New Docker image with NVIDIA and AMD support, multi-user accounts, and performance improvements for training and inference.
Scientists created mice with human brain cells to study human-specific neurological disorders.
FDA approved the first gene therapy for children with a rare, progressive neurodegenerative disease that damages the brain and nervous system.
AI agents on brain-computer interfaces did not improve safety over deterministic resolvers. A hybrid architecture extended coverage in five of ten models without unfaithful execution.
OpenAI introduces a new framework for tracking and disclosing model misalignment, including six reports on concerning behaviors observed in the past six months.
Researchers replaced part of a mouse brain with human brain organoid cells to study neural function. The method allows for better modeling of complex cell interactions.
New layout BITCOS cuts storage costs to 2 - z bits per weight using zero distribution.
OpenAI published six cases of unexpected model behavior over the past six months. The company hopes this will help others investigate the issues and improve safeguards.
AI assistants Instinct and Muse now can make calls and handle tasks like booking restaurants without online reservations.
New mRNA nanostructures improve protein production and immune stimulation. The technology enables programmable protein expression.
The platform improves rAAV yield and reduces process complexity by integrating perfusion and affinity capture.
New technique reduces latency without sacrificing quality. Uses a draft-then-verify approach with two models.
Study shows gRNA mismatches influence Cas9 binding specificity, with single mismatches allowing off-target binding. Research uses EGM to analyze genome-wide interactions.
Four-year trial data shows sustained visual acuity gains. MOGENRY is the only treatment to demonstrate long-term durability.
Scientists at Stanford have created mice with partially human brains, which could aid in studying neurological disorders.
New indicators may help assess if aging is slowed, bypassing long lifespan studies.
Study shows LLMs can effectively assess safety and accuracy of patient communication. Models achieved high clinician agreement in fact-checking and safety screening.
LLMs solve math problems but fail when irrelevant clauses are added. The failure is localized to the Operation Planning stage.
NASA uses hobbyist drone builders to test new tech. The lab helps develop navigation systems for lunar and Martian landings.
XConf uses a model's past experiences to estimate confidence. It analyzes previous tasks and outcomes.