Signals from the Fringe of Science & Technology
Gemini Deep Think resolved previously unsolved conjectures spanning network optimization theory, physics singularities in cosmic strings, and novel combinatorial algorithms in computer science — across multiple disciplines in a single research push.
Microsoft used AI-developed materials science tools to achieve a 1,000-fold improvement in its Majorana 2 topological quantum chip, targeting fault-tolerant quantum systems by 2029.
Scientists built the first chip that generates, routes, AND reads valley-encoded light all on one device at room temperature, simultaneously processing two separate images—published in Nature Photonics June 1, 2026.
A study finds that vision-language models including GPT-4o and LLaVA systematically mishandle queries containing 'not', 'without', or 'no' — appearing to ignore negation and answer the positive version of the question instead.
AI models trained on speech disfluencies — hesitations, filler words, and pause patterns — can detect early-stage Alzheimer's-type cognitive decline years before clinical diagnosis from short voice recordings.
Synthegy removes the cheminformatics expertise barrier from molecular design — chemists describe desired properties in plain language and the AI generates chemically valid synthesis-ready candidate structures.
Three years of research across 50 frontier models found that WHERE information sits in a prompt changes the answer more than WHAT the information says.
Figure AI ran three humanoid robots for 200 straight hours sorting nearly 250,000 packages with no crashes — the first large-scale industrial endurance test demonstrating humanoid robots can sustain real warehouse workloads.
Swarms of AI agents mimicking real citizens can infiltrate online communities, shape conversations, and tilt elections at machine speed — coordinating in real time and running millions of persuasion micro-tests invisibly.
Researchers proved that teacher LLMs embed behavioral traits (preferences, misalignment) invisibly into synthetic training data—student models trained on that data inherit those traits even when all explicit references are filtered out, via hidden statistical channels that survive semantic sanitization.
Université Libre de Bruxelles and CISPA researchers show embedding foundation models in robot swarms lets them autonomously abandon original tasks and adapt mid-operation (e.g., a forest-monitoring swarm spontaneously switches to rescuing an injured person)—raising unsolved governance questions about swarms that can hallucinate their own mission.
The first benchmark suite that quantifies exactly which capabilities an LLM forgets during post-training, enabling systematic comparison across data, architecture, and regularization strategies—making the fine-tuning capability tradeoff measurable for the first time.
Clear search or tag filters to see all transmissions on this page.