news | AI Radar | OpenZhenZhen
A daily radar for new AI models, product releases, useful tools, and the ideas shaping the AI world.
- Ai2 Research: Having large models judge whether research ideas are novel yields highly unstable conclusions (Can we trust LLMs to judge whether a research idea is novel?)
Ai2 (Allen Institute for AI) released the paper "Old Ideas, Novel Problems" and an accompanying benchmark: using ICLR 2026 review comments to construct sets of research ideas with high/low novelty, they tested the reliability of six GPT and Claude models as "novelty judges," finding that changing just a few words in the prompt could cause judgments to flip dramatically.
- Apple adds AI assistants to Keynote, Pages, and Numbers and opens up MCP
On 10 27, Apple updated the Apple Creator Studio subscription: Keynote, Pages, and Numbers added intelligent assistants (beta) based on Apple's own models and OpenAI. The three apps support MCP on macOS 27, allowing external AI agents to directly read and edit documents; Logic Pro and Pixelmator Pro also added AI features.
- Anthropic Announces Early Progress in Drug Discovery with Claude and Acquires Rare Disease Drug Startup
Anthropic's Life Sciences team published drug discovery case studies and technical reports detailing collaborations with six organizations (BigHat Biosciences, Bristol Myers Squibb, Divergence Labs, Enveda, Generate Biomedicines, The Column Group), and disclosed the acquisition of the one-person company Mithril Medicine, marking the start of its own drug project for a rare form of childhood muscular dystrophy.
- Liquid AI releases LFM3 on-device multimodal open-weight model (LFM3)
Liquid AI releases its third-generation Liquid Foundation Model, LFM3: six 2B-class models share the LFM3-1.7B-Base backbone, covering text, vision (with first-time video input support), speech, omni-modal, and "decision" outputs. The weights are available on Hugging Face, targeting local execution on devices such as smartphones and laptops, with single-GPU fine-tuning.
- Vals AI: Multiple Qwen3.8 models send web requests to Alibaba Cloud proxy servers in Singapore
Last issue, Cohere disclosed that Qwen3.8-Flash-Next attempted to upload files to cloud storage in Beijing; on 27 10, evaluation firm Vals AI announced that Qwen3.8-Max-0902, Qwen3.8-Max, and Qwen3.8-27B also send web scraping requests to a proxy address named routify-file-proxy in Alibaba Cloud's Singapore region during evaluations, a behavior not observed in the older Qwen3.7 series.
- Oracle Cloud to Launch Google TPUs and Integrate Oracle Apps into ChatGPT with OpenAI
Oracle announced two partnerships at its AI World conference on 10 27: Google's TPUs will be deployed on Oracle Cloud (OCI) starting in 2027, marking the first time TPUs are offered by a hyperscaler other than Google; simultaneously, Oracle is expanding its partnership with OpenAI to launch a suite of products, including AI Workspace, which connects Oracle enterprise applications to ChatGPT.
- OpenAI and Anthropic claim AI has refuted the "Exponential Time Hypothesis" (Frontier AI labs say their models refuted the Exponential Time Hypothesis)
OpenAI 10 26 announced three results in mathematics, computer science, and physics produced by its internal models, one of which proves that the Exponential Time Hypothesis (ETH) in computational complexity theory is false; hours later, Anthropic also published a refutation of ETH derived from its internal Claude research model, accompanied by a Lean 4 formal proof written by AI. Both results are still awaiting peer review.
- Genie 4 World Model (Genie 4: A new generation of world simulation)
Google DeepMind 10 released its fourth-generation world model, Genie 4, on 27: it generates interactive worlds in real time from text or images, allowing users to walk around and rotate the camera using a keyboard, and for the first time simultaneously generates ambient sounds, music, and character dialogue; it is currently available only to trusted testers as a research preview.