Figure AI is launching a Center for Humanoid Safety for transparency, testing, and third-party certification.
Figure AI announces plans for a Center for Humanoid Safety, focusing on transparency, rigorous testing, and third-party certification for safer robots.
While DeepSeek makes AI cheaper, seemingly without cutting corners on quality, a group is trying to figure out how to make tests for AI models that are hard enough. It’s ‘Humanity’s Last Exam’
If you’re looking for a new reason to be nervous about artificial intelligence, try this: Some of the smartest humans in the world are struggling to create tests that AI systems can’t pass.
For years, AI systems were measured by giving new models a variety of standardized benchmark tests. Many of these tests consisted of challenging, SAT-calibre problems in areas like math, science and logic. Comparing the models’ scores over time served as a rough measure of AI progress.
Researchers from Zhejiang University and HKUST (Guangzhou) have developed a cutting-edge AI model, ProtET, that leverages multi-modal learning to enable controllable protein editing through text-based instructions. This innovative approach, published in Health Data Science, bridges the gap between biological language and protein sequence manipulation, enhancing functional protein design across domains like enzyme activity, stability, and antibody binding.
Proteins are the cornerstone of biological functions, and their precise modification holds immense potential for medical therapies, synthetic biology, and biotechnology. While traditional protein editing methods rely on labor-intensive laboratory experiments and single-task optimization models, ProtET introduces a transformer-structured encoder architecture and a hierarchical training paradigm. This model aligns protein sequences with natural language descriptions using contrastive learning, enabling intuitive, text-guided protein modifications.
The research team, led by Mingze Yin from Zhejiang University and Jintai Chen from HKUST (Guangzhou), trained ProtET on a dataset of over 67 million protein–biotext pairs, extracted from Swiss-Prot and TrEMBL databases. The model demonstrated exceptional performance across key benchmarks, improving protein stability by up to 16.9% and optimizing catalytic activities and antibody-specific binding.
A team of investigators from Dana-Farber Cancer Institute, The Broad Institute of MIT and Harvard, Google, and Columbia University have created an artificial intelligence model that can predict which genes are expressed in any type of human cell. The model, called EpiBERT, was inspired by BERT, a deep learning model designed to understand and generate human-like language.
The work appears in Cell Genomics.
Every cell in the body has the same genome sequence, so the difference between two types of cells is not the genes in the genome, but which genes are turned on, when, and how many. Approximately 20% of the genome codes for regulatory elements determine which genes are turned on, but very little is known about where those codes are in the genome, what their instructions look like, or how mutations affect function in a cell.
To learn more about this book, visit: https://theconsciousnessexplosion.ai/
To design their improved materials, Serles and Filleter worked with Professor Seunghwa Ryu and PhD student Jinwook Yeo at the Korea Advanced Institute of Science & Technology (KAIST) in Daejeon, South Korea. This partnership was initiated through U of T’s International Doctoral Clusters program, which supports doctoral training through research engagement with international collaborators.
The KAIST team employed the multi-objective Bayesian optimization machine learning algorithm. This algorithm learned from simulated geometries to predict the best possible geometries for enhancing stress distribution and improving the strength-to-weight ratio of nano-architected designs.
Serles then used a two-photon polymerization 3D printer housed in the Centre for Research and Application in Fluidic Technologies (CRAFT) to create prototypes for experimental validation. This additive manufacturing technology enables 3D printing at the micro and nano scale, creating optimized carbon nanolattices.
Although Chinese AI, such as DeepSeek, might torpedo American megatech, the advent of vanishingly tiny costs might lead us further up that exponential curve to Superabundance.
S V3 model, DeepSeek-V2, triggered an AI model price war in China after it was released last May. ‘ + The fact that DeepSeek-V2 was open-source and unprecedentedly cheap, only 1 yuan.
⏩($0.14)⏪ per 1 million tokens👀 ‼️
S cloud unit announcing price cuts of up to 97% on a range of models. ‘.
BEIJING (Reuters)-Chinese tech company Alibaba on Wednesday released a new version of its Qwen 2.5 artificial intelligence model that it claimed surpassed the highly-acclaimed DeepSeek-V3. The unusual timing of the Qwen 2.5-Max’s release, on the first day of the Lunar New Year when most Chinese people are off work and with their families, points to the pressure Chinese AI startup DeepSeek’s meteoric rise in the past three weeks has placed on not just overseas rivals, but also its domestic competition. “Qwen 2.5-Max outperforms… almost across the board GPT-4o, DeepSeek-V3 and Llama-3.1-405B,” Alibaba’s cloud unit said in an announcement posted on its official WeChat account, referring to OpenAI and Meta’s most advanced open-source AI models.