- Sabine's Newsletter
- Posts
- Fast Fibres, Hallucination Explanation, Carbon Storage, and the Mystery of Units
Fast Fibres, Hallucination Explanation, Carbon Storage, and the Mystery of Units
This week’s science bits from SWTG
Marcus Gee-Lim
September 10, 2025
|
Optical Fibre Breakthrough Promises Better Data Transfer For Everyone
Artist’s illustration of the new optical fibre design. Credit: Prof Francesco Poletti and Dr Greg Jasion, University of Southampton.
Researchers from Microsoft Azure Fiber and the University of Southampton have presented a new type of hollow electrical fibre that could seriously advance long-distance data transfer. In their recently published paper, the researchers do not hold back on the enthusiasm: “We have reported what we believe to be one of the most noteworthy improvements in waveguided optical technology for the past 40 years,” they write.
The new fibre uses 5 nested cylinders inside the main fibre that trap light. In it, light is transmitted through air. This, the researchers say, cuts down in-medium losses by almost 50%, meaning that repeater stations (which re-amplify the signal) could be spaced further apart. The fibre also allows higher power transfer. Paper here. More here.
This week’s episode of Science News is about units of measurement. Units make the difference between maths and reality, between platonic ideals and physical quantities. But what are units, really? This is one of the most underrated questions in the foundations of physics – let’s take a look. This week’s episode comes with a companion article on Nautilus. Check it out here.
Also, you can now create and share your own quizzes on QuizWithIt for free! Each quiz has a unique URL, can be embedded into websites or newsletter, and be shared on social media. Happy quizzing!
A Solution To AI Hallucinations?
Researchers from OpenAI say they have found a simple explanation for hallucinations in Large Language Models (LLMs) and a solution, too. Hallucinations are those cases when the models confidently produce text about nonexistent entities or events. I recently encountered a severe case with Gemini 2.5, which made up science news items, all including quotes and references. But I swear this one is real…
The OpenAI researchers claim that hallucinations come down to bad training, that is, the models aren’t rewarded for admitting uncertainty. Consequently, they say, the problem could be much alleviated with better feedback.
I suspect that this will probably indeed weed out the most obvious hallucinations. I strongly doubt, though, that it will solve the underlying problem, which is that text allows different constructions of reality that are equally “correct” for what words are concerned. I believe it will take “world models” – as they have been dubbed – to solve this problem by constructing predictive models of reality. Paper here. More here.
Daily News for Curious Minds
Be the smartest person in the room by reading 1440! Dive into 1440, where 4 million Americans find their daily, fact-based news fix. We navigate through 100+ sources to deliver a comprehensive roundup from every corner of the internet – politics, global events, business, and culture, all in a quick, 5-minute newsletter. It's completely free and devoid of bias or political influence, ensuring you get the facts straight. Subscribe to 1440 today.
Carbon Storage Is Much More Difficult Than We Thought
A study led by researchers from NASA found that the world doesn’t have remotely enough safe places to store carbon dioxide to realize some of humanity’s more ambitious carbon capture plans. These plans typically assume that carbon dioxide, once captured, would be pumped under pressure underground into suitable rocks and sealed there. But finding rock formations where the carbon dioxide does not leak back out isn’t all that easy. The researchers also excluded any locations nearby human settlements due to the risk inherent in such storage, and any locations that would make storage too costly. They conclude that the places left could effectively reduce global warming by about 0.7°C if fully used – far less than the 5-6°C assumed to be possible in some scenarios (especially those favored by the fossil fuel industry).
While it’s good to quantify the problem, I think these numbers are redundant because the technology to remove carbon from the atmosphere (or even capture it effectively at fossil fuel power plants) is currently so inefficient and costly it won’t make any difference for global warming, certainly not in the near future, and quite possibly not ever. Press release here. Paper here.
