Interconnects
·
3 months ago
The author released the ATOM Report, an updated technical analysis of the open language model ecosystem that tracks adoption metrics across U.S. and Chinese models using a Relative Adoption Metric (RAM) score. The RLHF book was sent to production in the last week and will print in approximately 2 months, with pre-orders available on Amazon and Manning. The author is developing a free YouTube lecture series and course platform to expand the book into a complete learning experience for people transitioning to expertise in post-training.
Amazon Science
·
3 months ago
AWS and Johns Hopkins Engineering launched the Antibody Developability Benchmark, a public database for evaluating AI and machine learning models used in antibody drug design. The dataset contains 50 seed antibodies across four structural formats targeting 42 antigens, with up to 99 engineered variants per seed, making it 20 times more diverse than existing public benchmarks. The benchmark enables researchers to rigorously compare computational antibody prediction models against experimentally validated data before expensive wet-lab testing, potentially accelerating therapeutic antibody discovery and reducing costs.
OpenAI Blog
·
3 months ago
● 2 sources
OpenAI has expanded its Trusted Access for Cyber program by introducing GPT-5.4-Cyber, a specialized model available to vetted cybersecurity professionals. The model is designed specifically for defensive security work and includes enhanced safeguards to prevent misuse. This expansion allows approved organizations to access AI tools for threat detection and defense while maintaining access controls around the technology.
