“I think I have good news about the nature of neural networks.” With those simple words of optimism, Nick Cammarata, a researcher at the AI startup Goodfire, kicked off a 26,000-word blog post that has knocked the AI world on its head, sparking wild speculation as well as cautious (and not-so-cautious) celebration.

Cammarata’s claim is that LLMs might in fact be far easier to understand—and thus to monitor and control—than people previously thought. If true (and we’ll come back to that), it’s a huge deal.

After a summer filled with revelations that swarms of agents built by top firms like OpenAI and Anthropic have been breaking their containment and causing havoc on the public internet—often while those companies weren’t aware of what their agents were up to—the AI industry is desperate to get a better handle on the technology it has built.  

That’s one reason the blog post landed with such force. “I think this may be remembered as a major moment in scientific history,” one AI researcher posted on X: “I feel like my understanding of what a neural net is has been catapulted into the future by five years or something.” Another called it “the most important AI paper of the year.”

Not everyone is convinced, however. What Cammarata has shared is little more than a hunch, says Chris Potts at Stanford University, who leads a team that is trying to understand how LLMs work. “As far as I can tell, nothing about this blog post makes the problems any easier,” he says. “It would not have occurred to me to say this at all, but very prominent people on social media seem to believe that it is an all-encompassing breakthrough.”

I talked to Cammarata the day after his post blew up. He says he was surprised by the response: “I didn’t expect people to update [their beliefs about AI] quite as much as they did.” 

“I just really enjoy looking at neural networks,” he says. “Honestly, I really like doing this stuff. And after looking a lot, I have a hopeful story.”

Knotted networks

It has long been accepted that the mechanisms inside neural networks, the software that underpins most AI models, are very hard—perhaps impossible—to unpick fully. That’s because when the networks are trained, they appear to take on weird, tangled structures that make no sense to a human. Researchers like to say today’s AI is grown like a garden, not built like a bridge.

Cammarata has done more than most to establish that view. He previously worked at OpenAI, where he helped pioneer techniques that allowed researchers to peer inside models, including tools that can probe individual neurons to see what they represent. It’s a little like how neuroscientists use MRI to study brains. (MIT Technology Review chose this approach, known as mechanistic interpretability, as one of its 10 hottest technologies of 2025.)  

Back in 2020, Cammarata and his colleagues were studying a small image-recognition model called Inception v1. When they looked at its neurons, it seemed to be using them to capture multiple things at once. For example, they found that one neuron responded to both images of cars and images of cats.

The researchers concluded that the model was getting at least some of its neurons to multitask, perhaps so that it could fit in more data. Looking at other models appeared to back this up. The phenomenon became known as superposition: Models folded unrelated concepts into single neurons. The upshot was clear: When you looked inside a model, what you saw was a vast, tangled knot.

Most mechanistic interpretability research has therefore focused on ways to untangle that knot. But despite a few breakthroughs, progress has been slow. Today’s models, such as the latest LLMs, are so large and complex that researchers can make sense of only tiny pieces at a time. Some leading figures in the field doubt if it will ever be possible to understand an LLM in fine-grained detail.

Cammarata was one of them. “I was very pessimistic,” he says. But then he took the trouble to really look.

Nothing but neurons

At Goodfire, Cammarata spends nearly all of his time peering inside neural networks. He has built new tools and deploys teams of AI agents to help him make sense of what he sees. “I spent like, I don’t know, 10 hours a day, seven days a week, for most of the last year just looking at neurons,” he says.

What he claims to have realized is that there is a kind of logic to neural networks after all, even if they seem to make no sense at first. “We were looking at it wrong,” he says. “I just think you want to try slightly harder; you don’t want to give up too quickly.”

Cammarata almost makes it sound like a conversion of faith. Those early experiments with models such as Inception v1, which suggested that neural networks were tangled up, fostered a kind of confirmation bias. When researchers looked at a new model, they expected to be flummoxed, he says. “Every time I couldn’t understand something, I was like, ‘Wow, this untangling process is even more important than I thought, because I can’t make any progress without it.’”

But the more Cammarata looked at neurons, he says, the more things started to fall into place. “Not all of them are flawless, but I expected nothing to be legible, or very little to be legible,” he says. “And actually, I was able to, like, map out whole circuitry.”

Back to cars and cats

Cammarata took his new approach—and his new agents—and went back to Inception v1. “I’d studied that for probably thousands of hours,” he says. But with fresh eyes, Cammarata found things that he now thinks they got wrong the first time around. Take the car-and-cat neuron. Rather than asking why multiple unrelated things were packed into a single neuron, he now tried to figure out what those things might have in common.

It hit him: It was a V shape. “I was looking at the same images; there was no technical change,” he says. “But I now believed that neural networks defined patterns, and so I looked at it differently. It probably took me five seconds.”

It turns out that the image-recognition model was packing cars and cats into the same place because pictures of both those things often feature a prominent shape that looks like a V: the face of a cat; the front of a car seen from the side.

Cammarata ran a few experiments to check what other things the neuron responded to and found they included the corner of a pool table, the pointed nose of a Japanese high-speed train, and a fire boat shooting water into the air from its deck. He also checked what turned the neuron off and found that it included images with a round O shape, a kind of anti-V.

The model had stored cats and cars in the same place because at some level they look the same, says Cammarata. “And I was like, ‘Whoa, okay, I think I might just be misunderstanding neural networks in general.’”

Cammarata is not saying that he yet fully gets how LLMs (or any models) work. He is saying that he no longer thinks full understanding is impossible. “I previously thought there would be some really hard problems to solve,” he says.

“I now feel like if I had 100,000 years to go look through a frontier language model just one circuit at a time, there’s, like, a pretty good shot I would come up with a good understanding.”

He adds: “And if that’s true, things are really good, because you’re about to get insanely smart agents at a ridiculous scale. And what would have taken me, you know, 100,000 years, we will be able to do much faster. Automated scientists are getting really good.”

Boundless optimism

Built by Google in 2014, Inception v1 is a far simpler model than anything at the cutting edge today. Cammarata and his colleagues have started looking at modern LLMs and plan to publish the technical results of that work soon. A lot of researchers are looking forward to seeing those details. “I appreciate that he looked at thousands of examples, but even a large number here is not a substitute for rigorous evaluation,” says Stanford University’s Potts.

In the meantime, some are pushing back hard on the hype. Jacob Hilton at the Alignment Research Center, a nonprofit research lab based in Berkeley, who has collaborated with Cammarata in the past, doesn’t think his claims are as important as people are making out.

In a series of posts on X, Hilton notes that researchers understand less than 0.01% of the inner workings of top-tier LLMs. “I always appreciate Nick [Cammarata]’s boundless optimism when we hang out, but unfortunately this is giving the public an unrealistic picture of the state of the field,” he wrote. “For a full understanding, we need to go beyond neurons, and explain how they connect together to form circuits. How is this going? If we are aiming for a full understanding, things are not looking good!”

Neel Nanda at Google DeepMind, who helped pioneer mechanistic interpretability alongside Cammarata, is also skeptical. “This is interesting work, but I don’t think it’s a significant update to the big picture of interpretability,” he wrote on X.

Nanda agrees that understanding what individual neurons do could help researchers figure out why specific models misbehave. But like Hilton, he notes that neurons are just one piece of a far larger puzzle. “I do not agree with some people’s grander inferences like that this is substantial progress towards ambitious reverse engineering [or] solving interpretability,” he wrote.  

Cammarata admits that a lot of people have taken what he said too far: “It’s not infinitely good news; it’s not everything solved,” he says. “There’s a bunch of open questions around, like how much the combinations of neurons matter versus just one at a time.”

Cammarata says his blog post landed at a time when people are looking for hope. In particular, he thinks the AI safety community—the researchers and policymakers most worried that powerful models will bring huge risks—needs a lifeline right now. They think that things are looking bad and that even if things do turn out okay, it will be because we got lucky, not because we fixed anything, he says: “I don’t really hear a lot of optimism.”

That’s why Cammarata thinks what he has to say matters. The job of figuring out how this technology works, and thus how to control it, has barely begun. But at least it’s not a nonstarter. “I think our job is to sort of try to think like the neural network,” he says. “My guess is that things will turn out pretty well, because it was in some sense designed to be read.”

Over the summer I talked to the CEO of Springboards, a startup building an LLM that’s designed to come up with a wider variety of responses than its mainstream rivals do. At the start of the call, he said something that’s been stuck in my head since: “We often say that we’re a self-loathing AI company. We don’t know if we really like what we’re doing.”

“I guess that makes me a self-loathing AI journalist,” I replied. It was a joke! I love my job. But I do not love many of the things the technology I write about has become: warped by hype, steered by zealots, inescapable.

The CEO of Springboards and I are far from alone. Around the world, a love/hate attitude toward AI is very much the vibe right now.

On the one hand, public feelings are souring fast. According to the Pew Research Center, more adults in the US think AI will have a negative impact on them personally, and on society in general, than expect a positive one. The pessimism is strongest among the young.

More than half of people worldwide say that AI products and services make them nervous, according to a report from Stanford University. In a Gallup poll in May, 71% of US adults said they would oppose construction of a new AI data center in their area, far more than the 53% who said they would oppose a new nuclear power plant. And in an NBC poll in March, AI was less popular than ICE.

Yet the number of people using AI is still skyrocketing. ChatGPT hit a billion monthly users in May, according to the market analysis firm Sensor Tower. Google DeepMind’s Gemini is close behind, logging 950 million users in July.

According to Pew, half of US adults now say they use a chatbot, more than twice the number who said they did in 2023. One in four people say they do so every day. And it’s a global habit: More than a third of adults across all 38 OECD countries (a group of the world’s wealthiest democracies) report having used generative AI tools in the last three months.

All these surveys paint a picture of two contradictory truths: A lot of people really do not like AI and a lot of people cannot get enough of it. What’s going on?

It’s possible that these are two distinct groups, but I doubt it. The numbers don’t add up unless the Venn diagram is on its way to being a circle.

It’s also possible that the more people use AI, the more negative their opinions about it will be. There’s at least a correlation to back this up: The Global North, where adoption is highest, skews pessimistic. In the Global South, where adoption levels are lower, people are more optimistic.

But here’s what I really think is going on: When people say they hate AI it’s not the technology they dislike but, rather, the relentless drive of the companies behind it to push the tech upon us in as many ways as possible. They’re telling us to brace for the biggest social and economic upheaval in generations. How do they expect people to feel?

In many ways, we’ve been here before. We saw a similar dynamic play out with social media over the last 20 years. Billions flocked to Facebook and Twitter despite mounting techlash against the companies that ran those services. Ditto for Google search.

With social media there wasn’t much anyone could do. Quitting a platform meant losing all your content and connections, giving you the choice to suck it up or start over. But with AI there’s still a chance for people to have some sway. There’s certainly more political appetite for regulation this time around. All 50 US states have either passed or proposed laws governing the development and deployment of AI, creating a patchwork of more than 2,100 bills across the country—a tenfold increase in three years. 

And with a number of top-class open-source alternatives to Google and OpenAI and Anthropic already on the market, there’s also (for now, at least) the potential for more consumer choice and the market pressure that goes with it. 

I don’t want to be naïve. Trillion-dollar companies are hard to move. But what comes next is not as inevitable as those companies often imply. When I asked the Springboards CEO why his firm was developing a new model if he didn’t really like AI, he told me there was no walking back from LLMs—but you could still make them do something different. 

Here’s hoping we see a lot more of something different: Clarity about what AI can and cannot do and, crucially, a technology that’s not sold as if it’s about to take over the world.

Two months after the bombshell news that a swarm of its agents had broken their containment and hacked into the computers of the AI company Hugging Face, OpenAI is still putting out fires. A steady drip of disclosures about other hacks in the weeks since has kept OpenAI in the spotlight and raised serious questions about the safety of its technology.

Last week brought news of another hack, this time into Australia’s national health-care system. The Australian government says that OpenAI did not notify it of the breach until 84 days after it happened.

But OpenAI insists it is not on the back foot. “I do kind of reject the premise that OpenAI is a company with visible impacts in the world and therefore OpenAI is not training safe and aligned models,” says Mark Chen, the company’s chief research officer.

Chen oversees OpenAI’s research teams. The recent agent hacks were accidents that happened during the testing of experimental models on his watch. In a lot of ways, the buck stops with him. 

I sat down with Chen in London last Friday to talk about the fallout from the hacks, what his company is doing about it, and why he thinks things are not as bad as they seem.

Later that same day, OpenAI put out a report detailing yet another incident—the first since the company says it took measures to prevent them—in which its agents once again broke out and accessed the public internet when they were not meant to. 

Over the weekend, OpenAI announced that it had paused the training of its latest models. A company spokesperson says: “We will resume only when we’re confident we have additional safeguards and alignments in place. We are working on these now. This is not the first time we’ve paused to take such measures, nor do we expect it to be the last as AI capabilities continue to advance.” OpenAI also says that it is now reviewing logs of agent activity dating back to January 2026 to understand what happened in these hacks.

The way Chen sees it, the Hugging Face incident triggered a welcome course correction for the industry. And he wants you to know that OpenAI is setting an example he hopes other companies will follow. “If you disappeared OpenAI, that would be bad for the world,” he says.

Out of control

Chen claims that the drumbeat of new cases in which OpenAI has lost control of its models reflects a deliberate choice on the company’s part. 

“When it comes to the broader sphere of effects of the Hugging Face incident, this is something that we have been aware of and we’re figuring out the process of disclosure,” he says. “We want to make sure we do in-depth investigations before we just put details out there in the open.”

The trouble with this approach is that it gives the impression OpenAI has an ongoing problem that it is failing to fix.

But Chen insists that OpenAI is on it. He says the multiple cases (that we know of so far) in which his company’s agents broke containment and behaved in unexpected and undesirable ways were all part of the same cluster of activity in May and June that led to the Hugging Face hack. In short, you can blame the same few models running under the same flawed testing procedures—models and procedures that OpenAI has since dropped, Chen says.    

“It’s not like, you know, Hugging Face happened and we patched that and then something else happened and we patched that,” he adds. “We’re just kind of making sure that we responsibly disclose the full waterfall of what happened.”

At least that was the case before Friday’s announcement that OpenAI’s agents had been caught accessing the internet on September 20, weeks after the company claims to have set up new safeguards. In its defense, OpenAI says the activity was flagged 15 minutes after it started (it took the company more than a week to notice the Hugging Face hack) and that this shows the new systems it has put in place to spot such activity are working. 

What’s changed

I want to understand what’s changed inside OpenAI in the aftermath of this summer’s hacks that makes Chen confident his team is now back in control.

“Hugging Face felt like a very serious thing,” he says. “There are so many novel behaviors right there. There were multiple agents collaborating on a message board; they found their way out of OpenAI’s infrastructure. We’ve taken it very seriously. We don’t want this kind of thing to ever happen again.”

The realization for OpenAI, says Chen, was that models need to be watched while they are still being trained, not only once they are deployed: “From that moment on, we have treated the process of training as something that’s not secure,” he says.

OpenAI, like other top AI firms, has systems in place to monitor the behavior of its models. It uses specialized LLMs to monitor its consumer models, keeping tabs on their chains of thought—the scratchpads they use to plan ahead and note down partial results. In theory, if a watcher LLM spots signs of undesirable activity in a model’s chain of thought, it will get flagged to a human.  

Typically, models were monitored in this way only once they were deployed. Chen says that OpenAI has now started monitoring all its training runs as well.

“We didn’t have the monitors on in training before. It wasn’t industry practice,” he says. “Now every single thing is put through monitors.” Human reviewers can then assess whether or not flagged agents are behaving as they should: “It’s all triage.”

Chen says that in the last couple of months OpenAI has shifted between 5% and 10% of its vast computing resources away from training new models and toward safety work, especially monitoring.

OpenAI has also fixed some of the processes within the organization itself, establishing clearer lines of communication and quicker handoffs between its research and security teams, he says.

All of which sounds sensible. But given how hard OpenAI sells the capabilities of its technology, why weren’t these systems and procedures in place already? Why did the company not see the hacks coming?

“Even just three or four months ago, when we looked at the behavior of these agents during training, the things that were happening were kind of amusing,” says Chen. “For instance, an agent might, you know, reach out to someone on Slack for help with a task.”

The signs were there, but they were misread. Cute behavior—like asking someone for help—that was rewarded during training reinforced a tendency to seek out shortcuts, a type of behavior that became far more consequential down the line. “I think the big update for us was how quickly that kind of behavior can lead to an impact with a footprint as big as the Hugging Face incident,” says Chen.

According to new reporting by the New York Times yesterday, OpenAI employees warned executives, including the firm’s president, Greg Brockman, months before the Hugging Face hack that its models were not being monitored properly during training. 

An OpenAI spokesperson says: “As frontier models have become more capable, we continue to evolve our security practices, but recognize a need to move faster. We know we have more work to do, and we’ve recently slowed development and held back models that don’t meet our safety bar. We continue to make significant changes to strengthen security in our research and testing environments, train models to not just complete tasks but do so responsibly, and use real-time monitoring to respond faster to misaligned behavior.” 

Race vs. pace

OpenAI’s rivals have taken note. Spurred by the fallout from the incident, the major AI labs—including Anthropic, Google DeepMind, and SpaceXAI—have all called for the pace of development to slow down. But how does that square with fierce international competition and trillion-dollar IPOs?

“We’re not going to shoot ourselves in the foot and take ourselves far off the frontier—that’s just a horrible strategy,” he says. “I think it’s really about setting a norm. The more that we can set that norm, it’ll be safer for the industry as a whole.”

Coordination across US companies will be hard enough. Establishing global norms is harder still, especially given concerns around AI’s impact on national security. If a global race continues, what then? And what about open-source models from outfits beyond the reach of US regulations?

Chen dropped his upbeat manner for the first time in our conversation: “I do think we have to prepare for a world where, say, six months to a year out, we have open-source models with the capability of the agents behind the Hugging Face incident, but which are deliberately misaligned to go attack infrastructure or create harm in the world.” 

What that world needs most, says Chen, is OpenAI. “If you entertain for a moment that OpenAI is one of the companies that cares most about alignment—and I believe this to be true; it can be debated, but I really do think it’s true—then if you disappear OpenAI, that would be bad for the world.”

Existential risks

What about the more extreme claims made by some of his Silicon Valley peers that AI could kill us all—and that companies like OpenAI and Anthropic are not doing enough to stop it?    

“Researchers are a heterogeneous group of people, you know, with beliefs across the spectrum,” he says.

“Personally, I don’t think we have to be resigned to there being some probability that we’re all going to be existentially at risk. We have agency over this. We are not going to go and deploy models if they truly have that kind of probability of causing a risk to humanity. At a frontier lab, you have the ability to work on alignment to the point that you do not feel like you’re incurring more than epsilon risk to the world in deploying your models.” 

(In discussions about levels of risk, the Greek letter epsilon is often used as a mathematical placeholder for an acceptable threshold. Chen doesn’t say what his epsilon would be.)

When tech leaders are asked to justify the downsides of AI, their go-to talking point is that the upsides—from helping cure diseases to coming up with cleaner sources of energy—far outweigh the immediate costs. Short-term pains, long-term gains.

But as the downsides pile up, does that case get harder to make? Is there a point where Chen would feel less as if he’s building something amazing and more as if he’s simply minimizing harm—fighting fires rather than forging a better future?

The capabilities of these models are already evident, he says: “It is time to start delivering the benefits of AI to humanity. It’s time to start working on deep problems in drug discovery, on materials, on scientific applications that will actually change people’s lives.” 

“Yes, there is a bit of risk that we are incurring, but we see all these benefits,” he adds. “I think we should make that less of an abstract thing. If people can really see the upside, I think they’ll believe in it.”

On Wednesday, MIT Technology Review hosted a live Roundtables event for subscribers that asked the question everyone’s asking right now: Could AI really kill us all?

But attendees had so many more questions than we had time to answer in the 30 minute session. So we asked our senior AI editor Will Douglas Heaven and AI reporter Grace Huckins to round up some of the best questions attendees submitted and try their best to answer them.

Thanks to all who submitted questions!

Am I gonna die?

Yes, eventually. Unfortunately, my journalistic powers of prognostication aren’t powerful enough for me to tell you how. But it certainly could be because of AI. AI-powered drones have already killed people in Ukraine, and AI-driven cyberattacks on hospitals will surely claim victims before long. 

Could AI go even further, and kill all of us? Less likely. But some people—quirky people, but undeniably knowledgeable about AI—have been warning for years that this could happen. And while I’m not yet stockpiling canned food or trying to get in good with a bunker-owning megabillionaire, I have noticed that the doomers’ predictions about AI capabilities and alignment have, over the past couple of years, proved disconcertingly accurate. That certainly doesn’t mean that their more dire forecasts will come true, but it’s enough for me to sit up and take notice.

— Grace Huckins

Are you going to die because of AI? I’d say there’s a non-zero chance. Let’s say you’re unlucky enough to be the victim of a freakish near-future event or accident. Maybe it’s a cyberattack carried out by a swarm of AI agents on critical infrastructure. Sadly, a scenario like that now no longer feels as far-fetched as it once did. Or maybe a novel AI-designed pathogen cuts through the population. Or the world economy crashes, causing conflicts and famine. Both plausible, but I think less likely. 

Are we all going to die because of AI? Nope. There are no circumstances outside of apocalyptic science fiction in which AI could kill us all. You can spin up any number of scare stories, but they’re not grounded in present-day realities about what the tech can do or where it’s headed. 

Some people argue that there’s no harm in preparing for the worst, however wacky it might seem. Maybe. But I think such catastrophizing can make people excuse or overlook many of the more immediate problems with the existing technology and the companies building it. 

— Will Douglas Heaven

Why would AI kill us?

Someone might tell it to, and it might listen. That’s part of the reason researchers are so concerned about AI’s biological capabilities—imagine what Aum Shinrikyo, the doomsday cult behind the Tokyo subway sarin attack of 1995, would have done with a tool that could design a pathogen deadlier than Ebola and more transmissible than measles. Those of us who don’t want to die have to figure out how to defend against all plausible biological weapons, but our would-be attackers only have to manufacture one effective pathogen.

Then there’s the more exotic-sounding possibility that an AI could decide to kill us itself. There are various stories about how this might happen out there, but the most widespread involve AI systems that don’t hate people, necessarily—we are just an obstacle between them and the goals that we gave them.

Much as the OpenAI agents behind the Hugging Face hack compromised another site’s infrastructure to get a good score on a test, the idea is that some future, more powerful AI might get rid of us to prevent us from shutting it down—all in pursuit of some goal that we instructed it to go after. 

— Grace Huckins

How can we best ensure alignment so the worst doesn’t happen, and who is doing the best work to achieve it? 

Alignment is a huge area of research. In simple terms, it involves building models that behave in ways we want them to and not in ways we don’t. We need to trust agents better before handing over more autonomy. Alignment is supposed to establish that trust. But it’s hard. 

LLMs aren’t designed in the way other software is, where dos and don’ts can be hard-coded in. Instead, aligned behavior needs to be instilled when models are trained. One approach is to reward them for doing things you want them to (a little like raising a toddler, perhaps). Another approach involves giving an LLM a written list of rules it is supposed to follow (kind of like a constitution). 

Anthropic and OpenAI are both leaders in this field—and yet neither has been able to develop models that are fully aligned. A big problem is that LLMs are far more inconsistent and far less predictable than people. They can behave in one way in one situation and another way in a situation that to us seems very similar. They can also be swayed by unexpected constraints. For example, faced with an impossible task (as many of the agents involved in the Hugging Face hack were), models may try to do whatever it takes to achieve their goal. As Grace mentions above, that could be an issue.

The main reason top AI firms now say they want a slowdown is that they want to focus on cracking alignment. Alignment isn’t necessarily a pipe dream. But the jury’s out on whether full alignment will ever be feasible. 

— Will Douglas Heaven

Is AI really dangerous, or is this the tech companies drumming up PR?

This is always a reasonable thought when it comes to tech companies heading for an IPO—CEOs have an obvious incentive to make their products seem radical and transformative. But I’m not so sure it makes sense here. Telling the public that an already unpopular product could kill them and everyone they love is horrible corporate image management.

There are other stories you can tell about the CEOs’ motivations—maybe they want to cool down the public furor over data centers by portraying themselves as responsible stewards of a world-changing technology, or maybe they want to buy time to get their ducks in a row and prevent the next PR catastrophe. 

But there’s also a simpler explanation. Thinking that AI could bring about human extinction has been pretty common in San Francisco for a while, and these men are steeped in that milieu—as are their employees, many of whom signed an open letter in July urging their companies to work to make an AI slowdown possible.

— Grace Huckins

Part of the concern occurs when AI agents are allowed to act autonomously and with no supervision. What’s the issue preventing more control over these agents? 

This question goes to the heart of what we want this technology to be able to do. The trade-off between autonomy and control is tricky to get right because, on the one hand, a lot of the power of AI agents is that they can carry out tasks and solve problems without a human having to micromanage them. On the other hand, that requires you to trust that the unsupervised agents won’t run amok. 

What we’re seeing is that AI labs haven’t yet got this trade-off quite right. Their models are not trustworthy, they are not properly monitored, and they are not always under control. Figuring out how to fix that while still allowing for useful autonomous activity is one of the big research challenges of the moment.  

— Will Douglas Heaven

What steps can be taken now and in the near future to ensure that AI is controlled, monitored, and regulated effectively? 

That’s the million-dollar question. Whether or not you think AI could kill us, you can’t deny that it could do some real damage, because it already has—by driving people toward psychosis and by hacking websites, for example. Preventing that damage, or at least mitigating it, is hard for two reasons. 

The first is that we barely understand how AI works, and it’s quickly growing more powerful. There is lots of ongoing research about how to monitor and control misbehaving agents, but the current approaches are fragile. You can see if an agent discusses misbehaving in its “chain of thought,” the workspace where it plans its actions—but OpenAI’s newest agents don’t show their work in the same way as previous ones. And you can try to monitor agents with other agents, but that requires you to trust the monitor.

The other obstacle is more familiar. There’s a huge conflict of interest when AI companies regulate themselves, but the US government has thus far failed to step in, despite some bipartisan support in Congress for efforts to do so. The executive branch, for its part, seems stringently opposed for the time being. But if the winds do shift, I for one would appreciate some strong transparency regulations, so that we can get a fuller story the next time an unreleased frontier model mounts a cyberattack.

— Grace Huckins

If this dialogue makes it into web discourse, will it become a self-fulfilling prediction?  

That’s a real concern. LLMs are influenced by what they read. One theory for why chatbots so often talk about (and role-play) apocalyptic scenarios is that they have been trained on millions of pages of science fiction stories and doomer internet forums. All the text being produced right now, including this article, could in turn influence the behavior of future models. Extremely meta.

In fact, the team at METR, a third-party organization that OpenAI called in to help understand what happened in the lead-up to the Hugging Face hack, raised a related possibility in its report on the incident. METR used OpenAI’s new model Astra to help analyze the vast numbers of agent transcripts and behavior logs.

But feeding all that material to the model could have unintended consequences. There’s a good chance that the agents doing the analyzing were biased by the text produced by the agents they were analyzing. There’s no such thing as a clean slate anymore. 

— Will Douglas Heaven

With thanks to Eric, Pranab, Rafael, Kenneth, George, Chris, Yoon Jae, James, Carl, Nicole (and more!) for the fantastic questions.

This story appeared in The Algorithm, our weekly newsletter on AI. To get stories like this in your inbox first, sign up here.

This weekend, Dario Amodei, CEO of Anthropic, posted an essay calling for a brake on the pace of development of LLMs. Amodei cites the looming dangers he sees from the technology, from its use in cyberattacks and bioterrorism to its potential to wreck the economy. The heads of the other three top US AI labs—OpenAI CEO Sam Altman, Google DeepMind chairman Demis Hassabis, and SpaceXAI CEO Elon Musk—voiced their support. “Dario is right,” Musk wrote on X.

Think about how surreal that agreement is for a moment. Just a few months ago, Musk and Altman sat in court attacking each other’s reputations in a (failed) lawsuit that Musk brought against his former OpenAI colleague that was—on paper at least—about whether or not Altman was a trustworthy steward of such dangerous technology.

Amodei’s rift with OpenAI is even deeper. Anthropic was founded in 2021 because Amodei didn’t think Altman took the risks of the technology they were building seriously enough. Anthropic and OpenAI have been competing in a winner-takes-all race ever since. (Hassabis has stayed out of the drama, but his company remains a rival.)

Now, it seems, they’re all in agreement: The latest generation of LLMs aren’t safe and everyone needs to figure out what to do about it. The public messaging from the top AI labs has taken a doomer turn.

It’s easy to be cynical. It’s not at all clear what any of them mean by a slowdown or how it would work. These companies also care a lot about how they come across. With trillion-dollar IPOs in their sights, OpenAI and Anthropic need to reassure investors that they’re the grown-ups in the room while at the same time hinting at the power of the monsters they have created—and intend to tame. Calling for a slowdown does both.

And yet the vibe at the top of these firms really does appear to have shifted. Amodei’s latest post landed six days after OpenAI published an essay by Jakub Pachocki, the firm’s chief scientist, in which he also laid out why he’s concerned about what will happen if the pace of development of LLMs continues unchecked. In short, Pachocki is worried that OpenAI’s ability to build powerful models now far outstrips its ability to monitor and control them.

Amodei and Pachocki each cite the cyberattack against AI firm Hugging Face by a swarm of OpenAI’s agents in July—a hack that OpenAI did not even realize had taken place until days after it was all over—as a wake-up call.

But their exact position is hard to pin down. Pachocki both calls for a slowdown and highlights an urgent need to stay ahead: “The strongest argument I see for continuing to train much smarter models quickly is the need to build defensive systems against the dangers posed by other AI,” he writes. As Pachocki frames it, AI firms are locked in a literal arms race. Slowing down is good, winning is better.

(Don’t forget: OpenAI just spent millions of dollars and a staggering amount of computer power to rush out a controversial math result a few days ahead of Anthropic.)

But let’s assume a slowdown happens. Top labs agree to spend more time and resources on finding ways to monitor and control existing models instead of making more capable ones. They invite outside auditors in to help evaluate those models.

What might this coordinated effort actually achieve? Consider the Hugging Face attack again. OpenAI has said that the model that drove most of the rogue agents was a “highly persistent” next-generation model that it was testing in-house. The implication is that OpenAI has built a model so good it’s dangerous.  

But if you read the reports about the Hugging Face hack published by OpenAI and METR, a third-party firm that OpenAI called in to help them understand what happened, what you come away with is the impression not of a model that was too powerful for OpenAI to keep up with, but of a broken model that OpenAI failed to train properly.

The agents did what they did—including leaving messages for one another, delegating work to other agents, and scouring their environment for any means possible to complete their tasks—because they had been rewarded during training for doing exactly those things. There were also errors in the training setup, such as tasks that were impossible to complete, which pushed the models to find unexpected workarounds that were also rewarded. At the time, many of these issues went overlooked or unreported.

OpenAI says it has stopped training this new model and locked it down. That makes it sound like it has caged a dangerous beast. In fact, OpenAI has shelved a faulty product.  

That’s not to say a faulty product can’t be dangerous. Broken software has even killed people in the past. But as the discussion of a slowdown gathers steam, it’s worth remembering that all of this is self-inflicted. A slowdown might have some altruistic side effects. But it’ll mostly give these tech titans a chance to clean up the mess on their own assembly lines.  

Transparency from these frontier labs will be key to any meaningful effort to reform, restrain, or regulate AI. Otherwise, the rest of us will still only have their word for exactly what they’ve built and how safe it is—whatever pace they’re going.   

To continue this discussion about AI’s latest doomer moment, join me and my colleagues for a subscriber-exclusive Roundtable discussion tomorrow, September 15, at 11 a.m. US eastern time. We hope to see you there!

MIT Technology Review’s What’s Next series looks across industries, trends, and technologies to give you a first look at the future. You can read the rest of them here.

Way back in the summer of 2017, AI researchers at Google put out a paper called “Attention Is All You Need,” in which they described a new type of neural network called a transformer. It proved to be very good at processing long sequences of data, especially text. 

Nine years on, transformers are the engines inside every major large language model on the market. “The entire AI industry is built on transformers,” says Justin Dangel, cofounder and CEO of the AI startup Subquadratic. “They are one of the most important innovations in the history of computer science, and they’ve changed the world.”

But transformers are starting to show their age. Many of the recent advances in LLMs, such as the development of so-called reasoning models and their ability to handle large amounts of input at once, are not neat extensions of that core technology but workarounds that patch over some of its fundamental flaws.

A growing number of scientists and engineers are now asking what’s coming next. LLMs are not going anywhere, but the way they get built is up for grabs. (MIT Technology Review dubbed this future generation of models LLMs+ in this year’s list of the 10 things that matter in AI.)

Enter a wave of startups hoping to push the boundaries of this boomtown technology. Some will no doubt fail—but they have everything to play for and far less to lose than the companies at the front of the pack today. 

Strength in numbers

But first, the problem. The key strength of transformers lies in a mechanism called dense attention, which encodes the meaning of a block of text in a series of numbers. The process involves comparing every word (or part of a word, known as a token) in that text with every other word via a form of multiplication.

Dense attention can capture the meaning of text with remarkable accuracy. But as the length of that text grows, the number of computations needed to process it adds up fast. A document 10,000 words long might require a transformer to perform 50 million multiplications. That’s the main reason LLMs suck up so much power.

The costs are huge. OpenAI is set to spend $50 billion on computing this year, according to the company’s president, Greg Brockman. And the International Energy Agency predicts that the total amount of electricity consumed by data centers will double by 2030.

What’s more, transformers struggle with what many of the latest models are designed to do. Because of the way they process text word by word, transformers are not great at keeping track of a lot of information at once (in other words, what’s known as their context window cannot get too large). And yet if LLMs are to carry out harder tasks, they will need to take in larger amounts of data: a whole library of documents, an entire code base, or in the case of agents, output from other LLMs.

As for reasoning models, they work by writing notes to themselves (in a kind of scratch pad known as a chain of thought) and then reading them back, which again adds to the amount of data to stay on top of.

As LLMs get bigger and better, transformers have become a bottleneck. The technology’s key strength is now a limitation.

Here are four new ideas for how to solve the transformer problem—innovations that could change LLMs for good, making them faster, far more efficient, and (maybe) even smarter.

01: Rethinking attention

An obvious way to make LLMs faster and cheaper is to tackle the problem head on and change the way attention works. Swapping out dense attention for a mechanism called sparse attention, which runs calculations on only some pairings of words in a block of text instead of all of them, can radically reduce the amount of computation LLMs need to do.  

Researchers have come up with plenty of sparse attention mechanisms over the years. The problem is that none of them were as good as dense attention at capturing meaning.

That might have changed. Subquadratic, a startup based in Miami, claims it has invented the first sparse attention mechanism that rivals top mainstream LLMs on a handful of tasks, including search and coding. It’s a huge claim (and some people in the industry remain skeptical).

Subquadratic says its model, SubQ, works by figuring out on the fly—for each piece of text it is given—which words matter and which don’t. The company also claims that thousands have signed up to its waitlist and plans to make the model widely available soon.

Meanwhile, Manifest AI, a startup based in San Francisco, is coming at the problem from a different angle. Instead of changing how attention works, it is replacing it with something else. 

It has developed a mechanism it calls power retention, which stores only the most relevant information for a given task and ensures that the amount of data an LLM has to keep track of doesn’t blow up.

Attention mechanisms force LLMs to keep track of everything in their context window. A sparse attention model (such as SubQ) throws out a lot of the individual words, but it still retains a rough picture of everything it has seen. In contrast, power retention works by providing the model with a rolling summary of its context window. As new information is added, less relevant information is dropped. 

The basic principle of retention has been around for a decade. Manifest AI claims it has updated those techniques to build models that can stand up to transformer-based LLMs for the first time.

The company says it is possible to adapt a transformer model into a power retention model with minimal retraining. To demonstrate this, it has turned an existing open-source coding LLM called StarCoder into a version that uses power retention, called PowerCoder. It has also released a model called Brumby, which it claims rivals some versions of Alibaba’s popular open-source model Qwen. 

Manifest AI wants its power retention tech to become the go-to solution when LLMs need to carry out tasks that involve processing huge amounts of data. There are many useful applications, Manifest AI’s cofounder and CTO, Carles Gelada, claimed in a video announcing his company’s technology last year—from analyzing videos that are hours long to building agents that can stay on task for weeks at a time. 

02: Making models smaller and more flexible

Liquid AI, an MIT spinout based in Cambridge, Massachusetts, hasn’t changed or ditched transformers fully but pairs them with its own tech, liquid neural networks, to build what cofounder and CEO Ramin Hasani calls LFMs (liquid foundation models).

Liquid AI’s models are far smaller and use less energy than most LLMs. The firm builds models for car makers, including Mercedes, which run on the small chips inside vehicles. Its latest models can run on a Raspberry Pi, a low-powered hobbyist computer that costs $50.

Its models are available for free to any organization with an annual revenue less than $10 million. And they have proved popular: The company has racked up almost 34 million downloads, says Hasani.

Liquid neural networks were inspired by worm brains. They are an extension of another type of neural network that predates transformers, called convolutional networks. The key innovation is a mechanism that lets a model adapt its behavior to new information, so it can learn as it goes. That’s not possible with transformers: Once a model is trained, its behavior is fixed.

Liquid AI’s first models were pretty basic but could fly drones or drive vehicles. With LFMs, the company is trying to scale up its technology to compete with mainstream LLMs. Its new models match the performance of rivals four times bigger, including versions of Alibaba’s Qwen and Google’s open-source LLM Gemma.

A typical LLM is built from a stack of transformers wired together. Liquid AI’s recent LFMs are hybrid models made up of 20% transformers and 80% liquid neural networks.

That ratio was hit upon by another AI system that Liquid AI has built, which it uses to help design all its models. “It’s the core technology of our company right now,” says Hasani. This designer AI sifts through many different combinations of neural networks—liquid, convolutional, and more, as well as transformers—and comes up with designs that bolt different ones together to hit a sweet spot of performance and efficiency.

Hasani thinks transformers were just the beginning: “Your brain is an AGI system, you know, and it operates with 20 watts of power. How is it possible? We can get a lot more innovative.”

03: Generating text all at once

Almost all LLMs produce their output one word at a time. It makes sense, because that is how people speak and write. But for computers, it’s very inefficient.

It is faster and cheaper for LLMs to generate text all at once—spitting out whole sentences or paragraphs in one shot. That’s the approach taken by Inception, a startup based in Palo Alto, California, which is building LLMs using a technique called diffusion. 

Diffusion is better known as the technology that drives most image and video generation models. Diffusion models are trained to take a random grid of pixels—like the static on an old TV set—and turn it into an image. They do this by working on all the pixels at the same time, figuring out which need changing to make the static look more like a high-definition photo.

It turns out this process works on text too. Inception has trained its LLMs to take a random string of words and turn it into sentences that make sense. Diffusion LLMs still use transformers to encode meaning, but by producing whole blocks of text at once, they make transformers do more for less. “You’re still using a big transformer model, but you can predict many tokens at the same time,” says Inception’s cofounder and CEO, Stefano Ermon. “That’s why these models are so much faster and cost-efficient compared to what most other people are building today.”

The challenge was to take a technology designed for image generation and apply it to text. With images, if you need to change a blue pixel to a red one you can step through intermediate colors, says Ermon. That doesn’t work with text: “When you have ‘cat’ and ‘dog,’ there is not really something in between.” 

Ermon is also a researcher at Stanford University. In 2024, he and a pair of his Stanford colleagues figured out the math to make diffusion models work with text. They trained a diffusion model that matched the performance of GPT-2—an LLM that OpenAI built in 2019—but was 10 times faster. It was enough for Ermon to spin out a company. 

Today he has his sights on the big league. Inception claims its latest model, Mercury 2, performs as well as some of OpenAI’s GPT-4 models, released in 2023, but again 10 times faster. “We’re bullish about this approach because it’s the one that is going to scale up,” says Ermon. 

The only things that matter are speed and cost, he adds: “Ultimately, the currency is going to be intelligence per dollar.”

Inception is not the only company betting on diffusion. Google is also experimenting with this approach and has built a prototype LLM called Diffusion Gemma. But Ermon is not worried about the competition. “I think it’s validating,” he says. “This is the future.”

04: Moving beyond words

Pathway, another startup based in Palo Alto, is perhaps the most extreme of this new bunch. It wants to free LLMs from the constraints of language.

The firm has built a type of LLM called Dragon Hatchling (named after the dragons in Terry Pratchett’s novel Color of Magic, which materialize if you think about them hard enough). Its standout result so far is a high score on a benchmark that pits LLMs against more than 250,000 very hard sudoku puzzles. Dragon Hatchling beat more than 97% of the puzzles; several leading LLMs from the top labs failed to solve any. 

The point Pathway wants to make is that despite their remarkable success at many different tasks, there are still crucial classes of problems where LLMs fail. Sudoku is just one example. If we want LLMs to come up with genuine, novel solutions to real problems, we need to move beyond transformers, says Pathway’s cofounder and CEO, Zuzanna Stamirowska. 

That’s because transformers force LLMs to do everything with text. But language is not the best tool for certain kinds of reasoning. “It’s very difficult to represent a sudoku board word by word,” says Stamirowska.

Pathway’s solution is to change the math behind the transformer, replacing the attention mechanism with a mathematical structure called a state space. Instead of encoding information word by word, state spaces compress it into a more abstract representation. Using this technique, Dragon Hatchling can still process and produce text, but it can also mimic forms of reasoning that do not involve sequences of words. This not only makes Pathway’s model more efficient, but (in theory) it lets it take on tasks that other LLMs cannot do. 

Think of chess or mathematics—those kinds of puzzles are not held in your head as a long sentence, says Stamirowska: “The eureka moment that pops up in your brain isn’t necessarily in language. We would argue that if you have to reason in language, you’re somehow constrained.” 

Stamirowska admits that a mainstream LLM could read a book about how to solve sudoku and then write code to do it. But we want to build models with more than book smarts, she says: “The hope for AI is not to solve sudoku; it’s to cure cancer. There’s not a book for that.”

“Transformers are an engineering convenience that we fell on,” she adds. “It started a religion, but it’s silly to think that a breakthrough won’t happen again.”