Photonic Quantum Computing Hotter than Ever

The idea of using quanta of light – the photons – as computational units for quantum computing – the qubits – has been around for a long time, but it has only been in the past five years or so that technology has caught up with ambition. The first microscopic chips for single quantum processing have now been developed, and rudimentary computational power has been confirmed. With that step, photonic computing, as it’s called, has rapidly gained in popularity because it is believed to be much easier to scale up to high numbers of qubits by joining chips. This approach is most prominently pursued by the American company PsiQuantum. They just released a new paper in which they detail how to manufacture a scalable quantum computing platform. They also just received AUD 940 Million (USD 614 Million) from the Australian government. Definitely a company to keep an eye on.

This episode of Science News covers bird flu. The bird flu has spread to cattle in the United States. Inactive fragments of the virus have been found in milk. The World Health Organization rates the current risk for humans as low but epidemiologists all over the world are on high alert. I have the summary. You can now leave comments on our quizzes!

European Union is Behind with Electric Vehicle Chargers

The European Automobile Manufacturers' Association (ACEA) estimates that by 2030 the E.U. will need 8.8 million public charging places for electric vehicles, requiring an average 1.2 million new ones per year until 2030. New installations in 2023 were about 153k. This adds to the previously voiced worries by the International Energy Agency that the electric grid is the bottleneck in the clean energy transition. Full ACEA report here.

Use Your Technology For Something Good

Brilliant’s learning tools are designed by experts to be uniquely effective. Level up with real knowledge in math, data analysis, programming, and AI. Daily learning is the smartest use of your smartphone – you can set goals, see your progress, and join millions of people around the world on Brilliant’s easy-to-use app. Start your 30-day free trial now and get an extra 20% off a premium annual subscription.

Chatbots Think Mostly in English

Large Language Models need to be trained on vast amounts of text samples. These training data are currently mostly drawn from the internet, and dominated by English language sources. A team of researchers from the Swiss Federal Institute of Technology in Lausanne now used Meta’s open-source model Llama to study what happens if you query such a model in a different language than English. They found that these queries most often go through the part of the model that processes English. In other words, when looking for contextual understanding, the language models default to the biggest training body, which is English. The team only looked at Meta’s language model, but it is probably similar for the other currently available chatbots.This finding makes sense from a computational perspective, but brings the risk that grammatical constructs and verbal nuances of other languages can’t be put into proper context. It’s an issue to keep track of because chatbots are increasingly being used to translate language into code, which will bring in the imprecision of spoken language in general, English in particular. Paper here.

Our podcast “Science with Sabine” is available on Spotify, Amazon, Apple, Castbox, Google Podcasts, Pocket Casts, Radio Public, and YouTube.