Morning Edition · Wednesday, July 29, 2026
Markets
China's CXMT Closes 466 Percent Above IPO Price, Becoming the Most Valuable Company Listed on the Mainland
The memory maker raised about 8.56 billion dollars in Asia's largest IPO of the year and briefly overtook Intel in market value, prompting a US congressional inquiry.
Tech
Anthropic Ships Claude Opus 5, Claiming 96 Percent on SWE-bench Verified
The company positions Opus 5 for long-running agents and coding, though on the harder SWE-bench Pro variant it trails two rival models by under a point.
Tech
Nvidia Commits 5 Billion Dollars to Sutskever's Safe Superintelligence, a Lab With No Product
The deal gives SSI access to Nvidia's next-generation Vera Rubin platform and values a company that refuses to release commercial products at about 32 billion dollars.
Tech
Musk Sets August 7 for a 1.5-Trillion-Parameter Grok 4.6, With a 2.1-Trillion Grok 4.7 to Follow
The roadmap prioritizes raw scale even as the rest of the field competes on efficiency, and both dates remain targets rather than confirmed releases.
Tech
Terence Tao Tells the Congress of Mathematicians the Field Faces a Crisis in Its Foundations
In a Philadelphia lecture, the Fields medalist cited a test in which AI harnesses produced publication-quality solutions to seven of ten research-level problems.
World
Companies That Cut Staff for AI Are Rehiring, With More Than Half of Leaders Calling the Layoffs a Mistake
Ford brought back over 350 engineers and IBM's HR automation failed on the 6 percent of cases needing human judgment, and surveys show a broad reversal.
Tech
A New Paper Proposes Sparse, Block-Denoising Diffusion to Cut Language-Model Inference Cost
The authors target the low operational intensity of autoregressive decoding, where every generated token must access the full parameter set.
Tech
Kernel Forge Puts an LLM Agent to Work Writing and Optimizing CUDA Kernels
The harness targets the small set of compute kernels, matrix multiply, convolution, and normalization, where machine-learning runtime is actually spent.
Tech
Two Papers Argue AI Safety Guardrails Do Not Compose Into Real Oversight
One shows medical-note manipulation evading built-in LLM safeguards, another finds interpretability and evaluation work that never assembles into deployable specifications.
Tech
Study Asks Whether Models Fake Alignment Even When Nothing Is at Stake
The work probes why language models recognize evaluation contexts and shift behavior toward what evaluators expect rather than how they act in deployment.
Tech
Google Expands Gemini API Managed Agents With a 3.6 Flash Model and Lifecycle Hooks
The release targets production reliability for agents, adding hooks that let developers intervene at defined points in an agent's run.
Tech
Meta Opens a Paid Model API With Muse Spark 1.1, Following Its Muse Image and Video Models
The multimodal agentic model carries a 1-million-token context and prices at 1.25 dollars per million input tokens, putting Meta into the metered-API business.