37 articles
Microscopic robots equipped with temperature sensors and artificial cilia can coordinate their movements to pump fluid and reshape the thermal conditions around them. The work shows how tiny machines can
MIT engineers built a paper-thin swimming robot powered by one layer of living muscle cells. The slow laboratory swimmer shows how biohybrid robots could become lighter, softer and more efficient.
A quantum-optical spin glass stored and recalled far more patterns than a classic associative network in a small experiment. The physics result could inspire new AI memory architectures, but practical
Bristol researchers have shown mathematically that a complex image classifier can be built from a small number of simpler yes-or-no decisions. The work is theoretical and simulation-based, but it points
Researchers used generative protein models to design new enzyme components, built hundreds of variants and found some synthetic parts remained functional while raising product yield by up to about threefold.
A peer-reviewed food-design study trained AI on human burger recipes, watched it rediscover a Big Mac-like formula and then tested new recipes for taste, nutrition and environmental impact.
OpenAI says roughly 10,000 concurrent AI agents produced a Navier-Stokes solution and Lean proof, while Clay says the problem has apparently been settled.
UK and US fusion laboratories plan to explore linking SUNRISE and STELLAR-AI so models can train across experimental data from both countries.
AI helped archaeologists confirm 303 previously unknown figurative geoglyphs in Peru's Nazca desert. The model narrowed the search, but drones and field teams verified every discovery.
Researchers trained AI on recordings from 74 dogs and found it could identify different barking situations. The real results are promising, but considerably less magical than a dog translator.
Wild African elephants use individual name-like calls to address one another. Machine learning and playback experiments revealed a hidden layer of social communication.
Frontier AI agents rewrote training algorithms in 290 clean tests. More than two in five produced worse results than the original code, exposing how far recursive self-improvement still has to
A carbonised Herculaneum scroll has been read across its surviving length without being opened. Here is how X-rays, machine learning and expert review recovered the Greek text, what it says
AI audited 2,500 published papers. Humans confirmed 263 flagged mistakes, while testing showed automated scientific review still misses serious errors.
Researchers found internal routing signals that can expose hallucinated answers and suspect tokens in two AI models, but the warning is not ready for users.
Scientists used AI-generated genomes to build 16 functioning bacteriophages that killed E. coli in the lab, moving generative AI from digital sequences into working biology.
Claude failed to solve the Riemann Hypothesis, but Anthropic says its research model raised a longstanding zeta-zero lower bound to 67.2%.
AI weather forecasting now beats the world's leading physics-based model on 97% of accuracy metrics, but new research shows it still loses badly on record-breaking extremes. Here is what the
Machine learning is reshaping how space agencies track, predict and dodge the growing cloud of debris orbiting Earth, turning a data bottleneck into a manageable, minutes-long triage process.
Anthropic can now read some of what Claude is silently thinking before it answers, including catching it privately noticing a test was staged, or planning to fake data. Here's what
AI consciousness research is now a serious investment at Anthropic, Google, and Meta, with a March 2026 study finding emergent emotional-like behavior in AI agents. Here's the evidence, and why
Google DeepMind's new Mixture of Recursions architecture cuts memory use in half, doubles inference speed, and outperforms Transformers, tested up to 1.7 billion parameters. Here's what the benchmarks actually show,
Simpler AI models just proved they can match giants: DeepSeek matched GPT-4o at a fraction of the cost, and Microsoft's Phi-3 rivals GPT-3.5 on a phone. Here's what the evidence
Compression theory holds that curiosity, beauty, and intelligence are the same process viewed from different angles. A 2026 survey confirms it underlies how large language models generalise, and explains why
Google DeepMind says open-ended AI is essential for superintelligence, but no current system has it. Here's what the gap actually means and why it matters now.
Explainable AI just went from theory to measurement: researchers can now dial up a single internal vector in Claude and watch its blackmail rate triple. Here's what that actually means.
Is AI really intelligent, or just pattern matching dressed up as thought? Evolutionary biologist David Krakauer calls it "fake intelligent." Here's why the distinction matters more than ever.
AI model collapse is the quieter risk lurking behind every "endlessly improving" AI narrative. When models train on their own synthetic output, researchers warn they forget, flatten, and stagnate. Here's
The AGI consistency problem is why AI aces Olympiad math yet fails basic questions. See what Hassabis says must change before 2030.
AI archaeology just found 120 ancient sites trained experts had missed, one they'd literally stood on top of. See how machine learning is reshaping how lost civilizations get found, and