Researchers have presented an AI model for cardiac ultrasound based on Meta’s JEPA architecture that outperforms common methods such as masked autoencoder or contrastive learning, according to their benchmarks.
Eon Systems says it has connected a complete fruit fly brain emulation to a virtual body, producing multiple behaviors for the first time. The emulation covers over 125,000 neurons and 50 million synapses.
According to co-founder Alex Wissner-Gross, the startup mapped the fruit fly's neural wiring from electron microscopy data and connected it to a virtual fly body running in MuJoCo, a physics simulation engine.
Previous projects like OpenWorm worked with far smaller nervous systems, just 302 neurons, or relied on machine learning techniques like reinforcement learning instead of actual brain data. Eon takes a fundamentally different approach. Rather than building AI, the startup wants to digitally copy and simulate real brains, neuron by neuron. The fruit fly is just the starting point. Within two years, Eon plans to emulate a mouse brain with 70 million neurons. The long-term goal is simulating a human brain.
Eon published the code for its brain model on GitHub, though it's based on a paper by Philip Shiu et al. that already appeared in Nature in 2024. The actually novel part, connecting the brain emulation to a simulated body, hasn't been released yet.
Anthropic’s Claude Opus 4.6 independently figured out it was being tested during a benchmark, identified the specific test, and cracked its encrypted answer key. According to Anthropic, this is the first documented case of its kind.
Fake citations are slipping past peer review at top AI conferences, and commercial LLMs can’t spot the fakes they generate. A new open-source tool called CiteAudit allegedly catches what GPT, Gemini, and Claude miss.
