MIT Technology Review Explains: Let our writers untangle the complex, messy world of technology to help you understand what’s coming next. You can read more from the series here.

I am writing this because one of my editors woke up in the middle of the night and scribbled on a bedside notepad: “What is a parameter?” Unlike a lot of thoughts that hit at 4 a.m., it’s a really good question—one that goes right to the heart of how large language models work. And I’m not just saying that because he’s my boss. (Hi, Boss!)

A large language model’s parameters are often said to be the dials and levers that control how it behaves. Think of a planet-size pinball machine that sends its balls pinging from one end to the other via billions of paddles and bumpers set just so. Tweak those settings and the balls will behave in a different way.  

OpenAI’s GPT-3, released in 2020, had 175 billion parameters. Google DeepMind’s latest LLM, Gemini 3, may have at least a trillion—some think it’s probably more like 7 trillion—but the company isn’t saying. (With competition now fierce, AI firms no longer share information about how their models are built.)

But the basics of what parameters are and how they make LLMs do the remarkable things that they do are the same across different models. Ever wondered what makes an LLM really tick—what’s behind the colorful pinball-machine metaphors? Let’s dive in.  

What is a parameter?

Think back to middle school algebra, like 2a + b. Those letters are parameters: Assign them values and you get a result. In math or coding, parameters are used to set limits or determine output. The parameters inside LLMs work in a similar way, just on a mind-boggling scale. 

How are they assigned their values?

Short answer: an algorithm. When a model is trained, each parameter is set to a random value. The training process then involves an iterative series of calculations (known as training steps) that update those values. In the early stages of training, a model will make errors. The training algorithm looks at each error and goes back through the model, tweaking the value of each of the model’s many parameters so that next time that error is smaller. This happens over and over again until the model behaves in the way its makers want it to. At that point, training stops and the values of the model’s parameters are fixed.

Sounds straightforward …

In theory! In practice, because LLMs are trained on so much data and contain so many parameters, training them requires a huge number of steps and an eye-watering amount of computation. During training, the 175 billion parameters inside a medium-size LLM like GPT-3 will each get updated tens of thousands of times. In total, that adds up to quadrillions (a number with 15 zeros) of individual calculations. That’s why training an LLM takes so much energy. We’re talking about thousands of specialized high-speed computers running nonstop for months.

Oof. What are all these parameters for, exactly?

There are three different types of parameters inside an LLM that get their values assigned through training: embeddings, weights, and biases. Let’s take each of those in turn.

Okay! So, what are embeddings?

An embedding is the mathematical representation of a word (or part of a word, known as a token) in an LLM’s vocabulary. An LLM’s vocabulary, which might contain up to a few hundred thousand unique tokens, is set by its designers before training starts. But there’s no meaning attached to those words. That comes during training.  

When a model is trained, each word in its vocabulary is assigned a numerical value that captures the meaning of that word in relation to all the other words, based on how the word appears in countless examples across the model’s training data.

Each word gets replaced by a kind of code?

Yeah. But there’s a bit more to it. The numerical value—the embedding—that represents each word is in fact a list of numbers, with each number in the list representing a different facet of meaning that the model has extracted from its training data. The length of this list of numbers is another thing that LLM designers can specify before an LLM is trained. A common size is 4,096.

Every word inside an LLM is represented by a list of 4,096 numbers?  

Yup, that’s an embedding. And each of those numbers is tweaked during training. An LLM with embeddings that are 4,096 numbers long is said to have 4,096 dimensions.

Why 4,096?

It might look like a strange number. But LLMs (like anything that runs on a computer chip) work best with powers of two—2, 4, 8, 16, 32, 64, and so on. LLM engineers have found that 4,096 is a power of two that hits a sweet spot between capability and efficiency. Models with fewer dimensions are less capable; models with more dimensions are too expensive or slow to train and run. 

Using more numbers allows the LLM to capture very fine-grained information about how a word is used in many different contexts, what subtle connotations it might have, how it relates to other words, and so on.

Back in February, OpenAI released GPT-4.5, the firm’s largest LLM yet (some estimates have put its parameter count at more than 10 trillion). Nick Ryder, a research scientist at OpenAI who worked on the model, told me at the time that bigger models can work with extra information, like emotional cues, such as when a speaker’s words signal hostility: “All of these subtle patterns that come through a human conversation—those are the bits that these larger and larger models will pick up on.”

The upshot is that all the words inside an LLM get encoded into a high-dimensional space. Picture thousands of words floating in the air around you. Words that are closer together have similar meanings. For example, “table” and “chair” will be closer to each other than they are to “astronaut,” which is close to “moon” and “Musk.” Way off in the distance you can see “prestidigitation.” It’s a little like that, but instead of being related to each other across three dimensions, the words inside an LLM are related across 4,096 dimensions.

Yikes.

It’s dizzying stuff. In effect, an LLM compresses the entire internet into a single monumental mathematical structure that encodes an unfathomable amount of interconnected information. It’s both why LLMs can do astonishing things and why they’re impossible to fully understand.    

Okay. So that’s embeddings. What about weights?

A weight is a parameter that represents the strength of a connection between different parts of a model—and one of the most common types of dial for tuning a model’s behavior. Weights are used when an LLM processes text.

When an LLM reads a sentence (or a book chapter), it first looks up the embeddings for all the words and then passes those embeddings through a series of neural networks, known as transformers, that are designed to process sequences of data (like text) all at once. Every word in the sentence gets processed in relation to every other word.

This is where weights come in. An embedding represents the meaning of a word without context. When a word appears in a specific sentence, transformers use weights to process the meaning of that word in that new context. (In practice, this involves multiplying each embedding by the weights for all other words.)

And biases?

Biases are another type of dial that complement the effects of the weights. Weights set the thresholds at which different parts of a model fire (and thus pass data on to the next part). Biases are used to adjust those thresholds so that an embedding can trigger activity even when its value is low. (Biases are values that are added to an embedding rather than multiplied with it.) 

By shifting the thresholds at which parts of a model fire, biases allow the model to pick up information that might otherwise be missed. Imagine you’re trying to hear what somebody is saying in a noisy room. Weights would amplify the loudest voices the most; biases are like a knob on a listening device that pushes quieter voices up in the mix. 

Here’s the TL;DR: Weights and biases are two different ways that an LLM extracts as much information as it can out of the text it is given. And both types of parameters are adjusted over and over again during training to make sure they do this. 

Okay. What about neurons? Are they a type of parameter too? 

No, neurons are more a way to organize all this math—containers for the weights and biases, strung together by a web of pathways between them. It’s all very loosely inspired by biological neurons inside animal brains, with signals from one neuron triggering new signals from the next and so on. 

Each neuron in a model holds a single bias and weights for every one of the model’s dimensions. In other words, if a model has 4,096 dimensions—and therefore its embeddings are lists of 4,096 numbers—then each of the neurons in that model will hold one bias and 4,096 weights. 

Neurons are arranged in layers. In most LLMs, each neuron in one layer is connected to every neuron in the layer above. A 175-billion-parameter model like GPT-3 might have around 100 layers with a few tens of thousands of neurons in each layer. And each neuron is running tens of thousands of computations at a time. 

Dizzy again. That’s a lot of math.

That’s a lot of math.

And how does all of that fit together? How does an LLM take a bunch of words and decide what words to give back?

When an LLM processes a piece of text, the numerical representation of that text—the embedding—gets passed through multiple layers of the model. In each layer, the value of the embedding (that list of 4,096 numbers) gets updated many times by a series of computations involving the model’s weights and biases (attached to the neurons) until it gets to the final layer.

The idea is that all the meaning and nuance and context of that input text is captured by the final value of the embedding after it has gone through a mind-boggling series of computations. That value is then used to calculate the next word that the LLM should spit out. 

It won’t be a surprise that this is more complicated than it sounds: The model in fact calculates, for every word in its vocabulary, how likely that word is to come next and ranks the results. It then picks the top word. (Kind of. See below …) 

That word is appended to the previous block of text, and the whole process repeats until the LLM calculates that the most likely next word to spit out is one that signals the end of its output. 

That’s it?  

Sure. Well …

Go on.

LLM designers can also specify a handful of other parameters, known as hyperparameters. The main ones are called temperature, top-p, and top-k.

You’re making this up.

Temperature is a parameter that acts as a kind of creativity dial. It influences the model’s choice of what word comes next. I just said that the model ranks the words in its vocabulary and picks the top one. But the temperature parameter can be used to push the model to choose the most probable next word, making its output more factual and relevant, or a less probable word, making the output more surprising and less robotic. 

Top-p and top-k are two more dials that control the model’s choice of next words. They are settings that force the model to pick a word at random from a pool of most probable words instead of the top word. These parameters affect how the model comes across—quirky and creative versus trustworthy and dull.   

One last question! There has been a lot of buzz about small models that can outperform big models. How does a small model do more with fewer parameters?

That’s one of the hottest questions in AI right now. There are a lot of different ways it can happen. Researchers have found that the amount of training data makes a huge difference. First you need to make sure the model sees enough data: An LLM trained on too little text won’t make the most of all its parameters, and a smaller model trained on the same amount of data could outperform it. 

Another trick researchers have hit on is overtraining. Showing models far more data than previously thought necessary seems to make them perform better. The result is that a small model trained on a lot of data can outperform a larger model trained on less data. Take Meta’s Llama LLMs. The 70-billion-parameter Llama 2 was trained on around 2 trillion words of text; the 8-billion-parameter Llama 3 was trained on around 15 trillion words of text. The far smaller Llama 3 is the better model. 

A third technique, known as distillation, uses a larger model to train a smaller one. The smaller model is trained not only on the raw training data but also on the outputs of the larger model’s internal computations. The idea is that the hard-won lessons encoded in the parameters of the larger model trickle down into the parameters of the smaller model, giving it a boost. 

In fact, the days of single monolithic models may be over. Even the largest models on the market, like OpenAI’s GPT-5 and Google DeepMind’s Gemini 3, can be thought of as several small models in a trench coat. Using a technique called “mixture of experts,” large models can turn on just the parts of themselves (the “experts”) that are required to process a specific piece of text. This combines the abilities of a large model with the speed and lower power consumption of a small one.

But that’s not the end of it. Researchers are still figuring out ways to get the most out of a model’s parameters. As the gains from straight-up scaling tail off, jacking up the number of parameters no longer seems to make the difference it once did. It’s not so much how many you have, but what you do with them.

Can I see one?

You want to see a parameter? Knock yourself out: Here’s an embedding.

MIT Technology Review Explains: Let our writers untangle the complex, messy world of technology to help you understand what’s coming next. You can read more from the series here.

In July, a widely cited MIT study claimed that 95% of organizations that invested in generative AI were getting “zero return.” Tech stocks briefly plunged. While the study itself was more nuanced than the headlines, for many it still felt like the first hard data point confirming what skeptics had muttered for months: Hype around AI might be outpacing reality.

Then, in August, OpenAI CEO Sam Altman said what everyone in Silicon Valley had been whispering. “Are we in a phase where investors as a whole are overexcited about AI?” he said during a press dinner I attended. “My opinion is yes.” 

---

This story is part of MIT Technology Review’s Hype Correction package, a series that resets expectations about what AI is, what it makes possible, and where we go next.

---

He compared the current moment to the dot-com bubble. “When bubbles happen, smart people get overexcited about a kernel of truth,” he explained. “Tech was really important. The internet was a really big deal. People got overexcited.” 

With those comments, it was off to the races. The next day’s stock market dip was attributed to the sentiment he shared. The question “Are we in an AI bubble?” became inescapable.

Who thinks it is a bubble? 

The short answer: Lots of people. But not everyone agrees on who or what is overinflated. Tech leaders are using this moment of fear to take shots at their rivals and position themselves as clear winners on the other side. How they describe the bubble depends on where their company sits.

When I asked Meta CEO Mark Zuckerberg about the AI bubble in September, he ran through the historical analogies of past bubbles—railroads, fiber for the internet, the dot-com boom—and noted that in each case, “the infrastructure gets built out, people take on too much debt, and then you hit some blip … and then a lot of the companies end up going out of business.”

But Zuckerberg’s prescription wasn’t for Meta to pump the brakes. It was to keep spending: “If we end up misspending a couple of hundred billion dollars, I think that that is going to be very unfortunate, obviously. But I’d say the risk is higher on the other side.”

Bret Taylor, the chairman of OpenAI and CEO of the AI startup Sierra, uses a mental model from the late ’90s to help navigate this AI bubble. “I think the closest analogue to this AI wave is the dot-com boom or bubble, depending on your level of pessimism,” he recently told me. Back then, he explained, everyone knew e-commerce was going to be big, but there was a massive difference between Buy.com and Amazon. Taylor and others have been trying to position themselves as today’s Amazon.

Still others are arguing that the pain will be widespread. Google CEO Sundar Pichai told the BBC this month that there’s “some irrationality” in the current boom. Asked whether Google would be immune to a bubble bursting, he warned, “I think no company is going to be immune, including us.”

What’s inflating the bubble?

Companies are raising enormous sums of money and seeing unprecedented valuations. Much of that money, in turn, is going toward the buildout of massive data centers—on which both private companies like OpenAI and Elon Musk’s xAI and public ones such as Meta and Google are spending heavily. OpenAI has pledged that it will spend $500 billion to build AI data centers, more than 15 times what was spent on the Manhattan Project.

This eye-popping spending on AI data centers isn’t entirely detached from reality. The leaders of the top AI companies all stress that they’re bottlenecked by their limited access to computing power. You hear it constantly when you talk to them. Startups can’t get the GPU allocations they need. Hyperscalers are rationing compute, saving it for their best customers.

If today’s AI market is as brutally supply-constrained as tech leaders claim, perhaps aggressive infrastructure buildouts are warranted. But some of the numbers are too large to comprehend. Sam Altman has told employees that OpenAI’s moonshot goal is to build 250 gigawatts of computing capacity by 2033, roughly equaling India’s total national electricity demand. Such a plan would cost more than $12 trillion by today’s standards.

“I do think there’s real execution risk,” OpenAI president and cofounder Greg Brockman recently told me about the company’s aggressive infrastructure goals. “Everything we say about the future, we see that it’s a possibility. It is not a certainty, but I don’t think the uncertainty comes from scientific questions. It’s a lot of hard work.”

Who is exposed, and who is to blame?

It depends on who you ask. During the August press dinner, where he made his market-moving comments, Altman was blunt about where he sees the excess. He said it’s “insane” that some AI startups with “three people and an idea” are receiving funding at such high valuations. “That’s not rational behavior,” he said. “Someone’s gonna get burned there, I think.” As Safe Superintelligence cofounder (and former OpenAI chief scientist and cofounder) Ilya Sutskever put it on a recent podcast: Silicon Valley has “more companies than ideas.”

Demis Hassabis, the CEO of Google DeepMind, offered a similar diagnosis when I spoke with him in November. “It feels like there’s obviously a bubble in the private market,” he said. “You look at seed rounds with just nothing being tens of billions of dollars. That seems a little unsustainable.”

Anthropic CEO Dario Amodei also struck at his competition during the New York Times DealBook Summit in early December. He said he feels confident about the technology itself but worries about how others are behaving on the business side: “On the economic side, I have my concerns where, even if the technology fulfills all its promises, I think there are players in the ecosystem who, if they just make a timing error, they just get it off by a little bit, bad things could happen.”

He stopped short of naming Sam Altman and OpenAI, but the implication was clear. “There are some players who are YOLOing,” he said. “Let’s say you’re a person who just kind of constitutionally wants to YOLO things or just likes big numbers. Then you may turn the dial too far.”

Amodei also flagged “circular deals,” or the increasingly common arrangements where chip suppliers like Nvidia invest in AI companies that then turn around and spend those funds on their chips. Anthropic has done some of these, he said, though “not at the same scale as some other players.” (OpenAI is at the center of a number of such deals, as are Nvidia, CoreWeave, and a roster of other players.) 

The danger, he explained, comes when the numbers get too big: “If you start stacking these where they get to huge amounts of money, and you’re saying, ’By 2027 or 2028 I need to make $200 billion a year,’ then yeah, you can overextend yourself.”

Zuckerberg shared a similar message at an internal employee Q&A session after Meta’s last earnings call. He noted that unprofitable startups like OpenAI and Anthropic risk bankruptcy if they misjudge the timing of their investments, but Meta has the advantage of strong cash flow, he reassured staff.

How could a bubble burst?

My conversations with tech executives and investors suggest that the bubble will be most likely to pop if overfunded startups can’t turn a profit or grow into their lofty valuations. This bubble could last longer than than past ones, given that private markets aren’t traded on public markets and therefore move more slowly, but the ripple effects will still be profound when the end comes. 

If companies making grand commitments to data center buildouts no longer have the revenue growth to support them, the headline deals that have propped up the stock market come into question. Anthropic’s Amodei illustrated the problem during his DealBook Summit appearance, where he said the multi-year data center commitments he has to make combine with the company’s rapid, unpredictable revenue growth rate to create a “cone of uncertainty” about how much to spend.

The two most prominent private players in AI, OpenAI and Anthropic, have yet to turn a profit. A recent Deutsche Bank chart put the situation in stark historical context. Amazon burned through $3 billion before becoming profitable. Tesla, around $4 billion. Uber, $30 billion. OpenAI is projected to burn through $140 billion by 2029, while Anthropic is expected to burn $20 billion by 2027.

Consultants at Bain estimate that the wave of AI infrastructure spending will require $2 trillion in annual AI revenue by 2030 just to justify the investment. That’s more than the combined 2024 revenue of Amazon, Apple, Alphabet, Microsoft, Meta, and Nvidia. When I talk to leaders of these large tech companies, they all agree that their sprawling businesses can absorb an expensive miscalculation about the returns from their AI infrastructure buildouts. It’s all the other companies that are either highly leveraged with debt or just unprofitable—even OpenAI and Anthropic—that they worry about. 

Still, given the level of spending on AI, it still needs a viable business model beyond subscriptions, which won’t be able to  drive profits from billions of people’s eyeballs like the ad-driven businesses that have defined the last 20 years of the internet. Even the largest tech companies know they need to ship the world-changing agents they keep hyping: AI that can fully replace coworkers and complete tasks in the real world.

For now, investors are mostly buying into the hype of the powerful AI systems that these data center buildouts will supposedly unlock in the future. At some point the biggest spenders, like OpenAI, will need to show investors that the money spent on the infrastructure buildout was worth it.

There’s also still a lot of uncertainty about the technical direction that AI is heading in. LLMs are expected to remain critical to more advanced AI systems, but industry leaders can’t seem to agree on which additional breakthroughs are needed to achieve artificial general intelligence, or AGI. Some are betting on new kinds of AI that can understand the physical world, while others are focused on training AI to learn in a general way, like a human. In other words, what if all this unprecedented spending turns out to have been backing the wrong horse?

The question now

What makes this moment surreal is the honesty. The same people pouring billions into AI will openly tell you it might all come crashing down. 

Taylor framed it as two truths existing at once. “I think it is both true that AI will transform the economy,” he told me, “and I think we’re also in a bubble, and a lot of people will lose a lot of money. I think both are absolutely true at the same time.”

He compared it to the internet. Webvan failed, but Instacart succeeded years later with essentially the same idea. If you were an Amazon shareholder from its IPO to now, you’re looking pretty good. If you were a Webvan shareholder, you probably feel differently. 

“When the dust settles and you see who the winners are, society benefits from those inventions,” Amazon founder Jeff Bezos said in October. “This is real. The benefit to society from AI is going to be gigantic.”

Goldman Sachs says the AI boom now looks the way tech stocks did in 1997, several years before the dot-com bubble actually burst. The bank flagged five warning signs seen in the late 1990s that investors should watch now: peak investment spending, falling corporate profits, rising corporate debt, Fed rate cuts, and widening credit spreads. We’re probably not at 1999 levels yet. But the imbalances are building fast. Michael Burry, who famously called the 2008 housing bubble collapse (as seen in the film The Big Short), recently compared the AI boom to the 1990s dot-com bubble too.

Maybe AI will save us from our own irrational exuberance. But for now, we’re living in an in-between moment when everyone knows what’s coming but keeps blowing more air into the balloon anyway. As Altman put it that night at dinner: “Someone is going to lose a phenomenal amount of money. We don’t know who.”

Alex Heath is the author of Sources, a newsletter about the AI race, and the cohost of ACCESS, a podcast about the tech industry’s inside conversations. Previously, he was deputy editor at The Verge.

MIT Technology Review Explains: Let our writers untangle the complex, messy world of technology to help you understand what’s coming next. You can read more from the series here.

Weight-loss drugs have been back in the news this week. First, we heard that Eli Lilly, the company behind the drugs Mounjaro and Zepbound, became the first healthcare company in the world to achieve a trillion-dollar valuation.

Those two drugs, which are prescribed for diabetes and obesity respectively, are generating billions of dollars in revenue for the company. Other GLP-1 agonist drugs—a class that includes Mounjaro and Zepbound, which have the same active ingredient—have also been approved to reduce the risk of heart attack and stroke in overweight people. Many hope these apparent wonder drugs will also treat neurological disorders and potentially substance use disorders, too.

But this week we also learned that, disappointingly, GLP-1 drugs don’t seem to help people with Alzheimer’s disease. And that people who stop taking the drugs when they become pregnant can experience potentially dangerous levels of weight gain during their pregnancies. On top of that, some researchers worry that people are using the drugs postpartum to lose pregnancy weight without understanding potential risks.

All of this news should serve as a reminder that there’s a lot we still don’t know about these drugs. This week, let’s look at the enduring questions surrounding GLP-1 agonist drugs.

First a quick recap. Glucagon-like peptide-1 is a hormone made in the gut that helps regulate blood sugar levels. But we’ve learned that it also appears to have effects across the body. Receptors that GLP-1 can bind to have been found in multiple organs and throughout the brain, says Daniel Drucker, an endocrinologist at the University of Toronto who has been studying the hormone for decades.

GLP-1 agonist drugs essentially mimic the hormone’s action. Quite a few have been developed, including semaglutide, tirzepatide, liraglutide, and exenatide, which have brand names like Ozempic, Saxenda and Wegovy. Some of them are recommended for some people with diabetes.

But because these drugs also seem to suppress appetite, they have become hugely popular weight loss aids. And studies have found that many people who take them for diabetes or weight loss experience surprising side effects; that their mental health improves, for example, or that they feel less inclined to smoke or consume alcohol. Research has also found that the drugs seem to increase the growth of brain cells in lab animals.

So far, so promising. But there are a few outstanding gray areas.

Are they good for our brains?

Novo Nordisk, a competitor of Eli Lilly, manufactures GLP-1 drugs Wegovy and Saxenda. The company recently trialed an oral semaglutide in people with Alzheimer’s disease who had mild cognitive impairment or mild dementia. The placebo-controlled trial included 3808 volunteers.

Unfortunately, the company found that the drug did not appear to delay the progression of Alzheimer’s disease in the volunteers who took it.

The news came as a huge disappointment to the research community. “It was kind of crushing,” says Drucker. That’s despite the fact that, deep down, he wasn’t expecting a “clear win.” Alzheimer’s disease has proven notoriously difficult to treat, and by the time people get a diagnosis, a lot of damage has already taken place.

But he is one of many that isn’t giving up hope entirely. After all, research suggests that GLP-1 reduces inflammation in the brain and improves the health of neurons, and that it appears to improve the way brain regions communicate with each other. This all implies that GLP-1 drugs should benefit the brain, says Drucker. There’s still a chance that the drugs might help stave off Alzheimer’s in those who are still cognitively healthy.

Are they safe before, during or after pregnancy?

Other research published this week raises questions about the effects of GLP-1s taken around the time of pregnancy. At the moment, people are advised to plan to stop taking the medicines two months before they become pregnant. That’s partly because some animal studies suggest the drugs can harm the development of a fetus, but mainly because scientists haven’t studied the impact on pregnancy in humans.

Among the broader population, research suggests that many people who take GLP-1s for weight loss regain much of their lost weight once they stop taking those drugs. So perhaps it’s not surprising that a study published in JAMA earlier this week saw a similar effect in pregnant people.

The study found that people who had been taking those drugs gained around 3.3kg more than others who had not. And those who had been taking the drugs also appeared to have a slightly higher risk of gestational diabetes, blood pressure disorders and even preterm birth.

It sounds pretty worrying. But a different study published in August had the opposite finding—it noted a reduction in the risk of those outcomes among women who had taken the drugs before becoming pregnant.

If you’re wondering how to make sense of all this, you’re not the only one. No one really knows how these drugs should be used before pregnancy—or during it for that matter.

Another study out this week found that people (in Denmark) are increasingly taking GLP-1s postpartum to lose weight gained during pregnancy. Drucker tells me that, anecdotally, he gets asked about this potential use a lot.

But there’s a lot going on in a postpartum body. It’s a time of huge physical and hormonal change that can include bonding, breastfeeding and even a rewiring of the brain. We have no idea if, or how, GLP-1s might affect any of those.

How—and when—can people safely stop using them?

Yet another study out this week—you can tell GLP-1s are one of the hottest topics in medicine right now—looked at what happens when people stop taking tirzepatide (marketed as Zepbound) for their obesity.

The trial participants all took the drug for 36 weeks, at which point half continued with the drug, and half were switched to a placebo for another 52 weeks. During that first 36 weeks, the weight and heart health of the participants improved.

But by the end of the study, most of those that had switched to a placebo had regained more than 25% of the weight they had originally lost. One in four had regained more than 75% of that weight, and 9% ended up at a higher weight than when they’d started the study. Their heart health also worsened.

Does that mean that people need to take these drugs forever? Scientists don’t have the answer to that one, either. Or if taking the drugs indefinitely is safe. The answer might depend on the individual, their age or health status, or what they are using the drug for.

There are other gray areas. GLP-1s look promising for substance use disorders, but we don’t yet know how effective they might be. We don’t know the long-term effects these drugs have on children who take them. And we don’t know the long-term consequences these drugs might have for healthy-weight people who take them for weight loss.

Earlier this year, Drucker accepted a Breakthrough Prize in Life Sciences at a glitzy event in California. “All of these Hollywood celebrities were coming up to me and saying ‘thank you so much,’” he says. “A lot of these people don’t need to be on these medicines.”

This article first appeared in The Checkup, MIT Technology Review’s weekly biotech newsletter. To receive it in your inbox every Thursday, and read articles like this first, sign up here.

MIT Technology Review Explains: Let our writers untangle the complex, messy world of technology to help you understand what’s coming next. You can read more from the series here.

In mid-October, a mysterious object cracked the windshield of a packed Boeing 737 cruising at 36,000 feet above Utah, forcing the pilots into an emergency landing. The internet was suddenly buzzing with the prospect that the plane had been hit by a piece of space debris. We still don’t know exactly what hit the plane—likely a remnant of a weather balloon—but it turns out the speculation online wasn’t that far-fetched.

That’s because while the risk of flights being hit by space junk is still small, it is, in fact, growing. 

About three pieces of old space equipment—used rockets and defunct satellites—fall into Earth’s atmosphere every day, according to estimates by the European Space Agency. By the mid-2030s, there may be dozens. The increase is linked to the growth in the number of satellites in orbit. Currently, around 12,900 active satellites circle the planet. In a decade, there may be 100,000 of them, according to analyst estimates.

To minimize the risk of orbital collisions, operators guide old satellites to burn up in Earth’s atmosphere. But the physics of that reentry process are not well understood, and we don’t know how much material burns up and how much reaches the ground.

“The number of such landfall events is increasing,” says Richard Ocaya, a professor of physics at the University of Free State in South Africa and a coauthor of a recent paper on space debris risk. “We expect it may be increasing exponentially in the next few years.”

So far, space debris hasn’t injured anybody—in the air or on the ground. But multiple close calls have been reported in recent years. In March last year, an 0.7-kilogram chunk of metal pierced the roof of a house in Florida. The object was later confirmed to be a remnant of a battery pallet tossed out from the International Space Station. When the strike occurred, the homeowner’s 19-year-old son was resting in a next-door room.

And in February this year, a 1.5-meter-long fragment of SpaceX’s Falcon 9 rocket crashed down near a warehouse outside Poland’s fifth-largest city, Poznan. Another piece was found in a nearby forest. A month later, a 2.5-kilogram piece of a Starlink satellite dropped on a farm in the Canadian province of Saskatchewan. Other incidents have been reported in Australia and Africa. And many more may be going completely unnoticed. 

“If you were to find a bunch of burnt electronics in a forest somewhere, your first thought is not that it came from a spaceship,” says James Beck, the director of the UK-based space engineering research firm Belstead Research. He warns that we don’t fully understand the risk of space debris strikes and that it might be much higher than satellite operators want us to believe. 

For example, SpaceX, the owner of the currently largest mega-constellation, Starlink, claims that its satellites are “designed for demise” and completely burn up when they spiral from orbit and fall through the atmosphere.

But Beck, who has performed multiple wind tunnel tests using satellite mock-ups to mimic atmospheric forces, says the results of such experiments raise doubts. Some satellite components are made of durable materials such as titanium and special alloy composites that don’t melt even at the extremely high temperatures that arise during a hypersonic atmospheric descent. 

“We have done some work for some small-satellite manufacturers and basically, their major problem is that the tanks get down,” Beck says. “For larger satellites, around 800 kilos, we would expect maybe two or three objects to land.” 

It can be challenging to quantify how much of a danger space debris poses. The International Civil Aviation Organization (ICAO) told MIT Technology Review that “the rapid growth in satellite deployments presents a novel challenge” for aviation safety, one that “cannot be quantified with the same precision as more established hazards.” 

But the Federal Aviation Administration has calculated some preliminary numbers on the risk to flights: In a 2023 analysis, the agency estimated that by 2035, the risk that one plane per year will experience a disastrous space debris strike will be around 7 in 10,000. Such a collision would either destroy the aircraft immediately or lead to a rapid loss of air pressure, threatening the lives of all on board.

The casualty risk to humans on the ground will be much higher. Aaron Boley, an associate professor in astronomy and a space debris researcher at the University of British Columbia, Canada, says that if megaconstellation satellites “don’t demise entirely,” the risk of a single human death or injury caused by a space debris strike on the ground could reach around 10% per year by 2035. That would mean a better than even chance that someone on Earth would be hit by space junk about every decade. In its report, the FAA put the chances even higher with similar assumptions, estimating that “one person on the planet would be expected to be injured or killed every two years.”

Experts are starting to think about how they might incorporate space debris into their air safety processes. The German space situational awareness company Okapi Orbits, for example, in cooperation with the German Aerospace Center and the European Organization for the Safety of Air Navigation (Eurocontrol), is exploring ways to adapt air traffic control systems so that pilots and air traffic controllers can receive timely and accurate alerts about space debris threats.

But predicting the path of space debris is challenging too. In recent years, advances in AI have helped improve predictions of space objects’ trajectories in the vacuum of space, potentially reducing the risk of orbital collisions. But so far, these algorithms can’t properly account for the effects of the gradually thickening atmosphere that space junk encounters during reentry. Radar and telescope observations can help, but the exact location of the impact becomes clear with only very short notice.

“Even with high-fidelity models, there’s so many variables at play that having a very accurate reentry location is difficult,” says Njord Eggen, a data analyst at Okapi Orbits. Space debris goes around the planet every hour and a half when in low Earth orbit, he notes, “so even if you have uncertainties on the order of 10 minutes, that’s going to have drastic consequences when it comes to the location where it could impact.”

For aviation companies, the problem is not just a potential strike, as catastrophic as that would be. To avoid accidents, authorities are likely to temporarily close the airspace in at-risk regions, which creates delays and costs money. Boley and his colleagues published a paper earlier this year estimating that busy aerospace regions such as northern Europe or the northeastern United States already have about a 26% yearly chance of experiencing at least one disruption due to the reentry of a major space debris item. By the time all planned constellations are fully deployed, aerospace closures due to space debris hazards may become nearly as common as those due to bad weather.

Because current reentry predictions are unreliable, many of these closures may end up being unnecessary.

For example, when a 21-metric-ton Chinese Long March mega-rocket was falling to Earth in 2022, predictions suggested its debris could scatter across Spain and parts of France. In the end, the rocket crashed into the Pacific Ocean. But the 30-minute closure of south European airspace delayed and diverted hundreds of flights. 

In the meantime, international regulators are urging satellite operators and launch providers to deorbit large satellites and rocket bodies in a controlled way, when possible, by carefully guiding them into remote parts of the ocean using residual fuel. 

The European Space Agency estimates that only about half the rocket bodies reentering the atmosphere do so in a controlled way. 

Moreover, around 2,300 old and no-longer-controllable rocket bodies still linger in orbit, slowly spiraling toward Earth with no mechanisms for operators to safely guide them into the ocean.

“There’s enough material up there that even if we change our practices, we will still have all those rocket bodies eventually reenter,” Boley says. “Although the probability of space debris hitting an aircraft is small, the probability that the debris will spread and fall over busy airspace is not small. That’s actually quite likely.”