Markets · Large Language Models
NPR Newsletter Weighs Whether AI Bubble Bust Could Speed Job Losses
NPR examines whether an AI bubble burst could cause a recession that speeds AI-driven job losses.
Latest news, developments and reporting on Large Language Models.
NPR examines whether an AI bubble burst could cause a recession that speeds AI-driven job losses.
Starling partners Nvidia for a transactional foundation model, joining Revolut and Lendable in Europe's fintech AI race.
TechRadar analysis warns AI wrappers risk losing their moat when model providers absorb their features, citing Jasper.
Smart TV owners use Claude Code and ChatGPT Codex to remove ads and bloatware, Android Authority reports.
iQIYI is embedding generative AI across viewing and production, from a ‘Skip Watch’ feature that compresses long-form drama to script-evaluation and image-generation tools used by its studios.
Cloudflare previewed Clairic on Workers AI, an extraction model returning typed values in one pass; a test claims pipeline fell to 19.3 seconds from 61.7.
Trump administration bars civilian federal agencies from DeepSeek technology and procurement, citing security risks, DHS says.
MemTensor has released and open-sourced MemT 1.0, a large model that keeps per-user memory inside its architecture instead of an external retrieval layer, claiming gains on long-horizon memory and coding benchmarks.
Thore Graepel, ex-DeepMind and AlphaGo researcher, launches Metis Reasoning to build reasoning AI.
Next Claude reportedly targets root causes and reads intent better; Google and OpenAI ready new models.
Inception released Mercury 3, an API-only diffusion LLM it says runs at 1,068 tokens per second with a 1M-token context and leads GPT-6 Luna and Claude Haiku 5.5 on 8 of 12 own tests.
Dream Labs raised $90M and launched a public API for Dream-1, a robot model trained on video and language.
Nace AI has released Drex Max, a multimodal decision model for computer-use and browser agents that reports 67.59 on the public Decision Index 0.3.2; MiniMax also launched M3.1 Flash with faster inference and lower agent-task costs.
Google’s Gemini 4 Argon helped researchers extend exact mathematical solutions for magnetic confinement fusion, producing candidates within hours but not proving a commercial fusion breakthrough.
Hugging Face's Solaiman says embedded AI evaluators are no easy fix, citing reproducibility and standards gaps.
doubleAI raised $200M in a Series A with Nvidia and Dell participating; CTech valued it above $1B.
MiniMax released M3.1 Flash, claiming 70% lower agent-task costs and nearly triple the inference speed of M3.
Open-source early-reading app offers tap-to-sound phonics and MCP-based import of owned books via AI chatbots.
Nick Cammarata’s blog post argues LLMs may be easier to understand than assumed, drawing praise and skepticism.
Laminar launched flow-1, an RL-trained model for agent trace analysis, claiming GPT-6-sol-level detection 23x cheaper.
Nvidia launched open-weight Nemotron 3.5 Super VL for enterprise agents and Bolt models for local AI PCs.
Kalshi gives Anthropic 75% odds of a 2026 IPO announcement, the only major company above 50% as rivals delay.
GPT-6 Astra decoded an 1809 military letter in six hours; Claude traced five NSA elliptic-curve seeds in 30 minutes.
MiniMax launches M3.1 Flash, a 428B MoE model with 1M context at $0.15/M input; benchmarks are vendor-reported.