When I opened the email telling me I’d been accepted to run the London Marathon, I felt elated. And then terrified. Barely six months on from my last marathon, I knew how dedicated I’d have to be to keep running day after day, week after week, month after month, through rain, cold, tiredness, grumpiness, and hangovers.

The marathon is the easy part. It’s the constant grind of the training that kills you—and finding ways to keep it fresh and interesting is part of the challenge. Some exercise nuts think they’ve found a way to live their routines up: by using the AI chatbot ChatGPT as a sort of proxy personal trainer.

Its appeal is obvious. ChatGPT answers questions in seconds, saving the need to sift through tons of information, and asking follow-up questions will give you a more detailed and personalized answer. But is ChatGPT really the future of how we work out? Or is it just a confident bullshitter? Read the full story.

—Rhiannon Williams

How new technologies could clean up air travel

Aviation is a notorious “hard-to-decarbonize” sector. It makes up about 3% of the world’s greenhouse-gas emissions, and airline traffic could more than double from today’s levels by 2050. 

When it comes to flying, the technical challenge of cutting emissions is especially steep. Fuels for planes need to be especially light and compact, so planes can make it into the sky and still have room for people or cargo. But the industry has some promising ideas for cleaning up its act—and some of them are already taking off. Read the full story.

Hitting the gym

Despite the variable quality of ChatGPT’s fitness tips, some people have actually been following its advice in the gym. 

John Yu, a TikTok content creator based in the US, filmed himself following a six-day full-body training program courtesy of ChatGPT. He instructed it to give him a sample workout plan each day, tailored to which bit of his body he wanted to work (his arms, legs, etc), and then did the workout it gave him. 

The exercises it came up with were perfectly fine, and easy enough to follow. However, Yu  found that the moves lacked variety. “Strictly following what ChatGPT gives me is something I’m not really interested in,” he says. 

Lee Lem, a bodybuilding content creator based in Australia, had a similar experience. He asked ChatGPT to create an “optimal leg day” program. It suggested the right sorts of exercises—squats, lunges, deadlifts, and so on—but the rest times between them were far too brief. “It’s hard!” Lem says, laughing. “It’s very unrealistic to only rest 30 seconds between squat sets.”

Lem hit on the core problem with ChatGPT’s suggestions: they fail to consider human bodies. As both he and Yu found out, repetitive movements quickly leave us bored or tired. Human coaches know to mix their suggestions up. ChatGPT has to be explicitly told.

For some, though, the appeal of an AI-produced workout is still irresistible—and something they’re even willing to pay for. Ahmed Mire, a software engineer based in London, is selling ChatGPT-produced plans for $15 each. People give him their workout goals and specifications, and he runs them through ChatGPT. He says he’s already signed up customers since launching the service last month and is considering adding the option to create diet plans too. ChatGPT is free, but he says people pay for the convenience. 

What united everyone I spoke to was their decision to treat ChatGPT’s training suggestions as entertaining experiments rather than serious athletic guidance. They all had a good enough understanding of fitness, and what does and doesn’t work for their bodies, to be able to spot the model’s weaknesses. They all knew they needed to treat its answers skeptically. People who are newer to working out might  be more inclined to take them at face value.

The future of fitness?

This doesn’t mean AI models can’t or shouldn’t play a role in developing fitness plans. But it does underline that they can’t necessarily be trusted. ChatGPT will improve and could learn to ask its own questions. For example, it might ask users if there are any exercises they hate, or inquire about any niggling injuries. But essentially, it can’t come up with original suggestions, and it has no fundamental understanding of the concepts it is regurgitating

Batteries could power planes, at least for short distances. Some companies have been trying out test flights of electric planes powered this way, mostly small eVTOL (electric vertical take-off and landing) aircraft that can carry just a few people. Unlike combustion-powered aircraft, electric planes wouldn’t produce pollution, and they could reach zero emissions if charged with renewable energy.

Batteries have the benefit of being a technology that’s widely used today in electric vehicles, and they’ve gotten much better over their decades of development. But batteries will have to keep improving dramatically for electric planes to carry a significant number of people any significant distance. (Check out my story from last year on electric planes for more.)

Hydrogen could be a versatile fuel for aviation in the future. Planes might use hydrogen in two different ways. It could be burned in combustion engines, similar to how jet fuel is used today. Alternatively, hydrogen could be used in fuel cells, where chemical reactions generate electricity. We love to have options. 

Hydrogen’s environmental impact and feasibility will depend on how it’s being used. Combustion will lead to some tailpipe emissions, though these would be mostly water. Hydrogen-electric planes, like aircraft powered by batteries, could be free from climate pollution depending on how the hydrogen is produced. 

In either case, hydrogen has one key thing going for it: it contains a lot of energy without being too heavy (unlike batteries). When a vehicle has to lug its power source 30,000 feet into the air, it’s better for that power source to be really light—and hydrogen, as the lightest element on the periodic table, fits this bill perfectly. 

The problem is, while hydrogen is light, it also takes up a lot of space. In order to get it into a small enough volume to carry onboard a plane, hydrogen will likely need to be cooled to cryogenic temperatures (below -250 °C). Designing these systems and getting them onto planes will be difficult. So will sourcing and distributing large amounts of hydrogen made with renewable energy. And there’s the small fact that while there have been some experiments with flying hydrogen-powered planes over the years, the technology still needs work. It’s hard to remake an industry, which is why SAFs, the drop-in solution, are probably the most likely to be adopted in the near future, while hydrogen will take decades to break through. 

But there’s been some exciting movement on using hydrogen for aviation over the past couple of years, with big players like Airbus getting into the game and announcing planned test flights. 

And last week, startup ZeroAvia was in the news again, announcing it had completed a test flight of a 19-seat Dornier 228, the largest plane flown partly on hydrogen fuel cells. Before this test, the company had tested a smaller, nine-seat aircraft. 

The news: Eight years ago, a patient lost her power of speech because of ALS, or Lou Gehrig’s disease, which causes progressive paralysis. Now, after volunteering to receive a brain implant, the woman has been able to rapidly communicate phrases at a rate approaching normal speech.

Why it matters: Even in an era of keyboards, thumb-typing, emojis, and internet abbreviations, speech remains the fastest form of human-to-human communication. The scientists from Stanford University say their volunteer smashed previous records by using the brain-reading implant to communicate at a rate of 62 words a minute, three times the previous best. 

What’s next: Although the study has not been formally reviewed, experts have hailed the results as a significant breakthrough. The findings could pave the way for experimental brain-reading technology to leave the lab and become a useful product soon. Read the full story.

—Antonio Regalado

Resolving to live the Year of the Rabbit to the fullest

By Zeyi Yang, China reporter

This past Sunday was the Lunar New Year, the most important holiday for Chinese and several other Asian cultures. It’s supposed to be an opportunity for us to reset and seize new opportunities.

In that spirit, I’ve recently revisited some of my favorite China-focused MIT Technology Review stories from the last year and gone back to the people I interviewed. I asked them whether they’d resolved any troubling challenges, and what they’re hoping for in the Year of the Rabbit. 

6. The history of Zhongguancun, China’s Silicon Valley, explained. (Wired $)

7. A Chinese state-owned bank in Hong Kong is enticing new clients from the mainland with the possibility of getting mRNA vaccine shots. (Financial Times $)

Lost in translation

The new year is for new changes, and as Chinese tech publication Baobian reported, many Chinese Big Tech workers are quitting the industry and reflecting on how they ended up working pointless “bullshit jobs.”

Even though the country’s tech industry is relatively young, these companies, like their Western counterparts, have grown into gigantic corporations burdened with bureaucracy and low efficiency. A main source of frustration for staffers is feeling that they are spending months working on insignificant product changes that could be vetoed at the last minute. For example, making a simple UI design change requires two weeks of opposition research, and there’s little originality in the final product. Some workers also feel they are losing their individual purpose while helping the company optimize its money-making machinery.

Luyi, who worked for Tencent, Alibaba, and ByteDance in different positions, felt that she was chasing abstract numbers based on unreliable data analytics, and ultimately achieving nothing. Last year, she finally decided to quit the tech industry and went to work for an art gallery in Beijing. “When I successfully organize an art exhibit, there’s an immense sense of achievement. I can get a lot of positive feedback on the scene,” she said. That’s the feeling she was missing when she worked in Big Tech.

One more thing

To celebrate the transition from the Year of the Tiger to the Year of the Rabbit, a zoo in western China organized a ceremony on Friday in which a tiger cub and a rabbit were placed on the same table. But the video was promptly cut when the tiger went for the rabbit’s neck, the correspondent began shouting in panic, and the scene descended into chaos. Fortunately, the rabbit was reportedly unharmed. Otherwise it would have been a terrible omen for the new year.

In the new research, the Stanford team wanted to know if neurons in the motor cortex contained useful information about speech movements, too. That is, could they detect how “subject T12” was trying to move her mouth, tongue, and vocal cords as she attempted to talk?

These are small, subtle movements, and according to Sabes, one big discovery is that just a few neurons contained enough information to let a computer program predict, with good accuracy, what words the patient was trying to say. That information was conveyed by Shenoy’s team to a computer screen, where the patient’s words appeared as they were spoken by the computer.

The new result builds on previous work by Edward Chang at the University of California, San Francisco, who has written that speech involves the most complicated movements people make. We push out air, add vibrations that make it audible, and form it into words with our mouth, lips, and tongue. To make the sound “f,” you put your top teeth on your lower lip and push air out—just one of dozens of mouth movements needed to speak.  

A path forward

Chang previously used electrodes placed on top of the brain to permit a volunteer to speak through a computer, but in their preprint, the Stanford researchers say their system is more accurate and three to four times faster.

“Our results show a feasible path forward to restore communication to people with paralysis at conversational speeds,” wrote the researchers, who included Shenoy and neurosurgeon Jaimie Henderson.

David Moses, who works with Chang’s team at UCSF, says the current work reaches “impressive new performance benchmarks.” Yet even as records continue to be broken, he says, “it will become increasingly important to demonstrate stable and reliable performance over multi-year time scales.” Any commercial brain implant could have a difficult time getting past regulators, especially if it degrades over time or if the accuracy of the recording falls off.

The path forward is likely to include both more sophisticated implants and closer integration with artificial intelligence. 

The current system already uses a couple of types of machine learning programs. To improve its accuracy, the Stanford team employed software that predicts what word typically comes next in a sentence. “I” is more often followed by “am” than “ham,” even though these words sound similar and could produce similar patterns in someone’s brain. 

When it comes to the climate, the picture can look bleak.

Emissions of the greenhouse gasses that cause climate change are estimated to have reached new heights in 2022. Meanwhile, climate disasters, from record heat waves in China and Europe to devastating floods in Pakistan, seem to be hitting at a breakneck pace.

But a close look at global data shows that there are a few bright spots of good news, and a lot of potential progress ahead. Renewable sources make up a growing fraction of the energy supply, and they’re getting cheaper every year. Countries are also  setting new targets for emissions reductions, and unprecedented public investments could unlock more technological advances. 

So despite what can feel like a barrage of bad news, there are at least a few reasons to be hopeful. Read the full story.

—Casey Crownhart

These simple design rules could turn the chip industry on its head

Since the computer was invented, humans have devised many programming languages to command them to do our bidding. For a chip to execute your code, software must translate it into instructions a chip can use. So engineers designate specific binary sequences to prompt the hardware to perform certain actions, known as the computer’s instruction set.  

This is a far cry from the field’s reputation in the 1990s, when Wooldridge was finishing his PhD. AI was still seen as a weird, fringe pursuit; the wider tech sector viewed it in a similar way to how established medicine views homeopathy, he says. 

Today’s AI research boom has been fueled by neural networks, which saw a big breakthrough in the 1980s and work by simulating the patterns of the human brain. Back then, the technology hit a wall because the computers of the day weren’t powerful enough to run the software. Today we have lots of data and extremely powerful computers, which makes the technique viable. 

New breakthroughs, such as the chatbot ChatGPT and the text-to-image model Stable Diffusion, seem to come every few months. Technologies like ChatGPT are not fully explored yet, and both industry and academia are still working out how they can be useful, says Stone. 

Instead of a full-blown AI winter, we are likely to see a drop in funding for longer-term AI research and more pressure to make money using the technology, says Wooldridge. Researchers in corporate labs will be under pressure to show that their research can be integrated into products and thus make money, he adds.

That’s already happening. In light of the success of OpenAI’s ChatGPT, Google has declared a “code red” threat situation for its core product, Search, and is looking to aggressively revamp Search with its own AI research. 

Stone sees parallels to what happened at Bell Labs. If Big Tech’s AI labs, which dominate the sector, turn away from deep, longer-term research and focus too much on shorter-term product development, exasperated AI researchers may leave for academia, and these big labs could lose their grip on innovation, he says. 

That’s not necessarily a bad thing. There are a lot of smart people looking for jobs at the moment. Venture capitalists are looking for new startups to invest in as crypto fizzles out, and generative AI has shown how the technology can be made into products. 

This moment presents the AI sector with a once-in-a-generation opportunity to play around with the potential of new technology. Despite all the gloom around the layoffs, it’s an exciting prospect. 

But the silicon switches in your laptop’s central processor don’t inherently understand the word “for” or the symbol “=.” For a chip to execute your Python code, software must translate these words and symbols into instructions a chip can use.  

Engineers designate specific binary sequences to prompt the hardware to perform certain actions. The code “100000,” for example, could order a chip to add two numbers, while the code “100100” could ask it to copy a piece of data. These binary sequences form the chip’s fundamental vocabulary, known as the computer’s instruction set. 

For years, the chip industry has relied on a variety of proprietary instruction sets. Two major types dominate the market today: x86, which is used by Intel and AMD, and Arm, made by the company of the same name. Companies must license these instruction sets—which can cost millions of dollars for a single design. And because x86 and Arm chips speak different languages, software developers must make a version of the same app to suit each instruction set. 

Lately, though, many hardware and software companies worldwide have begun to converge around a publicly available instruction set known as RISC-V. It’s a shift that could radically change the chip industry. RISC-V proponents say that this instruction set makes computer chip design more accessible to smaller companies and budding entrepreneurs by liberating them from costly licensing fees. 

“There are already billions of RISC-V-based cores out there, in everything from earbuds all the way up to cloud servers,” says Mark Himelstein, the CTO of RISC-V International, a nonprofit supporting the technology. 

In February 2022, Intel itself pledged $1 billion to develop the RISC-V ecosystem, along with other priorities. While Himelstein predicts it will take a few years before RISC-V chips are widespread among personal computers, the first laptop with a RISC-V chip, the Roma by Xcalibyte and DeepComputing, became available in June for pre-order.

What is RISC-V?

You can think of RISC-V (pronounced “risk five”) as a set of design norms, like Bluetooth, for computer chips. It’s known as an “open standard.” That means anyone—you, me, Intel—can participate in the development of those standards. In addition, anyone can design a computer chip based on RISC-V’s instruction set. Those chips would then be able to execute any software designed for RISC-V. (Note that technology based on an “open standard” differs from “open-source” technology. An open standard typically designates technology specifications, whereas “open source” generally refers to software whose source code is freely available for reference and use.)

A group of computer scientists at UC Berkeley developed the basis for RISC-V in 2010 as a teaching tool for chip design. Proprietary central processing units (CPUs) were too complicated and opaque for students to learn from. RISC-V’s creators made the instruction set public and soon found themselves fielding questions about it. By 2015, a group of academic institutions and companies, including Google and IBM, founded RISC-V International to standardize the instruction set.