via MIT Technology Review https://ift.tt/3MJqc7k

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. 

This week, I wrote about an external stimulator that delivers electrical pulses to the spine to help improve hand and arm function in people who are paralyzed. This isn’t a cure. In many cases the gains were relatively modest. One participant said it increased his typing speed from 23 words a minute to 35. Another participant was newly able to use scissors with his right hand. A third used her left hand to release a seatbelt.

The study didn’t garner as much media attention as previous, much smaller studies that focused on helping people with paralysis walk. Tech that allows people to type slightly faster or put their hair in a ponytail unaided just doesn’t have the same allure. “The image of a paralyzed person getting up and walking is almost biblical,” Charles Liu, director of the Neurorestoration Center at the University of Southern California, once told a reporter. 

For the people who have spinal cord injuries, however, incremental gains can have a huge impact on quality of life. 

So today in The Checkup, let’s talk about this tech and who it serves.

In 2004, Kim Anderson-Erisman, a researcher at Case Western Reserve University, who also happens to be paralyzed, surveyed more than 600 people with spinal cord injuries. Wanting to better understand their priorities, she asked them to consider seven different functions—everything from hand and arm mobility to bowel and bladder function to sexual function. She asked respondents to rank these functions according to how big an impact recovery would have on their quality of life. 

Walking was one of the functions, but it wasn’t the top priority for most people. Most quadriplegics put hand and arm function at the top of the list. For paraplegics, meanwhile, the top priority was sexual function. I interviewed Anderson-Erisman for a story I wrote in 2019 about research on implantable stimulators as a way to help people with spinal cord injuries walk. For many people, “not being able to walk is the easy part of spinal cord injury,” she told me. “[If] you don’t have enough upper-extremity strength or ability to take care of yourself independently, that’s a bigger problem than not being able to walk.” 

One of the research groups I focused on was at the University of Louisville. When I visited in 2019, the team had recently made the news because two people with spinal cord injuries in one of their studies had regained the ability to walk, thanks to an implanted stimulator. “Experimental device helps paralyzed man walk the length of four football fields,” one headline had trumpeted.

But when I visited one of those participants, Jeff Marquis, in his condo in Louisville, I learned that walking was something he could only do in the lab. To walk he needed to hold onto parallel bars supported by other people and wear a harness to catch him if he fell. Even if he had extra help at home, there wasn’t enough room for the apparatus. Instead, he gets around his condo the same way he gets around outside his condo: in a wheelchair. Marquis does stand at home, but even that requires a bulky frame. And the standing he does is only for therapy. “I mostly just watch TV while I’m doing that,” he said.  

That’s not to say the tech has been useless. The implant helped Marquis gain some balance, stamina, and trunk stability. “Trunk stability is kind of underrated in how much easier that makes every other activity I do,” he told me. “That’s the biggest thing that stays with me when I have [the stimulator] turned off.”  

What’s exciting to me about this latest study is that the tech gave the participants skills they could use beyond the lab. And because the stimulator is external, it is likely to be more accessible and vastly cheaper. Yes, the newly enabled movements are small, but if you listen to the palpable excitement of one study participant as he demonstrates how he can move a small ball into a cup, you’ll appreciate that incremental gains are far from insignificant. That’s according to Melanie Reid, one of the participants in the latest trial, who spoke at a press conference last week. “There [are] no miracles in spinal injury, but tiny gains can be life-changing.”

---

Now read the rest of The Checkup

Read more from MIT Technology Review’s archive

In 2017, we hailed as a breakthrough technology electronic interfaces designed to reverse paralysis by reconnecting the brain and body. Antonio Regalado has the story. 

An implanted stimulator changed John Mumford’s life, allowing him to once again grasp objects after a spinal cord injury left him paralyzed. But when the company that made the device folded, Mumford was left with few options for keeping the device running. “Limp limbs can be reanimated by technology, but they can be quieted again by basic market economics,” wrote Brian Bergstein in 2015. 

In 2014, Courtney Humphries covered some of the rat research that laid the foundation for the technological developments that have allowed paralyzed people to walk. 

From around the web

Lots of bird flu news this week. A second person in the US has tested positive for the illness after working with infected livestock. (NBC)

The livestock industry, which depends on shipping tens of millions of live animals, provides some ideal conditions for the spread of pathogens, including bird flu. (NYT)

Long read: How the death of a nine-year-old boy in Cambodia triggered a global H5N1 alert. (NYT)

You’ve heard about tracking viruses via wastewater. H5N1 is the first one we’re tracking via store-bought milk. (STAT) 

The first organ transplants from pigs to humans have not ended well, but scientists are learning valuable lessons about what they need to do better. (Nature) 

Another long read that’s worth your time: an inside look at just how long 3M knew about the pervasiveness of “forever chemicals.” (New Yorker) 

via TechSpot https://ift.tt/IGkr7tE
The latest annual survey of corporate images by the Harris Poll and Axios surveyed 16,500 Americans about the reputations of companies. Participants were first asked to name two companies with the best reputations and two with the worst, determining the 100 most visible firms. Later, they were asked to pick...

Read Entire Article

via MIT Technology Review https://ift.tt/tnz0moA

This article is from The Spark, MIT Technology Review’s weekly climate newsletter. To receive it in your inbox every Wednesday, sign up here.

Tech companies keep finding new ways to bring AI into every facet of our lives. AI has taken over my search engine results, and new virtual assistants from Google and OpenAI announced last week are bringing the world eerily close to the 2013 film Her (in more ways than one).

As AI has become more integrated into our world, I’ve gotten a lot of questions about the technology’s rising electricity demand. You may have seen the headlines proclaiming that AI uses as much electricity as small countries, that it’ll usher in a fossil-fuel resurgence, and that it’s already challenging the grid.  

So how worried should we be about AI’s electricity demands? Well, it’s complicated. 

Using AI for certain tasks can come with a significant energy price tag. With some powerful AI models, generating an image can require as much energy as charging up your phone, as my colleague Melissa Heikkilä explained in a story from December. Create 1,000 images with a model like Stable Diffusion XL, and you’ve produced as much carbon dioxide as driving just over four miles in a gas-powered car, according to the researchers Melissa spoke to. 

But while generated images are splashy, there are plenty of AI tasks that don’t use as much energy. For example, creating images is thousands of times more energy-intensive than generating text. And using a smaller model that’s tailored to a specific task, rather than a massive, all-purpose generative model, can be dozens of times more efficient. In any case, generative AI models require energy, and we’re using them a lot. 

Electricity consumption from data centers, AI, and cryptocurrency could reach double 2022 levels by 2026, according to projections from the International Energy Agency. Those technologies together made up roughly 2% of global electricity demand in 2022. Note that these numbers aren’t just for AI—it’s tricky to nail down AI’s specific contribution, so keep that in mind when you see predictions about electricity demand from data centers. 

There’s a wide range of uncertainty in the IEA’s projections, depending on factors like how quickly deployment increases and how efficient computing processes get. On the low end, the sector could require about 160 terawatt-hours of additional electricity by 2026. On the higher end, that number might be 590 TWh. As the report puts it, AI, data centers, and cryptocurrency together are likely adding “at least one Sweden or at most one Germany” to global electricity demand. 

In total, the IEA projects, the world will add about 3,500 TWh of electricity demand over that same period—so while computing is certainly part of the demand crunch, it’s far from the whole story. Electric vehicles and the industrial sector will both be bigger sources of growth in electricity demand than data centers in the European Union, for example. 

Still, some big tech companies are suggesting that AI could get in the way of their climate goals. Microsoft pledged four years ago to bring its greenhouse-gas emissions to zero (or even lower) by the end of the decade. But the company’s recent sustainability report shows that instead, emissions are still ticking up, and some executives point to AI as a reason. “In 2020, we unveiled what we called our carbon moonshot. That was before the explosion in artificial intelligence,” Brad Smith, Microsoft’s president, told Bloomberg Green.

What I found interesting, though, is that it’s not AI’s electricity demand that’s contributing to Microsoft’s rising emissions, at least on paper. The company has agreements in place and buys renewable-energy credits so that electricity needs for all its functions (including AI) are met with renewables. (How much these credits actually help is questionable, but that’s a story for another day.) 

Instead, infrastructure growth could be adding to the uptick in emissions. Microsoft plans to spend $50 billion between July 2023 and June 2024 on expanding data centers to meet demand for AI products, according to the Bloomberg story. Building those data centers requires materials that can be carbon intensive, like steel, cement, and of course chips. 

Some important context to consider in the panic over AI’s energy demand is that while the technology is new, this sort of concern isn’t, as Robinson Meyer laid out in an April story in Heatmap.

Meyer points to estimates from 1999 that information technologies were already accounting for up to 13% of US power demand, and that personal computers and the internet could eat up half the grid’s capacity within the decade. That didn’t end up happening, and even at the time, computing was actually accounting for something like 3% of electricity demand. 

We’ll have to wait and see if doomsday predictions about AI’s energy demand play out. The way I see it, though, AI is probably going to be a small piece of a much bigger story. Ultimately, rising electricity demand from AI is in some ways no different from rising demand from EVs, heat pumps, or factory growth. It’s really how we meet that demand that matters. 

If we build more fossil-fuel plants to meet our growing electricity demand, it’ll come with negative consequences for the climate. But if we use rising electricity demand as a catalyst to lean harder into renewable energy and other low-carbon power sources, and push AI to get more efficient, doing more with less energy, then we can continue to slowly clean up the grid, even as AI continues to expand its reach in our lives. 

---

Now read the rest of The Spark

Related reading

Check out my colleague Melissa’s story on the carbon footprint of AI from December here. 

For a closer look at Microsoft’s new sustainability report and the effects of AI, give this Bloomberg Green story from reporters Akshat Rathi and Dina Bass a read. 

Robinson Meyer at Heatmap dug into the context around the AI energy demand in this April piece. 

Another thing

Missed our event last week on thermal batteries? Good news—the recording is now available for subscribers!

For the latest in our Roundtables series, I spoke with Amy Nordrum, MIT Technology Review executive editor, about how the technology works, who the crucial players are, and what I’m watching for next. Check it out here. 

Keeping up with climate  

Changing how we generate heat in industry will be crucial to cleaning up that sector in China, according to a new report. Thermal batteries and heat pumps could meet most of the demand. (Axios)

Form Energy is known for its iron-air batteries, which could help unlock cheap energy storage on the grid. Now, the company is working on research to produce green iron. (Canary Media)

The NET Power pilot in Texas is working to generate electricity with natural gas while capturing the vast majority of emissions. But carbon capture technology in power plants is far from proven. (Cipher News)

MIT spinoff Electrified Thermal Solutions is working to bring its thermal battery technology to commercial use. The company’s product is roughly the size of an elevator and can reach temperatures up to 1,800 °C. (Inside Climate News)

Mexico City has seen constant struggles over water. Now groundwater is drying up, and a system of dams and canals may soon be unable to provide water to the city. (New York Times)

Sodium-ion batteries could offer cheap energy storage while avoiding material crunches for metals like lithium, nickel, and cobalt. China has a massive head start, leaving other countries scrambling to catch up. (Latitude Media)

→ Here’s how this abundant material could unlock cheaper energy storage. (MIT Technology Review)

Biochar is made by heating up biomass like wood and plants in low-oxygen environments. It’s a simple approach to carbon removal, but it doesn’t always get as much attention as other carbon removal technologies. (Heatmap)

This startup wants ships to capture their own emissions by bubbling exhaust through seawater and limestone and dumping it into the ocean. Experts caution that some components of the exhaust could harm sea life if they’re not handled properly. (New Scientist)

via TechSpot https://ift.tt/7saQFnR
Signal 65's Ryan Shrout put the Qualcomm Snapdragon X Elite SoC through a series of tests and benchmarks to see if it lives up to the promises Microsoft made during its recent event introducing the Copilot+ PCs, including the latest Surface line.

Read Entire Article

via MIT Technology Review https://ift.tt/vTeQo83

This story first appeared in China Report, MIT Technology Review’s newsletter about technology in China. Sign up to receive it in your inbox every Tuesday.

Last week’s release of GPT-4o, a new AI “omnimodel” that you can interact with using voice, text, or video, was supposed to be a big moment for OpenAI. But just days later, it feels as if the company is in big trouble. From the resignation of most of its safety team to Scarlett Johansson’s accusation that it replicated her voice for the model against her consent, it’s now in damage-control mode. 

Add to that another thing OpenAI fumbled with GPT-4o: the data it used to train its tokenizer—a tool that helps the model parse and process text more efficiently—is polluted by Chinese spam websites. As a result, the model’s Chinese token library is full of phrases related to pornography and gambling. This could worsen some problems that are common with AI models: hallucinations, poor performance, and misuse. 

I wrote about it on Friday after several researchers and AI industry insiders flagged the problem. They took a look at GPT-4o’s public token library, which has been significantly updated with the new model to improve support of non-English languages, and saw that more than 90 of the 100 longest Chinese tokens in the model are from spam websites. These are phrases like “_free Japanese porn video to watch,” “Beijing race car betting,” and “China welfare lottery every day.”

Anyone who reads Chinese could spot the problem with this list of tokens right away. Some such phrases inevitably slip into training data sets because of how popular adult content is online, but for them to account for 90% of the Chinese language used to train the model? That’s alarming.

“It’s an embarrassing thing to see as a Chinese person. Is that just how the quality of the [Chinese] data is? Is it because of insufficient data cleaning or is the language just like that?” says Zhengyang Geng, a PhD student in computer science at Carnegie Mellon University. 

It could be tempting to draw a conclusion about a language or a culture from the tokens OpenAI chose for GPT-4o. After all, these are selected as commonly seen and significant phrases from the respective languages. There’s an interesting blog post by a Hong Kong–based researcher named Henry Luo, who queried the longest GPT-4o tokens in various different languages and found that they seem to have different themes. While the tokens in Russian reflect language about the government and public institutions, the tokens in Japanese have a lot of different ways to say “thank you.”

But rather than reflecting the differences between cultures or countries, I think this explains more about what kind of training data is readily available online, and the websites OpenAI crawled to feed into GPT-4o.

After I published the story, Victor Shih, a political science professor at the University of California, San Diego, commented on it on X: “When you try not [to] train on Chinese state media content, this is what you get.”

It’s half a joke, and half a serious point about the two biggest problems in training large language models to speak Chinese: the readily available data online reflects either the “official,” sanctioned way of talking about China or the omnipresent spam content that drowns out real conversations.

In fact, among the few long Chinese tokens in GPT-4o that aren’t either pornography or gambling nonsense, two are “socialism with Chinese characteristics” and “People’s Republic of China.” The presence of these phrases suggests that a significant part of the training data actually is from Chinese state media writings, where formal, long expressions are extremely common.

OpenAI has historically been very tight-lipped about the data it uses to train its models, and it probably will never tell us how much of its Chinese training database is state media and how much is spam. (OpenAI didn’t respond to MIT Technology Review’s detailed questions sent on Friday.)

But it is not the only company struggling with this problem. People inside China who work in its AI industry agree there’s a lack of quality Chinese text data sets for training LLMs. One reason is that the Chinese internet used to be, and largely remains, divided up by big companies like Tencent and ByteDance. They own most of the social platforms and aren’t going to share their data with competitors or third parties to train LLMs. 

In fact, this is also why search engines, including Google, kinda suck when it comes to searching in Chinese. Since WeChat content can only be searched on WeChat, and content on Douyin (the Chinese TikTok) can only be searched on Douyin, this data is not accessible to a third-party search engine, let alone an LLM. But these are the platforms where actual human conversations are happening, instead of some spam website that keeps trying to draw you into online gambling.

The lack of quality training data is a much bigger problem than the failure to filter out the porn and general nonsense in GPT-4o’s token-training data. If there isn’t an existing data set, AI companies have to put in significant work to identify, source, and curate their own data sets and filter out inappropriate or biased content. 

It doesn’t seem OpenAI did that, which in fairness makes some sense, given that people in China can’t use its AI models anyway. 

Still, there are many people living outside China who want to use AI services in Chinese. And they deserve a product that works properly as much as speakers of any other language do. 

How can we solve the problem of the lack of good Chinese LLM training data? Tell me your idea at zeyi@technologyreview.com.

---

Now read the rest of China Report

Catch up with China

1. China launched an anti-dumping investigation into imports of polyoxymethylene copolymer—a widely used plastic in electronics and cars—from the US, the EU, Taiwan, and Japan. It’s widely seen as a response to the new US tariff announced on Chinese EVs. (BBC)

• Meanwhile, Latin American countries, including Mexico, Chile, and Brazil, have increased tariffs on Chinese-imported steel, testing China’s relationship with the region. (Bloomberg $)

2. China’s solar-industry boom is incentivizing farmers to install solar panels and make some extra cash by selling the electricity they generate. (Associated Press)

3. Hedging against the potential devaluation of the RMB, Chinese buyers are pushing the price of gold to all-time highs. (Financial Times $)

4. The Shanghai government set up a pilot project that allows data to be transferred out of China without going through the much-dreaded security assessments, a move that has been sought by companies like Tesla. (Reuters $)

5. China’s central bank fined seven businesses—including a KFC and branches of state-owned corporations—for rejecting cash payments. The popularization of mobile payment has been a good thing, but the dwindling support for cash is also making life harder for people like the elderly and foreign tourists. (Business Insider $)

6. Alibaba and Baidu are waging an LLM price war in China to attract more users. (Bloomberg $) 

7. The Chinese government has sanctioned Mike Gallagher, a former Republican congressman who chaired the Select Committee on China and remains a fierce critic of Beijing. (NBC News)

Lost in translation

China’s National Health Commission is exploring the relaxation of stringent rules around human genetic data to boost the biotech industry, according to the Chinese publication Caixin. A regulation enacted in 1998 required any research that involves the use of this data to clear an approval process. And there’s even more scrutiny if the research involves foreign institutions. 

In the early years of human genetic research, the regulation helped prevent the nonconsensual collection of DNA. But as the use of genetic data becomes increasingly important in discovering new treatments, the industry has been complaining about the bureaucracy, which can add an extra two to four months to research projects. Now the government is holding discussions on how to revise the regulation, potentially lifting the approval process for smaller-scale research and more foreign entities, as part of a bid to accelerate the growth of biotech research in China.

One more thing

Did you know that the Beijing Capital International Airport has been employing birds of prey to chase away other birds since 2019? This month, the second generation of Beijing’s birdy employees started their work driving away the migratory birds that could endanger aircraft. The airport even has different kinds of raptors—Eurasian hobbies, Eurasian goshawks, and Eurasian sparrowhawks—to deal with the different bird species that migrate to Beijing at different times.

via TechSpot https://ift.tt/Sdcjkv2
In the original Dune book, full-body stillsuits consisted of layers that absorbed body moisture from sweating and urination. This was then filtered so users could drink the reclaimed water from a tube attached to the neck or, as is the case in the recent movie adaptations, inserted into the nostrils....

Read Entire Article

via MIT Technology Review https://ift.tt/e3JpA8K

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

I’m excited to spend this week in Cambridge, Massachusetts. I’m visiting the mothership for MIT Technology Review’s annual flagship AI conference, EmTech Digital, on May 22-23. 

Between the world leaders gathering in Seoul for the second AI Safety Summit this week and Google and OpenAI’s launches of their supercharged new models, Astra and GPT-4o, the timing could not be better. AI feels hotter than ever.  

This year’s EmTech will be all about how we can harness the power of generative AI while mitigating its risks,and how the technology will affect the workforce, competitiveness, and democracy. We will also get a sneak peek into the AI labs of Google, OpenAI, Adobe, AWS, and others. 

This year’s top speakers include Nick Clegg, the president of global affairs at Meta, who will talk about what the platform intends to do to curb misinformation. In 2024, over 40 national elections will happen around the world, making it one of the most consequential political years in history. At the same time, generative AI has enabled an entirely new age of misinformation. And it’s all coalescing, with major shake-ups at social media companies and information platforms. MIT Technology Review’s executive editor Amy Nordrum will press Clegg on stage about what this all means for democracy.

Here are some other sessions I am excited about.

A Peek Inside Google’s plans 
Jay Yagnik, a vice president and engineering fellow at Google, will share what the history of AI can teach us about where the technology is going next and discuss Google’s vision for how to harness generative AI.  

From the Labs of OpenAI
Srinivas Narayanan, the vice president of applied AI at OpenAI, will share what the company has been building recently and what is coming next. In another session, Connor Holmes, who led work on video-generation AI Sora, will talk about how video-generation models could work as world simulators, and what this means for future AI models. 

The Often-Overlooked Privacy Problems in AI
Language models are prone to leaking private data. In this session Patricia Thaine, cofounder and CEO of Private AI, will explore methods that keep secrets secret and help organizations maintain compliance with privacy regulations. 

A Word Is Worth a Thousand Pictures
Cynthia Lu, senior director and head of applied research at Adobe, will walk us through the AI technology that Adobe is building and the ethical and legal implications of generated imagery. I’ve written about Adobe’s efforts to build generative AI in a non-exploitative way and how they’re paying off, so I’ll be interested to hear more about that.  

AI in the ER
Advances in medical image analysis are now enabling doctors to interpret radiology reports and automate incident documentation. This session by Polina Golland, the associate director of the MIT Computer Science and AI Laboratory, will explore both the challenges of working with sensitive personal data and the benefits of AI-assisted health care for patients.

Future Compute
On Tuesday, May 21, we are also hosting Future Compute, a day looking at how businesses and technical leaders navigate adopting AI. We have tech leaders from Salesforce, Stack Overflow, Amazon, and more, discussing how they are managing the AI transformation, and what pitfalls to avoid. 

I’d love to see you there, so if you can make it, sign up and come along! Readers of The Algorithm get 30% off tickets with the code ALGORITHMD24.

---

Now read the rest of The Algorithm

Deeper Learning

To kick off this busy week in AI, heavyweights such as Turing Prize winners Geoffrey Hinton and Yoshua Bengio, and a slew of other prominent academics and writers, have just written an op-ed published in Science calling for more investment in AI safety research. The op-ed, timed to coincide with the Seoul AI Safety Summit, represents the group’s wish list for leaders meeting to discuss AI. Many of the researchers behind the text have been heavily involved in consulting with governments and international organizations on the best approach to building safer AI systems. 

They argue that tech companies and public funders should invest at least a third of their AI R&D budgets into AI safety, and that governments should mandate stricter AI safety standards and assessments rather than relying on voluntary measures. The piece calls for them to establish fast-acting AI oversight bodies and provide them with funding comparable to the budgets of safety agencies in other sectors. It also says governments should require AI companies to prove that their systems cannot cause harm. 

But it’s hard to see this op-ed shifting things much. Tech companies have little incentive to spend money on measures that might slow down innovation and, crucially, product launches. Over the past few years, we’ve seen teams working on responsible AI take the hit during mass layoffs. Governments have shown more willingness to regulate AI in the last year or so, with the EU passing its first piece of comprehensive AI legislation, but this op-ed calls for them to go much further and faster. 

Despite that, focusing on the hypothetical existential risks posed by AI remains controversial among researchers, with some experts arguing that it distracts from the very real problems AI is causing today. As my colleague Will Douglas Heaven wrote last June when the AI safety debate was at a fever pitch: “The Overton window has shifted. What were once extreme views are now mainstream talking points, grabbing not only headlines but the attention of world leaders.”

Even Deeper Learning

GPT-4o’s Chinese token-training data is polluted by spam and porn websites

Last Monday OpenAI released GPT-4o, an AI model that you can communicate with in real time via live voice conversation, video streams from your phone, and text. But just days later, Chinese speakers started to notice that something seemed off about it: the tokens it uses to parse text were full of phrases related to spam and porn.

Oops, AI did it again: Humans read in words, but LLMs analyze tokens—distinct units in a sentence. When it comes to the Chinese language, the new tokenizer used by GPT-4o has introduced a disproportionate number of meaningless phrases. In one example, the longest token in GPT-4o’s public token library literally means “_free Japanese porn video to watch.” Experts say that’s likely due to insufficient data cleaning and filtering before the tokenizer was trained. (MIT Technology Review) 

Bits and Bytes

What’s next in chips
Thanks to the boom in artificial intelligence, the world of chips is on the cusp of a huge tidal shift. We outline four trends to look for in the year ahead that will define what the chips of the future will look like, who will make them, and which new technologies they’ll unlock. (MIT Technology Review) 

OpenAI and Google are launching supercharged AI assistants. Here’s how you can try them out.
OpenAI unveiled its GPT-4o assistant last Monday, and Google unveiled its own work building supercharged AI assistants just a day later. My colleague James O’Donnell walks you through what you should know about how to access these new tools, what you might use them for, and how much it will cost. 

OpenAI has lost its cofounder and dissolved the team focused on long-term AI risks
Last week OpenAI cofounder Ilya Sutskever and Jan Leike, the co-lead of the startup’s superalignment team, announced they were leaving the company. The superalignment team was set up less than a year ago to develop ways to control superintelligent AI systems. Leike said he was leaving because OpenAI’s “safety culture and processes have taken a backseat to shiny products.” In Silicon Valley, money always wins. (CNBC) 

Meta’s plan to win the AI race: give its tech away for free
Mark Zuckerberg’s bet is that making powerful AI technology free will drive down competitors’ prices, making Meta’s tech more widespread while others build products on top of it—ultimately giving him more control over the future of AI. (The Wall Street Journal) 

Sony Music Group has warned companies against using its content to train AI
The record label says it opts out of indiscriminate AI training and has started sending letters to AI companies prohibiting them from mining text or data, scraping the internet, or using Sony’s content without licensing agreements. (Sony) 

What do you do when an AI company takes your voice?
Two voice actors are suing Lovo, a startup, claiming it illegally took recordings of their voices to train their AI model. (The New York Times) 

via MIT Technology Review https://ift.tt/RYh6k7v

This is today’s edition of The Download, our weekday newsletter that provides a daily dose of what’s going on in the world of technology.

GPT-4o’s Chinese token-training data is polluted by spam and porn websites

Soon after OpenAI released GPT-4o last Monday, some Chinese speakers started to notice that something seemed off about this newest version of the chatbot: the tokens it uses to parse text were full of spam and porn phrases.

Humans read in words, but LLMs read in tokens, which are distinct units in a sentence that have consistent and significant meanings. GPT-4o is supposed to be better than its predecessors at handling multi-language tasks, and many of the advances were achieved through a new tokenization tool that does a better job compressing texts in non-English languages.

But, at least when it comes to the Chinese language, the new tokenizer used by GPT-4o has introduced a disproportionate number of meaningless phrases—and experts say that’s likely due to insufficient data cleaning and filtering before the tokenizer was trained. If left unresolved, it could lead to hallucinations, poor performance, and misuse. Read the full story.

—Zeyi Yang

Astronomers are enlisting AI to prepare for a data downpour

In deserts across Australia and South Africa, astronomers are planting forests of metallic detectors that will together scour the cosmos for radio signals. When it boots up in five years or so, the Square Kilometer Array Observatory will look for new information about the universe’s first stars and the different stages of galactic evolution. 

But after synching hundreds of thousands of dishes and antennas, astronomers will quickly face a new challenge: combing through some 300 petabytes of cosmological data a year—enough to fill a million laptops. So in preparation for the information deluge, astronomers are turning to AI for assistance. Read the full story.

—Zack Savitsky

Join us for Future Compute

If you’re interested in learning more about how to navigate the rapid changes in technology, Future Compute is the conference for you. It’s designed to help teach leaders strategic vision, agility, and a deep understanding of emerging technologies, and is held tomorrow, May 21, on MIT’s campus. Join us in-person or online by registering today.

EmTech Digital kicks off this week

The pace of AI development is truly breakneck these days—and we’ve got a sneak peek at what’s coming next. If you want to learn about how Google plans to develop and deploy AI, come and hear from its vice president of AI, Jay Yagnik, at our flagship AI conference, EmTech Digital. 

We’ll hear from OpenAI about its video generation model Sora too, and Nick Clegg, Meta’s president of global affairs, will also join MIT Technology Review’s executive editor Amy Nordrum for an exclusive interview on stage. 

It’ll be held at the MIT campus and streamed live online this week on May 22-23. Readers of The Download get 30% off tickets with the code DOWNLOADD24—here’s how to register. See you there!

The must-reads

I’ve combed the internet to find you today’s most fun/important/scary/fascinating stories about technology.

1 Apple is teaming up with OpenAI to overhaul iOS18 
In the hopes it’ll give Apple an edge over rivals Google and Microsoft. (Bloomberg $)
+ OpenAI and Google recently launched their own supercharged AI assistants. (MIT Technology Review)

2 Blue Origin took six customers to the edge of space on Sunday
It’s the company’s first tourist flight in almost two years. (CNN)
+ Space tourism hasn’t exactly got off the ground yet. (WP $)

3 How TikTok users are skirting around its weight-loss drug promotion ban
Talking in code is becoming increasingly common. (WP $)
+ A new kind of weight-loss therapy is on the horizon. (Fast Company $)
+ What don’t we know about Ozempic? Quite a lot, actually. (Vox)
+ Weight-loss injections have taken over the internet. But what does this mean for people IRL? (MIT Technology Review)

4 Chinese companies are pushing ‘AI-in-a-box’ products
They’re sold as all-in-one cloud computing solutions, much to cloud providers’ chagrin. (FT $)

5 Microscopic blood clots could explain the severity of long covid 
But doctors are calling for rigorous peer review before any solid conclusions can be made. (Undark Magazine)
+ Scientists are finding signals of long covid in blood. They could lead to new treatments. (MIT Technology Review)

6 How hackers saved stalled Polish trains
It looks as though the locomotives’ manufacturer could be behind the breakdown. (WSJ $)

7 We’re getting closer to making an HIV vaccine
A successful trial is giving researchers new hope. (Wired $)
+ Three people were gene-edited in an effort to cure their HIV. The result is unknown. (MIT Technology Review)

8 Most healthy people don’t need to track their blood glucose
That doesn’t stop companies trying to sell you their monitoring services, though. (The Guardian)

9 Filming strangers is public is not okay
And yet, people keep doing it. Why? (Vox)

10 Beware the spread of AI slop
Spam is no longer a strong enough term—the latest wave of AI images is slop. (The Guardian)

Quote of the day

“It’s a process of trust collapsing bit by bit, like dominoes falling one by one.”

—An anonymous OpenAI insider tells Vox that safety-minded employees are losing faith in the company’s CEO Sam Altman.

The big story

What does GPT-3 “know” about me?

August 2022

One of the biggest stories in tech is the rise of large language models that produce text that reads like a human might have written it.

These models’ power comes from being trained on troves of publicly available human-created text hoovered up from the internet. If you’ve posted anything even remotely personal in English on the internet, chances are your data might be part of some of the world’s most popular LLMs.

Melissa Heikkilä, MIT Technology Review’s AI reporter, wondered what data these models might have on her—and how it could be misused. So she put OpenAI’s GPT-3 to the test. Read about what she found.

We can still have nice things

A place for comfort, fun and distraction to brighten up your day. (Got any ideas? Drop me a line or tweet ’em at me.)

+ Sea urchins just love tiny hats
+ There’s nothing better than a Lego optical illusion of sorts.
+ Waking up each morning can be tough. Maybe a better alarm is the way forward?
+ Out of the way: it’s the annual worm charming championships!

via TechSpot https://ift.tt/3P7XMlf
From Microsoft's Project Natick to China's Hainan Undersea Data Center Demonstration Development Project, we've seen data centers being placed underwater for years, saving precious land space and providing a dust-free, oxygen-free environment, which can protect electronics and help reduce faults.

Read Entire Article

via MIT Technology Review https://ift.tt/1LwGBN7

In deserts across Australia and South Africa, astronomers are planting forests of metallic detectors that will together scour the cosmos for radio signals. When it boots up in five years or so, the Square Kilometer Array Observatory will look for new information about the universe’s first stars and the different stages of galactic evolution. 

But after synching hundreds of thousands of dishes and antennas, astronomers will quickly face a new challenge: combing through some 300 petabytes of cosmological data a year—enough to fill a million laptops. 

It’s a problem that will be repeated in other places over the coming decade. As astronomers construct giant cameras to image the entire sky and launch infrared telescopes to hunt for distant planets, they will collect data on unprecedented scales. 

“We really are not ready for that, and we should all be freaking out,” says Cecilia Garraffo, a computational astrophysicist at the Harvard-Smithsonian Center for Astrophysics. “When you have too much data and you don’t have the technology to process it, it’s like having no data.”

In preparation for the information deluge, astronomers are turning to AI for assistance, optimizing algorithms to pick out patterns in large and notoriously finicky data sets. Some are now working to establish institutes dedicated to marrying the fields of computer science and astronomy—and grappling with the terms of the new partnership.

In November 2022, Garraffo set up AstroAI as a pilot program at the Center for Astrophysics. Since then, she has put together an interdisciplinary team of over 50 members that has planned dozens of projects focusing on deep questions like how the universe began and whether we’re alone in it. Over the past few years, several similar coalitions have followed Garraffo’s lead and are now vying for funding to scale up to large institutions.

Garraffo recognized the potential utility of AI models while bouncing between career stints in astronomy, physics, and computer science. Along the way, she also picked up on a major stumbling block for past collaboration efforts: the language barrier. Often, astronomers and computer scientists struggle to join forces because they use different words to describe similar concepts. Garraffo is no stranger to translation issues, having struggled to navigate an English-only school growing up in Argentina. Drawing from that experience, she has worked to put people from both communities under one roof so they can identify common goals and find a way to communicate. 

Astronomers had already been using AI models for years, mainly to classify known objects such as supernovas in telescope data. This kind of image recognition will become increasingly vital when the Vera C. Rubin Observatory opens its eyes next year and the number of annual supernova detections quickly jumps from hundreds to millions. But the new wave of AI applications extends far beyond matching games. Algorithms have recently been optimized to perform “unsupervised clustering,” in which they pick out patterns in data without being told what specifically to look for. This opens the doors for models pointing astronomers toward effects and relationships they aren’t currently aware of. For the first time, these computational tools offer astronomers the faculty of “systematically searching for the unknown,” Garraffo says. In January, AstroAI researchers used this method to catalogue over 14,000 detections from x-ray sources, which are otherwise difficult to categorize.

Another way AI is proving fruitful is by sniffing out the chemical composition of the skies on alien planets. Astronomers use telescopes to analyze the starlight that passes through planets’ atmospheres and gets soaked up at certain wavelengths by different molecules. To make sense of the leftover light spectrum, astronomers typically compare it with fake spectra they generate based on a handful of molecules they’re interested in finding—things like water and carbon dioxide. Exoplanet researchers dream of expanding their search to hundreds or thousands of compounds that could indicate life on the planet below, but it currently takes a few weeks to look for just four or five compounds. This bottleneck will become progressively more troublesome as the number of exoplanet detections rises from dozens to thousands, as is expected to happen thanks to the newly deployed James Webb Space Telescope and the European Space Agency’s Ariel Space Telescope, slated to launch in 2029. 

Processing all those observations is “going to take us forever,” says Mercedes López-Morales, an astronomer at the Center for Astrophysics who studies exoplanet atmospheres. “Things like AstroAI are showing up at the right time, just before these faucets of data are coming toward us.”

Last year López-Morales teamed up with Mayeul Aubin, then an undergraduate intern at AstroAI, to build a machine-learning model that could more efficiently extract molecular composition from spectral data. In two months, their team built a model that could scour thousands of exoplanet spectra for the signatures of five different molecules in 31 seconds, a feat that won them the top prize in the European Space Agency’s Ariel Data Challenge. The researchers hope to train a model to look for hundreds of additional molecules, boosting their odds of finding signs of life on faraway planets. 

AstroAI collaborations have also given rise to realistic simulations of black holes and maps of how dark matter is distributed throughout the universe. Garraffo aims to eventually build a large language model similar to ChatGPT that’s trained on astronomy data and can answer questions about observations and parse the literature for supporting evidence. 

“There’s this huge new playground to explore,” says Daniela Huppenkothen, an astronomer and data scientist at the Netherlands Institute for Space Research. “We can use [AI] to tackle problems we couldn’t tackle before because they’re too computationally expensive.” 

However, incorporating AI into the astronomy workflow comes with its own host of trade-offs, as Huppenkothen outlined in a recent preprint. The AI models, while efficient, often operate in ways scientists don’t fully understand. This opacity makes them complicated to debug and difficult to identify how they may be introducing biases. Like all forms of generative AI, these models are prone to hallucinating relationships that don’t exist, and they report their conclusions with an unfounded air of confidence. 

“It’s important to critically look at what these models do and where they fail,” Huppenkothen says. “Otherwise, we’ll say something about how the universe works and it’s not actually true.”

Researchers are working to incorporate error bars into algorithm responses to account for the new uncertainties. Some suggest that the tools could warrant an added layer of vetting to the current publication and peer-review processes. “As humans, we’re sort of naturally inclined to believe the machine,” says Viviana Acquaviva, an astrophysicist and data scientist at the City University of New York who recently published a textbook on machine-learning applications in astronomy. “We need to be very clear in presenting results that are often not clearly explicable while being very honest in how we represent capabilities.”

Researchers are cognizant of the ethical ramifications of introducing AI, even in as seemingly harmless a context as astronomy. For instance, these new AI tools may perpetuate existing inequalities in the field if only select institutions have access to the computational resources to run them. And if astronomers recycle existing AI models that companies have trained for other purposes, they also “inherit a lot of the ethical and environmental issues inherent in those models already,” Huppenkothen says.

Garraffo is working to get ahead of these concerns. AstroAI models are all open source and freely available, and the group offers to help adapt them to different astronomy applications. She has also partnered with Harvard’s Berkman Klein Center for Internet & Society to formally train the team in AI ethics and learn best practices for avoiding biases. 

Scientists are still unpacking all the ways the arrival of AI may affect the field of astronomy. If AI models manage to come up with fundamentally new ideas and point scientists toward new avenues of study, it will forever change the role of the astronomer in deciphering the universe. But even if it remains only an optimization tool, AI is set to become a mainstay in the arsenal of cosmic inquiry. 

“It’s going to change the game,” Garraffo says. “We can’t do this on our own anymore.” 

Zack Savitsky is a freelance science journalist who covers physics and astronomy. 

via TechSpot https://ift.tt/xzsh53f
Intel has tried to dethrone AMD's Ryzen Z1-series chips as the dominant supplier for handheld gaming consoles running Windows 11 with little success, but that may change with this year's expected release of its next-gen Core Ultra Series 2 Lunar Lake chips. Looking at the predicted specs, they appear power-efficient...

Read Entire Article