Technology

➚ MOVE FAST: Pinky promise. Uber had been signaling for months that it would reward investors after pledging to put profitability over growth. But when it announced an earnings milestone last week, shareholders were disappointed. Today, it came through with a $7 billion share buyback, a first for the company.

➘ BREAK THINGS: Fat finger. A clerical error that added a zero on a profitability margin metric sent Lyft’s shares soaring, and then plummeting yesterday. The mistake eclipsed its improving financial performance. But the “year of efficiency” that has taken over the tech industry and extended into 2024 has paid off, though jobs are a casualty.

After the consumer internet exploded in the mid-1990s, it took decades before it became truly global, touching nearly every corner of the world. The rise of large language models looks like it will become international far sooner.

Cohere for AI, the nonprofit arm of the LLM provider, worked with 3,000 independent researchers in 119 countries to create a dataset and model that is fluent in 101 languages. While there are more than 7,000 languages spoken on the planet, 100 covers a large chunk of the global population.

The model, called Aya, is completely open source, allowing researchers around the world to use it for their own purposes.

While Cohere for AI’s effort was a big undertaking, it’s striking how little text was needed to teach the model. It’s another sign that language barriers between nations are coming down.

Little known startup takes the AI weather prediction crown

THE SCOOP

A team of Stanford graduates in their 20s has overtaken tech giants like Huawei, Nvidia, and Google DeepMind in the competitive field of using artificial intelligence to predict the weather.

Startup WindBorne Systems announced Wednesday that it surpassed DeepMind, the current leader in AI weather prediction, in key benchmarks set by U.S. and European government weather models.

In an exclusive interview with Semafor, the co-founders of the firm, backed by Khosla Ventures and Footwork, said they used an in-air fleet of around 100 inexpensive, hand-built weather balloons made from plastic purchased from a restaurant supply company to gather granular data, which they then analyzed using the same AI techniques that power ChatGPT.

WindBorne has already earned millions of dollars in revenue by contracting with government agencies like the National Oceanic and Atmospheric Association and the U.S. Air Force. With the new forecasting capability, it plans to go after the commercial market.

Last week, a computer monitor at the company’s headquarters showed six of its balloons rafting toward California in an atmospheric river that was pummeling the southern part of the state. The data the balloons gathered was fed into the company’s algorithms, and to NOAA and the Scripps Institute of Oceanography as part of a research project.

WindBorne says its sensors could help governments better prepare for extreme weather events that can cause billions in damage, such as Hurricane Otis, which last year went from a blip on the radar to a full-blown hurricane in less than 24 hours.

Balloon capabilities were prominent in the news last year when the U.S. shot down what it claimed was a Chinese surveillance balloon over American land; Beijing said it was a weather balloon.

KNOW MORE

WindBorne’s WeatherMesh system takes advantage of two technology trends: The rapid evolution of AI algorithms and the precipitous decline in the cost and size of computer hardware and wireless equipment.

WindBorne’s weather balloons, which cost about as much to manufacture as an inexpensive mobile phone, can orbit around the earth for weeks, using AI to precisely control their paths. WindBorne says it already employs the world’s largest constellation of weather balloons, and will increase it 100-fold to 10,000 balloons, giving a tiny startup as much visibility into the earth’s weather systems as heavily funded government agencies.

“As far as I’m aware, this makes us the first company to apply AI-based weather forecasting at scale in the real world,” said WindBorne co-founder Kai Marshland, who said the company markets its forecasts to a wide variety of potential customers, from maritime shipping companies to energy traders. The technology could also provide valuable data to climate researchers and help businesses save fuel, thereby reducing emissions.

Traditional computer-based weather forecasts are based on physics models that create atmospheric simulations. They require massive amounts of compute power, and are much slower compared to newer methods that utilize a different approach: deep learning.

Instead of using physics to understand what is happening in the atmosphere, the deep learning technique takes in vast amounts of data, from wind speed to barometric pressure, and picks up on patterns and cues that would be impossible for a human to find. Once it’s been sufficiently trained, the model can look at real-time weather data and predict where those metrics are likely to go, even without any prior knowledge of physics.

In November, DeepMind announced that GraphCast, a model trained on 40 years of weather data, was more accurate than the European Centre for Medium-Range Weather Forecasts, which is known as the gold standard in weather modeling.

WindBorne says its WeatherMesh model is 11% more accurate than DeepMind’s in the key forecasting metrics.

Another benefit of using deep learning instead of physics models is that although they require a lot of compute power for the initial training of the model, they are relatively fast and cheap to run after that initial process. DeepMind’s GraphCast and WindBorne’s WeatherMesh can both run on consumer-grade computers, and can complete in a matter of seconds work that would usually require a supercomputer under traditional models.

WindBorne is using a transformer model, a deep learning technique pioneered by Google and used in large language models like ChatGPT, which use next-word prediction to complete a sentence. WindBorne uses it to project what comes next in a weather forecast.

The company says it plans to combine both deep learning and the physics model approach to gain an advantage. Using proprietary data gathered by its balloons, it plans to create its own physics models for certain regions and then use the output of those models as data to pre-train its deep-learning model creating more detailed weather forecasts over certain geographies.

Over the weekend, WindBorne conducted a simulation of how its WeatherMesh model would perform in predicting the path of Hurricane Ian, a Category 5 storm that wreaked havoc in the Southeastern U.S. in 2022. The model followed the storm so perfectly that employees at first thought something must have been off, like the hurricane itself made it into the training dataset, giving it an unfair advantage. But the numbers checked out; the traditional models were off by around 200 kilometers.

The number of trackers Mozilla Research found that a romantic AI chatbot had deployed within the first 60 seconds of interactions, giving new meaning to the term “pillow talk” in the generative AI era. Happy Valentine’s Day!

Learn what matters in tech, AI, and venture capital. The Generalist delivers in-depth analysis and insights on the world’s most successful companies and technologies. Whether you’re an investor, entrepreneur, or operator, The Generalist equips you with the knowledge to capitalize on the future. Join 100k+ readers by signing up here.

Mark Zuckerberg posted a clever video touting all the ways his Meta Quest 3 headset is better than Apple’s new Vision Pro. The image that came to mind was the now infamous Steve Ballmer interview in which he dismisses the iPhone as an overpriced toy.

The thing is, Ballmer wasn’t wrong in saying phones with physical keyboards were better for writing emails. (As someone who writes a lot, I still kind of miss my BlackBerry.) But he misjudged that most people wanted to use their phones for so much more than just work.

Zuckerberg may also be right that the Quest is the overall best headset today, but possibly wrong about consumer preferences in the long run.

When you step back, this debate says so much about how Apple has changed as a company since the iPhone first came out. The roles are reversed, with Apple playing the part of the behemoth trying to sell what is essentially a work tool.

Zuckerberg, in his video, sounded more like Apple did in 2007 than Apple does today. He seemed to be really passionate about what consumers can do with their device, like play Xbox and other video games.

Meanwhile, the Vision Pro feels more limited, mainly because of Apple’s walled garden, making its products incompatible with devices and services from other companies.

Whether one platform wins or both products end up getting stuffed into office drawers, the devices are a reflection of how Apple’s image has changed.

• The Chinese e-commerce app Temu dominated Super Bowl advertising this year.
• How Washington reacted to Biden joining TikTok.
• What we know about Russia’s alleged use of Starlink satellites.