Morning Edition · Monday, August 3, 2026
Tech
Alibaba Ships Qwen3.8-Max and Claims It Trails Only Anthropic's Top Model, Without Publishing the Numbers
The 2.4-trillion-parameter model went live with a vendor claim of "second only to Fable 5" and no benchmark table, model card, or license to verify it.
Tech
Anthropic's Opus 5 Matches Its Own Flagship on Coding at Half the Cost per Task
The company reports Opus 5 more than doubles Opus 4.8 on a software-engineering evaluation and reaches near-Fable 5 coding scores using a fraction of the reasoning tokens.
Markets
Berkshire's $339 Billion Treasury Position Is the Bear Case on AI Capex That Buffett Won't Say Directly
With Berkshire's short-term Treasury bills earning about $20 billion a year, the firm's cash stance is a direct comparison against the hundreds of billions labs are spending on models and data centers.
Tech
A 6,000-Line C Engine Claims to Run the Full Kimi K3 Weights on a 64-Gigabyte Laptop
The method is streaming a roughly one-terabyte model from a solid-state drive (SSD) rather than holding it in memory, and the open question is throughput, not whether it loads.
Tech
A New Paper Names the Networking Bottleneck No Disaggregated Inference System Solves Correctly
When the prefill and decode stages run on separate graphics-processing-unit (GPU) pools, moving the key-value cache between them becomes a data-center data-movement problem, and the authors argue current systems handle it incorrectly.
Tech
Researchers Propose a Pipeline That Uses Language Models to Generate and Validate Mathematical Conjectures
The framework aims to systematize the intuition-heavy step of proposing conjectures worth proving, with the stated ambition of identifying candidates in the class of major open problems.
Tech
A Benchmark Study Asks Whether AI Can Judge the Quality of AI-Generated Research
The proposal uses automated multi-model review to score autonomous research systems, confronting the problem that evaluation, not generation, is now the difficult part.
World
Study Finds 40 Percent of Top TikTok Health Videos Are AI-Generated, Rising to 84 Percent for 'Health Tips' Searches
Researchers say synthetic clips featuring fabricated doctors average 2.5 million views each and are shared more often than human-made ones.
Tech
Meta Opens a Paid Frontier API With Muse Spark 1.1, Ending Its Open-Only Posture
The multimodal agentic model is available through an OpenAI- and Anthropic-compatible API priced at $1.25 and $4.25 per million input and output tokens.
Tech
Meta Puts Segment Anything and DINO Into National-Lab Science Projects
The open-vocabulary perception stack is being applied to Department of Energy research and to assistive robotics, extending promptable vision beyond fixed label sets into production.
Tech
Paper Proposes Cross-Model Auditing to Harden LLM Judges Against Their Own Biases
Chain-of-Models routes a judgment through multiple models to identify the cognitive biases that prompt-based debiasing fails to fix.
Crypto
Researchers Show Wallet Transaction-Simulation Previews Can Be Spoofed to Phish Crypto Users
The safety feature that lets wallets like MetaMask preview a transaction's effect becomes an attack surface when the preview can be manipulated to differ from what executes.