IBM Research
·
1 month ago
IBM researchers developed an evolutionary workflow guided by large language models to discover quantum error correction codes, which are mathematical constructions that protect quantum information by introducing redundancy. The framework discovered 465 new error correction codes, including one with a logical qubit count of 50 compared to the previous record of 16, and codes requiring only 72 physical qubits. The work demonstrates that LLMs can accelerate the exploration of quantum error correction code design space, potentially enabling researchers to more comprehensively understand trade-offs between code properties and identify candidates for fault-tolerant quantum computing systems.
Ben's Bites
·
1 month ago
Anthropic released Fable, a new Claude model described as a safer version of the unreleased Mythos model, which performs better than Opus 4.8 on benchmarks and is priced at 2x the cost of Opus. Fable is available only until June 22nd in Claude subscription plans before transitioning to paid credits, and the model shows particular strength in spawning multiple subagents while maintaining context. The release sparked controversy over Anthropic's policy to restrict the model's use for ML/AI work, which the company partially walked back after backlash.
IEEE Spectrum AI
·
1 month ago
Isomorphic Labs, a Google DeepMind spin-off, developed the Isomorphic Drug Design Engine to identify hidden protein binding pockets and predict drug-protein interactions beyond what existing AlphaFold models can achieve. The system successfully predicted a previously unknown cryptic pocket on cereblon protein that was only recently published in a Nature paper in January. The tool aims to make more disease-associated proteins tractable as drug targets by modeling interactions with small molecules, antibodies, peptides, and other therapeutic compounds.
AI Snake Oil
·
1 month ago
Software engineer employment has continued growing despite AI adoption, with a Federal Reserve study finding only a 3 percentage point annual slowdown in growth compared to pre-AI projections. High-profile layoffs attributed to AI (at Block, Snap, and Intuit) were actually driven by financial pressures, with New York's WARN Act filings showing only one company citing AI for layoffs affecting roughly 0.2% of workers. The claim that AI replacing code-writing will eliminate jobs misunderstands software engineering, where decision-making and delivery verification remain human bottlenecks even as AI compresses the execution layer.
Hugging Face Blog
·
1 month ago
PyTorch's profiler traces show that nn.Linear uses an optimized cuBLAS kernel (addmm) that folds bias addition into the matrix multiplication via an epilogue, eliminating the need for a separate add kernel. When torch.compile optimizes a single Linear layer, it removes only CPU dispatch overhead by pre-computing strides at compile time—the GPU kernel remains identical. In a three-layer MLP with GeGLU activation, the forward pass launches exactly five GPU kernels: three GEMMs for the linear projections plus one each for GeLU and element-wise multiplication.
OpenAI Blog
·
1 month ago
OpenAI has committed to supporting the EU Code of Practice on AI content transparency by developing provenance standards and tools. The company will work to help users identify and understand content generated by AI systems. This effort aims to build trust in AI systems across Europe by establishing clearer mechanisms for content attribution.
OpenAI Blog
·
1 month ago
BBVA deployed ChatGPT Enterprise across its workforce of 100,000 employees to integrate AI into core banking operations. The bank scaled the tool to all staff members and partnered with OpenAI to develop AI-powered banking services. BBVA plans to use this infrastructure to accelerate customer-facing applications and internal efficiency improvements across its global operations.
OpenAI Blog
·
1 month ago
Chi-kwan Chan, an astrophysicist, uses OpenAI's Codex to develop simulations that model black hole behavior and test predictions from Einstein's general relativity. Codex accelerates the code-writing process for complex computational physics models that would otherwise require substantial manual development time. This approach enables faster iteration on black hole simulations, allowing researchers to explore extreme physics scenarios more efficiently.
OpenAI Blog
·
1 month ago
OpenAI is acquiring Ona, a company that builds secure cloud environments for running code. The acquisition will integrate Ona's technology into Codex to support long-running AI agents that can persist state across enterprise tasks. This enables OpenAI's models to execute multi-step workflows without losing context between operations.
