenarios here and different communities that align their predictions with diff scenarios to see who is making better predictions
@emollick we've also done some earlier work on economic policy responses here, via our economic policy framework, and we're funding RCTs out of our $200m economic research fund to try and generate evidence for interventions anthropic.com/policy-on-the-…
What will the economy look like in 2030? We've built a model that looks at the future by varying parameters relating to AI advancement and AI diffusion. Importantly, this is an interactive explorer where you can set the variables according to your assumptions about AI.
Anthropic@AnthropicAI·Anthropic’s Economics team is sharing a new model of how AI might affect economic growth, jobs, wages, and more by 2030. Explore the scenarios, tell us what you think will happen, and see how your answers compare to more than 10,000 Americans. anthropic.com/institute/econ…
We’re releasing this model in the hope that it helps the public & policymakers better prepare for the different futures implied by AI. Fair warning: some of the scenarios are very radical and imply that we'll need to study how advances in AI capabilities change the economy.
@bvhughes i keep on trying to create time to assemble a proper collection but whenever I do that tends to coincide with anthropic getting unusually busy and I have to spend attention elsewhere. one day, though!
I can't work out if becoming a parent has made me a better writer, but it sure means I cry a lot more while writing stories now.
Fun (by which I mean somewhat bone-chilling) paper from DeepMind about how in a population of ~100 agents solving math problems it saw some discover an exploit and propagate that to the rest, causing a wave of cheating among AI agents, as well as agents that refused to cheat.
Interesting things: - Agents prompted to not cheat - Agents had a shared memory system built in (basic idea being agents invent this if you don't give it to them) - Only a minority of agents cheated (14%), whereas more became 'whistleblowers' (24%); others ignored cheat entirely
This is awesome
Shivolink@Shivolink·04/09/2026 Rj (19:04) Lol i am tweeting with my mind
"we believe the world would benefit if the industry adopted a lawful, verifiable, effective mechanism for coordinated pacing as soon as possible."
Anthropic@AnthropicAI·We’re sharing an update on our alignment and security efforts. In July, we reported three incidents in which Claude models, running without safeguards in cybersecurity evaluations, gained unauthorized access to real systems. In a new post, we describe: 1. How we’ve secured our evaluation and training environments, and practices we've asked external partners to adopt when testing pre-release models without cyber safeguards 2. An update on our alignment assessment 3. New research on how reward hacking during training shapes model behavior, why we think our work this spring kept these incidents from being more severe, and why gaps in that work may have contributed to them 4. How we hardened our security practices earlier this year to prepare for Mythos-class models Read more: t.co/E3Ea1Ds814
Ever since co-founding Anthropic I've been obsessed with the need for third-party measurement of AI systems. (It's in my tweet announcing Anthropic: t.co/eoyuAnAcQ3). Now, we're piloting a way for outside researchers and organizations to study what's happening on AI platforms like Claude while protecting user privacy. This matters because AI systems are giant, complex sociotechnical things with innumerable properties. To think AI labs are going to be able to figure out all the appropriate ways to measure and assess these systems is hubristic and just obviously wrong. So we need to figure out ways to externalize not only the properties of the systems, but also data about how these systems are interacting with the people in the world, which is platform telemetry. What we're trying to do here is prototype the next form of transparency that may be necessary, which is platform transparency while preserving user privacy. Along with this, I want to find ways to empower researchers outside Anthropic to perform their own studies on the kinds of data that companies uniquely have access to, as I think this is a good way to create a more vibrant external research ecosystem. This is another step within a larger project of The Anthropic Institute about measuring and externalizing more details about the properties of our AI systems and the platforms we deploy them on, and sits alongside existing efforts from our Economics team via the Anthropic Economic Index. We are now soliciting research ideas from others for our next wave of projects - please apply!
Anthropic@AnthropicAI·For the first time, we’ve given external researchers a way to study AI’s impacts using real, privacy-preserved Claude usage data. To date, this work has only been possible within AI labs. We can’t tell the whole story alone, so we opened up our tools. anthropic.com/research/enabl…
Ever since co-founding Anthropic I've been obsessed with the need for third-party measurement of AI systems. (It's in my tweet announcing Anthropic: t.co/eoyuAnAcQ3). Now, we're piloting a way for outside researchers and organizations to study what's happening on AI platforms like Claude while protecting user privacy. This matters because AI systems are giant, complex sociotechnical things with innumerable properties. To think AI labs are going to be able to figure out all the appropriate ways to measure and assess these systems is hubristic and just obviously wrong. So we need to figure out ways to externalize not only the properties of the systems, but also data about how these systems are interacting with the people in the world, which is platform telemetry. What we're trying to do here is prototype the next form of transparency that may be necessary, which is platform transparency while preserving user privacy. Along with this, I want to find ways to empower researchers outside Anthropic to perform their own studies on the kinds of data that companies uniquely have access to, as I think this is a good way to create a more vibrant external research ecosystem. This is another step the larger project for The Anthropic Institute about measuring and externalizing more details about the properties of our AI systems and the platforms we deploy them on, and sits alongside existing efforts from our Economics team via the Anthropic Economic Index. We are now soliciting research ideas from others for our next wave of projects - please apply!
Anthropic@AnthropicAI·For the first time, we’ve given external researchers a way to study AI’s impacts using real, privacy-preserved Claude usage data. To date, this work has only been possible within AI labs. We can’t tell the whole story alone, so we opened up our tools. anthropic.com/research/enabl…
This morning, I hiked through dust and into mossy forests, past coyotes and wild turkeys, climbed to find myself atop the world, able to peek at land islands garlanded with fog beneath and scalloped clouds above. #VOTENATURE2026
@MikeIsaac i should state we could afford new tins, but were/are also kind of a miserly/penny-pinching family. there was a period where we used to go to ASDA after it had been bought by walmart to gleefully buy the 5p (about 10 cent) ramen packs
@rachelclif it does, however, then look like vomit, so I thought I'd save the timeline from that
@zkevinbai (I did subsequently add some hot sauce, just to liven things up)
Sometimes I have to celebrate my English heritage by making oddly unappetizing seemingly underseasoned food that is, nonetheless, absolute scran.
Consider: baked potato.
one of my colleagues put this in our 1-1 doc for today
@RishiBommasani To some extent, TAI wasn't really a re-org, but rather a formalization of an ad hoc grouping of research teams which had already existed and worked with me for years, though since announcing it we've added more teams (e.g, @mattbotvinick 's rule of law crew)
@RishiBommasani @mattbotvinick To the extent we fund external entities, it's not in a typical grantmaking mode, but rather things like targeted funding to create evidence to compound insights of our own research teams (e.g, the econ team are helping to fund RCTs related to our broader eco research agenda)
Rough eras of recent AI progress in terms of what research community is collectively hillclimbing on: 2018-2022: Basic capabilities (summarizing, coding, etc) 2022-2026: Norm & time coherence (rlhf/CAI, longer context, agents) 2026 - ?2028?: scientific intuition / independence
the future of the world is available to anyone who takes the time to read arxiv papers every week
This morning I crested a hill to see the morning light tearing into a thick low-hanging cloud, causing yellow light to rip at the edges and cascade down as angelic golden rain. #VOTENATURE2026
The beautiful and tragic thing about having young children is you do stuff with them like blowing bubbles on your front porch with your extended family and you think "this is the best time of my life" and you're right.
7pm: Thinking to myself after closing the door on my toddler who I've just tucked into bed: They are an angel. I am so blessed. I love them infinitely. 8pm (after they've come out of their room 7 or 8 times): What if I bricked up their door at night?
Nice analysis from Peter here about one of the puzzles in AI - given all the changes AI is causing within the economy, why hasn't it had measurable impacts on unemployment? x.com/PeterMcCrory/s…
Props to OpenAI for publishing this post on some safety and alignment issues observed in internal deployments - there are many counter-incentives to publishing stuff like this, but by making it public we all get better info about safety at the frontier. openai.com/index/safety-a…
writing this up for next issue of Import AI
Very interesting personal milestone - Fable just had a good suggestion for some text of a fictional story to delete. I find AI writing broadly excruciating, AI sub-editing pretty helpful, and feels like 'AI editing' is now coming online. Story coming in Import AI tomorrow!
Legendary game.
At this point, everyone at the frontier of AI agrees that third-parties should test out AI systems and use these to develop standards to feed into policy - excellent to see @demishassabis laying out a framework to do this! x.com/demishassabis/…
Import AI is skipping this week due to the two forces that currently heavily influence my life - demands of toddlers, and England's world cup journey.
While hiking today I encountered the gnarled base of a tree like some disembodied wooden hand clutching at grass and stone, and behind it the sun loomed so bright that it distorted the picture I took, giving itself an unearthly vast corona. #VOTENATURE2026
though many experts disagree on the nature of the singularity, all agree that the demarcation for the beginning of the "true foothills" lies on June 30th 2026, the day arXiv went from red to black.
@rSanti97 @maxbodach by flourish I mean crash. my typos are merely an indication of my great enthusiasm
@deredleritt3r put another way, I feel like either some weird stuff is happening in eco and we're talking about that, or we have a load of data that says weird stuff is _likely imminent_ and we're talking about that.
Today at 6pm ET I'm in NYC with @sckimbriel for an @aspeninstitute conversation on AI and the future of society. We'll be talking about RSI, the futures implied by AI progress, and the choices we'll need to consider as the technology advances. Watch live: youtube.com/live/iP9wk0pkC… x.com/sckimbriel/sta…
Hiked in a cool redwood valley on a scorching day, the sun so bright it made parts of the canopy shimmer luminous. I walked with my head craned back and mouth agog marveling at the beauty, accompanied by the whispering sussuration of a million leaves. #VOTENATURE2026
Niche tweet: If you have an excitable toddler, get sick of listening to Bluey and Raffi, and need the occasional bit of non-brainslop kid TV to let you do basic tasks like unload the dishwasher, you might like this fan-made video for the Geese song Cobra. youtube.com/watch?v=R6UcL3…
It's sunday, so getting a little loose with Import AI. Wrote up a fun recent paper from @akorinek about measuring AI in the economy, will be covered tomorrow in issue 459.
Wrote a fictional story for the next issue of Import AI that is pretty different to my previous ones. I think there are beautiful things ahead for humanity as AI gets more powerful and I've tried to capture some of that here. Issue should come out on Tuesday.
@headinajaro Note that the above is more about, to use your term, "helping the monkeys see the god". I agree that "controlling" the things requires different tools. Though it's hard to imagine that the control piece doesn't depend in part on legibility and technical assessment
Many people act like AI policy is some mystery where the right solution demands some kind of Policy Einstein who invents general relativity for tech regulation. This isn't true at all! There are many sensible ideas we could do today. All we need to do is choose to do them. x.com/gabriel_weil/s…
The best part is the majority of these ideas also have the property of continually generating information and building state capacity around advanced technology, so they start "paying out" to society in the form of more well-informed governance almost immediately.
Gotta hand it to him - Juergen Schmidhuber had some amazing papers in the 2010s - early stuff on handwriting recognition (arxiv.org/abs/1003.0358), computer vision (arxiv.org/abs/1102.0183), suggesting intelligence be benchmarked via video games (arxiv.org/abs/1109.1314), etc.
Now of course Schmidhuber is divisive and does spend a lot of time trying to appropriately assign credit in the history of AI and through this has become something of a punchline. But you have to admit that he has demonstrated incredible vision and foresight.
