Podcast Episode
NewsPodLM Daily Podcast 29 Jan 2026
January 29, 2026
Audio archived. Episodes older than 60 days are removed to save server storage. Story details remain below.
This week's AI news features SoftBank potentially pushing OpenAI's valuation to eight hundred and thirty billion dollars, Yann LeCun dramatically leaving Meta to chase world models, Google Search getting a Gemini 3 upgrade, and a stunning revelation that over one hundred fake citations slipped through peer review at NeurIPS.
SoftBank Eyes Record Investment in OpenAI
SoftBank Group is in discussions to invest an additional thirty billion dollars in OpenAI, which would push the AI company's valuation to approximately eight hundred and thirty billion dollars. This comes just months after SoftBank completed a forty one billion dollar investment giving it an eleven percent stake. The funding round could raise up to one hundred billion dollars total, with Amazon and Nvidia also reportedly in talks to participate.Yann LeCun Exits Meta After Clash with Zuckerberg
After more than a decade as Meta's chief AI scientist, Yann LeCun has left the company to launch Advanced Machine Intelligence Labs, targeting a three billion dollar valuation. LeCun cited frustration with being forced to report to Alexandr Wang, the former Scale AI CEO who now leads Meta's Superintelligence Labs after a fourteen billion dollar acquisition deal. LeCun maintains that large language models are a dead end for achieving superintelligence and is betting on world models trained on video and spatial data.Google Search Defaults to Gemini 3
Google has made Gemini 3 the default model for AI Overviews globally, now reaching over two billion users monthly. The upgrade brings conversational follow-up capabilities directly from search results, allowing users to seamlessly transition into AI Mode for deeper exploration. Gemini 3 can dynamically create visual layouts with interactive tools and simulations tailored to each query.NeurIPS Papers Found to Contain Over One Hundred Hallucinated Citations
AI detection company GPTZero scanned all four thousand eight hundred and forty one papers accepted at NeurIPS twenty twenty five and found over one hundred confirmed hallucinated citations across fifty one papers. These fabricated references, including fake author names and non-existent journals, survived peer review at one of the world's most prestigious AI conferences. The finding highlights growing concerns about AI-generated content infiltrating scientific literature.AI2 Releases Open Source Coding Agents
The Allen Institute for AI has released SERA, an open-source coding agent family that solves fifty four point two percent of SWE-Bench Verified problems. The breakthrough allows anyone to train their own coding agent for about four hundred dollars, making AI coding assistance accessible to researchers and smaller organisations. The models are optimised for integration with Claude Code out of the box.Moonshot AI Launches Kimi K2.5 with Agent Swarm
Chinese AI company Moonshot AI has released Kimi K2.5, a multimodal model capable of self-directing swarms of up to one hundred sub-agents across fifteen hundred coordinated tool calls. The open-weight model was trained on fifteen trillion vision-language tokens and can understand screenshots, designs, and generate code from visual specifications.Published January 29, 2026 at 12:30pm