AI Reasoning, Two Origins of Life, & The Riemann Hypothesis

Researchers at the Chinese tech company Alibaba have found that artificial intelligence models often answered difficult maths and science questions by guessing and verifying, rather than straight-forward reasoning. They tested 11 leading artificial intelligence models (including GPT-5.5, Claude Opus 4.7, Gemini 3.1 Pro Preview, and Alibaba’s own Qwen 3.7 Max) on difficult maths, physics and chemistry questions. The six best-performing models obtained 8% to 24% of correct answers by guessing the answer and then verifying it. This method is logically valid, but it is only likely to succeed if similar answers have been in the training data. The result therefore suggests that looking only at correct answers might overestimate the reasoning abilities of the current large language models. It’s probably correct, 2/10 on the bullshit meter.

Paper here.

Researchers from Heinrich Heine University Düsseldorf in Germany say that life had two origins. According to their hypothesis, the common ancestor of all forms of life depended on minerals in hot seafloor vents to survive. This ancestor then split into two types, bacteria and archaea, each of which was still dependent on the vents’ minerals. Each group then separately evolved the ability to live independently, without the vents’ minerals. Much later, a descendant of the archaea absorbed a bacterium, which became the energy-producing part of complex cells that eventually gave rise to plants, animals, and humans.

The researchers say they inferred this by comparing proteins from almost 1,000 modern microorganisms. They also supported their hypothesis by using the metal palladium to reproduce two chemical reactions needed for life in warm alkaline water. But palladium is extremely rare and their experiments required unnaturally high concentrations of another chemical, so the scenario is rather speculative. 7/10 on the bullshit meter, more evidence is needed.

Paper here. Press release here.

Researchers from the Beijing Academy of Quantum Information Sciences in China have proposed using a quantum computer to search for zeroes of the Riemann function, potentially settling a 167-year-old claim about the distribution of prime numbers. They claim their algorithm can vastly speed up the calculation of the zeroes if the number of qubits becomes large enough. As a demonstration, the researchers programmed just 5 qubits and reproduced the first five known zeroes of the Riemann function. This method could solve the Riemann hypothesis in case the conjecture is false by finding a counterexample.

Paper here.