HapnGoWhat Happens & What Goes
EN
English Română Italiano Deutsch Français Español Português

Interface in 7 languages. Any other language uses Google Translate, loaded only when you ask for it.

News

Models & Labs

New models, releases and benchmarks from OpenAI, Anthropic, Google, Meta, Mistral and the open-weight world.

627 stories · Every headline links to the original publisher. HapnGo curates and groups; we do not copy articles.

Models & Labs NVIDIA Technical Blog

Memory-Enforced GPU Sharing in Kubernetes with NVIDIA CUDA 13.4

AI summary NVIDIA has published GPU Fractioning Operator, an open-source Kubernetes tool hosted on GitHub under the KAI Scheduler organization, which works next to the NVIDIA GPU Operator to run MPS on each GPU node and apply memory caps per workload. It relies on CUDA 13.4, which adds application-transparent memory limits, including fixed sizes or a guaranteed minimum with a higher burst ceiling, plus a choice of time-slicing or sm-sharing compute. KAI Scheduler v0.18.0 is the first scheduler to support it through a mode called NvFractions, enabled with a Helm flag at install time. In one example, an inference server reserves 40Gi on each of four GPUs and can expand to 60Gi when memory is free.

Models & Labs The Decoder

Google Deepmind's Genie 4 puts a modern spin on agent designs that predate LLMs

AI summary Google Deepmind has unveiled Genie 4, a successor to the Genie 3 model shown in August 2025, which builds interactive worlds frame by frame in real time and now creates its own ambient sound and character speech. Gemini models work behind the scenes, with Gemini 3.8 Flash splitting goals into steps and adjusting plans as the scene or user input changes. The company calls it experimental, noting that details can drift in longer sessions. Only select testers have access for now, with Google AI Ultra subscribers due to follow in the Project Genie web app.

Models & Labs Hugging Face

Build specialized multimodal models with LFM3

AI summary Liquid has released LFM3, a family of open-weight models of roughly 2B parameters for text, vision, audio, omni and decision tasks, all derived from LFM3-1.7B-Base and each tunable on a single GPU. The base model was pre-trained on 22T tokens, supports a 256K context window, and the 1.7B text model beats LFM2.5-2.6B despite having 35% fewer parameters, scoring 38.42 versus 3.02 on WebArena-Verified. LFM3-VL-2B now accepts video and lifts its average over 62 vision benchmarks to 71.4 from 57.6, while LFM3-Omni-2B is the company's first model combining vision, audio and speech output. The models are available on Hugging Face along with fine-tuning notebooks.

Models & Labs Google DeepMind

Genie 4: A new generation of world simulation

AI summary Genie 4, a new world model from Google DeepMind, builds interactive simulations on the fly and pairs them with generated audio such as music, background sounds and characters talking to each other. It relies on several Gemini models: Gemini Omni makes scenes look and act realistically, while the others keep each simulation consistent. People move around with a keyboard and can type changes, like switching the weather or placing a person in the scene. For now only trusted testers can try this research preview; Google AI Ultra subscribers will later get it through Project Genie, as will early Gemini API and Google AI Studio partners.

Models & Labs Search Engine Journal

Looking Things Up Now Rivals Writing In ChatGPT, OpenAI Data Shows

AI summary OpenAI's Signals V3.0 data, published Oct. 26, shows that requests for information made up 20.7% of consumer ChatGPT messages in August, edging past writing at 20.6% for the first time since tracking began in July 2024. Writing regained the lead in September with 21.4% against 20.2%, while practical guidance stayed the biggest topic at 28.1%. The figures come from a monthly sample of 1M messages from adult users on Free, Go, Plus and Pro plans, excluding Enterprise and Codex. Signals shows a smaller rise in information seeking than a 2025 OpenAI and Harvard paper, which put the share at 24% by mid-2025 versus 18.3% in Signals for June 2025.

Models & Labs The Decoder

Former Deepmind researcher thinks AlphaGo's playbook can cure AI hallucinations

AI summary Thore Graepel, a former Deepmind researcher on AlphaGo, has founded Metis Reasoning and unveiled its approach at Nvidia's GTC conference in Berlin. Instead of storing facts in model weights, the startup wants an outside ledger in which a dedicated module checks every entry against proofs, calculations or empirical evidence, with AlphaGo-like search picking each next move. Metis promises fewer hallucinations and source-backed answers, though its white paper shows no functioning system or benchmark results so far. According to Bloomberg, Graepel is seeking a round in the low tens of millions of dollars and is preparing a later one worth several hundred million.

Models & Labs The Decoder

OpenAI's partner content arrives in ChatGPT pre-summarized while the rest of the web gets raw scraps

AI summary Rather than fetching pages from OpenAI's publishing partners in real time, ChatGPT is fed summaries written beforehand, whereas regular websites only get through as raw snippets of text. Konitzny examined 1,000 cases and found partner content placed before all web results in the prompt. In ranked lists, items in positions one to three were covered by the summaries 70 percent of the time, compared with only 23 percent for positions seven through ten. In 2024 OpenAI promised its partners greater visibility, and a later study from Hamburg showed Springer titles turning up much more frequently through the web version than through the API.

Free, daily

The HapnGo Briefing, in your inbox

One email a day with the AI stories that matter, why they matter, and one tool worth trying. No spam, one-click unsubscribe.

We only use your address to send the briefing and, if you vote, to count one vote per reader. Privacy.