Category: Science & Tech

Over time, Trump’s tariffs may indeed compel companies to bring more of their manufacturing operations back to the US and help diversify the global supply chain for crucial goods, UC San Diego’s Victor says.

The tariffs are likely to fuel more mining and processing of critical minerals like lithium and nickel in the US, too, given both the increased costs on imported materials and the administration’s plans to roll back environmental and permitting rules. 

“They love extractive sectors,” says Jonas Nahm, an associate professor at the Johns Hopkins School of Advanced International Studies.

But the “big concern” is that Trump’s plans to boost tariffs, cut government spending, and enact other policy changes could stall the broader economy, says Rachel Slaybaugh, a partner at DCVC, a San Francisco venture firm.

Indeed, the combined effects of Trump’s proposals, including his pledge to deport hundreds of thousands to millions of workers, may drive up US inflation more than 4% by 2026 while cutting gross domestic product by at least 1.3%, according to an analysis by the Peterson Institute for International Economics, a nonpartisan research firm in Washington, DC. 

The tariffs alone could cost typical households an extra $2,600 per year. They may also trigger retaliatory measures by other nations, including China, which could impose their own steeper fees on US products or cut off the flow of crucial goods.

Slaybaugh expects to see a continued slowdown in venture investments into cleantech companies in the coming months, as investors wait to see how aggressively the Trump administration implements the various pledges he made on the campaign trail. That pause alone will make it harder for startups to secure the capital they need to scale up or sustain operations. 

Even if the tariffs do eventually push US businesses to produce more of the goods currently being delivered cheaply and efficiently from elsewhere, it leaves a big problem when it comes to the clean energy transition: Given the higher expenses of US labor, land, and materials, it will simply cost far, far more to build the modern, low-emissions energy and transportation systems the nation now needs, Nahm says. 

At this point, after China has spent decades and vast sums locking down global supply chains, scaling up production, and driving down manufacturing costs, it’s foolhardy to believe that US businesses can easily step in and crank out these essential goods in relative global isolation, as Victor and his colleague, Michael Davidson, argued in a recent Brookings essay. 

“Collaboration and competition, not hostility, are how we can catch up to the world’s largest supplier of clean technology products,” they wrote. 

In that first trial, Brännström’s team transplanted uteruses into nine women, each of whom had IVF to create and store embryos beforehand. The woman who was the first to give birth had IVF over a 12-month period, which ended six months before her surgery. It took a little over 10 hours to remove the uterus from the donor, and just under five hours to stitch it into the recipient.

The recipient started getting her period 43 days after her transplant. Doctors transferred one of her embryos into the uterus a year after her surgery. Three weeks later, a pregnancy test confirmed she was pregnant.

At 31 weeks, she was admitted to hospital with preeclampsia, a serious medical condition that can develop during pregnancy, and her baby was delivered by C-section 16 hours later. She was discharged from hospital after three days, although the baby spent 16 days in the hospital’s neonatal unit.

Despite those difficulties, her story is considered a success. Other uterus recipients have also experienced pregnancy complications, and some have had surgical complications. And all transplant recipients must adhere to a regimen of immunosuppressant drugs, which can have side effects.

The uteruses aren’t intended to last forever, either. Surgeons remove them once the recipients have completed their families, often after one or two children. The removal is also a significant operation.

Given all that, uterus transplants are not to be taken lightly. And there are other paths to parenthood. Some ethicists are concerned that in pursuing uterus transplantation as a fertility treatment, we might reinforce ideas that define a woman’s value in terms of her reproductive potential, Natasha Hammond-Browning, a legal scholar at Cardiff University in Wales, said at the event. “There is debate around whether we should be giving greater preference to adoption, to surrogacy, and to supporting children who already exist and who need care,” she said.

We also need to consider whether there is a “right to gestate,” and if there is, who has that right, said Hammond-Browning. And these concerns need to be balanced with the importance of reproductive autonomy—the idea that people have the right to decide and control their own reproductive efforts.

Further questions remain over whether uterus transplants might ever be an option for trans women, who not only lack a uterus but also have a different pelvic anatomy. I asked the speakers if the surgery might ever be feasible. They weren’t hugely optimistic that it would, at least in the near future.

When OpenAI tested DALL-E 3 last year, it used an automated process to cover even more variations of what users might ask for. It used GPT-4 to generate requests producing images that could be used for misinformation or that depicted sex, violence, or self-harm. OpenAI then updated DALL-E 3 so that it would either refuse such requests or rewrite them before generating an image. Ask for a horse in ketchup now, and DALL-E is wise to you: “It appears there are challenges in generating the image. Would you like me to try a different request or explore another idea?”

In theory, automated red-teaming can be used to cover more ground, but earlier techniques had two major shortcomings: They tend to either fixate on a narrow range of high-risk behaviors or come up with a wide range of low-risk ones. That’s because reinforcement learning, the technology behind these techniques, needs something to aim for—a reward—to work well. Once it’s won a reward, such as finding a high-risk behavior, it will keep trying to do the same thing again and again. Without a reward, on the other hand, the results are scattershot. 

“They kind of collapse into ‘We found a thing that works! We’ll keep giving that answer!’ or they’ll give lots of examples that are really obvious,” says Alex Beutel, another OpenAI researcher. “How do we get examples that are both diverse and effective?”

A problem of two parts

OpenAI’s answer, outlined in the second paper, is to split the problem into two parts. Instead of using reinforcement learning from the start, it first uses a large language model to brainstorm possible unwanted behaviors. Only then does it direct a reinforcement-learning model to figure out how to bring those behaviors about. This gives the model a wide range of specific things to aim for. 

Beutel and his colleagues showed that this approach can find potential attacks known as indirect prompt injections, where another piece of software, such as a website, slips a model a secret instruction to make it do something its user hadn’t asked it to. OpenAI claims this is the first time that automated red-teaming has been used to find attacks of this kind. “They don’t necessarily look like flagrantly bad things,” says Beutel.

Will such testing procedures ever be enough? Ahmad hopes that describing the company’s approach will help people understand red-teaming better and follow its lead. “OpenAI shouldn’t be the only one doing red-teaming,” she says. People who build on OpenAI’s models or who use ChatGPT in new ways should conduct their own testing, she says: “There are so many uses—we’re not going to cover every one.”

For some, that’s the whole problem. Because nobody knows exactly what large language models can and cannot do, no amount of testing can rule out unwanted or harmful behaviors fully. And no network of red-teamers will ever match the variety of uses and misuses that hundreds of millions of actual users will think up. 

That’s especially true when these models are run in new settings. People often hook them up to new sources of data that can change how they behave, says Nazneen Rajani, founder and CEO of Collinear AI, a startup that helps businesses deploy third-party models safely. She agrees with Ahmad that downstream users should have access to tools that let them test large language models themselves. 

This is a comment I get all the time on the topic of climate change, both in conversations and on whatever social media site is currently en vogue. Usually, it comes in response to some statement about how the US and Europe are addressing the issue (or how they need to be).

Sometimes I think people ask this in bad faith. It’s a rhetorical way to throw up your hands, imply that the US and Europe aren’t the real problem, and essentially say: “if they aren’t taking responsibility, why should we?” However, amid the playground-esque finger-pointing there are some undeniable facts: China emits more greenhouse gases than any other country, by far. It’s one of the world’s most populous countries and a climate-tech powerhouse, and its economy is still developing. 

With many complicated factors at play, how should we think about the country’s role in addressing climate change?

China’s emissions are the highest in the world, topping 12 billion tons of carbon dioxide in 2023, according to the International Energy Agency.

There’s context missing if we just look at that one number, as I wrote in my latest story that digs into recent global climate data. Since carbon dioxide hangs around in the atmosphere for centuries, we should arguably consider not just a country’s current emissions, but everything it’s produced over time. If we do that, the US still takes the crown for the world’s biggest climate polluter.

However, China is now in second place, according to a new analysis from Carbon Brief released this week. In 2023, the country exceeded the EU’s 27 member states in historical emissions for the first time.

This reflects a wider trend that we’re seeing around the world: Developing nations are starting to account for a larger fraction of emissions than they used to. In 1992, when countries agreed to the UN climate convention, industrialized countries (a category called Annex I) made up about one-fifth of the world’s population but were responsible for a whopping 61% of historical emissions. By the end of 2024, though, those countries’ share of global historical emissions will fall to 52%, and it is expected to keep ticking down.

China, like all nations, will need to slash its emissions for the world to meet global climate goals. One crucial point here is that while its emissions are still huge, there are signs that the nation is making some progress. 

Artists and writers have launched several lawsuits against AI companies, arguing that their work has been scraped into databases for training AI models without consent or compensation. Tech companies have responded that anything on the public internet falls under fair use. But it will be years until we have a legal resolution to the problem. 

Unfortunately, there is little you can do if your work has been scraped into a data set and used in a model that is already out there. You can, however, take steps to prevent your work from being used in the future. 

Here are four ways to do that. 

Mask your style 

One of the most popular ways artists are fighting back against AI scraping is by applying “masks” on their images, which protect their personal style from being copied. 

Tools such as Mist, Anti-DreamBooth, and Glaze add tiny changes to an image’s pixels that are invisible to the human eye, so that if and when images are scraped, machine-learning models cannot decipher them properly. You’ll need some coding skills to run Mist and Anti-DreamBooth, but Glaze, developed by researchers at the University of Chicago, is more straightforward to apply. The tool is free and available to download as an app, or the protection can be applied online. Unsurprisingly, it is the most popular tool and has been downloaded millions of times. 

But defenses like these are never foolproof, and what works today might not work tomorrow. In computer security, breaking defenses is standard practice among researchers, as this helps people find weaknesses and make systems safer. Using these tools is a calculated risk: Once something is uploaded online, you lose control of it and can’t retroactively add protections to images. 

Rethink where and how you share 

Popular art profile sites such as DeviantArt and Flickr have become gold mines for AI companies searching for training data. And when you share images on platforms such as Instagram, its parent company, Meta, can use your data to build its models in perpetuity if you’ve shared it publicly. (See opt-outs below.) 

One way to prevent scraping is by not sharing images online publicly, or by making your social media profiles private. But for many creatives that is simply not an option; sharing work online is a crucial way to attract clients. 

Led by Joon Sung Park, a Stanford PhD student in computer science, the team recruited 1,000 people who varied by age, gender, race, region, education, and political ideology. They were paid up to $100 for their participation. From interviews with them, the team created agent replicas of those individuals. As a test of how well the agents mimicked their human counterparts, participants did a series of personality tests, social surveys, and logic games, twice each, two weeks apart; then the agents completed the same exercises. The results were 85% similar. 

“If you can have a bunch of small ‘yous’ running around and actually making the decisions that you would have made—that, I think, is ultimately the future,” Joon says. 

In the paper the replicas are called simulation agents, and the impetus for creating them is to make it easier for researchers in social sciences and other fields to conduct studies that would be expensive, impractical, or unethical to do with real human subjects. If you can create AI models that behave like real people, the thinking goes, you can use them to test everything from how well interventions on social media combat misinformation to what behaviors cause traffic jams. 

Such simulation agents are slightly different from the agents that are dominating the work of leading AI companies today. Called tool-based agents, those are models built to do things for you, not converse with you. For example, they might enter data, retrieve information you have stored somewhere, or—someday—book travel for you and schedule appointments. Salesforce announced its own tool-based agents in September, followed by Anthropic in October, and OpenAI is planning to release some in January, according to Bloomberg. 

The two types of agents are different but share common ground. Research on simulation agents, like the ones in this paper, is likely to lead to stronger AI agents overall, says John Horton, an associate professor of information technologies at the MIT Sloan School of Management, who founded a company to conduct research using AI-simulated participants. 

“This paper is showing how you can do a kind of hybrid: use real humans to generate personas which can then be used programmatically/in-simulation in ways you could not with real humans,” he told MIT Technology Review in an email. 

The research comes with caveats, not the least of which is the danger that it points to. Just as image generation technology has made it easy to create harmful deepfakes of people without their consent, any agent generation technology raises questions about the ease with which people can build tools to personify others online, saying or authorizing things they didn’t intend to say. 

The evaluation methods the team used to test how well the AI agents replicated their corresponding humans were also fairly basic. These included the General Social Survey—which collects information on one’s demographics, happiness, behaviors, and more—and assessments of the Big Five personality traits: openness to experience, conscientiousness, extroversion, agreeableness, and neuroticism. Such tests are commonly used in social science research but don’t pretend to capture all the unique details that make us ourselves. The AI agents were also worse at replicating the humans in behavioral tests like the “dictator game,” which is meant to illuminate how participants consider values such as fairness. 

Even then, though, there’s another factor to consider: population. Dividing a country’s total emissions by its population reveals how the average individual in each nation is contributing to climate change today. 

Countries with smaller populations and economies that are heavily reliant on oil and gas tend to top this list, including Saudi Arabia, Bahrain, and the United Arab Emirates.

Among the larger nations, Australia has the highest per capita emissions from fossil fuels, with the US and Canada close behind. Meanwhile, other countries that have high total emissions are farther down the list when normalized by population: China’s per capita emissions are just over half that of the US, while India’s is a small fraction.

Understanding the complicated picture of global emissions is crucial, especially during ongoing negotiations (including the current meeting at COP29 in Baku, Azerbaijan) over how to help developing nations pay for efforts to combat climate change. 

Looking at current emissions, one might expect the biggest emitter, China, to contribute more than any other country to climate finance. But considering historical contributions, per capita emissions, and details about national economies, other nations like the US, UK, and members of the EU emerge as those experts tend to say should feature prominently in the talks. 

What is clear is that when it comes to the emissions blame game, it’s more complicated than just pointing at today’s biggest polluters. Ultimately, addressing climate change will require everyone to get on board—we all share an atmosphere, and we’re all going to continue feeling the effects of a changing climate. 

Notes on data methodology: 

• Emissions data is from the Global Carbon Project, which estimates carbon emissions based on energy use. Territorial emissions take into account energy and some industry, but don’t include land use emissions. 
• Data from the European Union is the sum of its current 27 member states. The bloc is represented together because the EU generally negotiates together on the international stage. 
• Historical emissions for some countries are disaggregated from former borders, including the former USSR and Yugoslavia. 
• The per capita emissions map uses official World Bank boundaries, with the exception of Taiwan, which has separate emissions data in the Global Carbon Project. 
• Western Sahara’s energy data are reported by Morocco, so its emissions are included in that total. Per capita emissions for Morocco are also used for Western Sahara on the map. 
• More detailed information about the Global Carbon Project methods (including the particulars on how territorial emissions are broken down) is available here.