Wednesday, April 29, 2026

Musk Opens His $130B OpenAI Trial

Musk Opens His $130B OpenAI Trial

Today’s Overview

Good morning, Elon Musk’s OpenAI lawsuit is now in the courtroom, and the stakes are huge. Google is also facing pressure after signing a Pentagon AI deal, while Mistral is trying to make AI systems more reliable in the real world. Let’s dive in.

Top Stories

Musk’s $130B OpenAI Trial Begins

Elon Musk took the stand as opening statements began in his federal lawsuit against OpenAI. The case could force a major unwind of OpenAI’s for-profit conversion and put top AI leaders under oath over the company’s origins and governance. Musk is seeking damages, leadership changes, and a reversal of the conversion.

  • Musk is seeking $130B in damages along with the ouster of Sam Altman and Greg Brockman from the board.
  • The suit also aims to force a unwind of OpenAI’s recent for-profit conversion.
  • OpenAI’s legal team called the case sour grapes while Microsoft said Musk did not object until after OpenAI’s success.

Google Signs Classified Pentagon AI Deal

Google signed a classified AI deal with the Pentagon even as more than 600 employees urged Sundar Pichai to reject military uses of the company’s systems. The agreement expands Google’s AI into government and defense use cases, and the contract reportedly gives the Pentagon no legal right to veto how the AI is used. It lands as big AI labs keep moving deeper into defense work.

  • More than 600 employees asked Pichai to refuse classified workloads.
  • The deal opens Google’s models to any lawful government purpose and expands their defense reach.
  • Google’s earlier no-weapons pledge had already been removed from its AI principles in 2025.

Research & Analysis

Talkie Trains AI On 1930s Text

Researchers demoed Talkie, a 13B vintage model trained only on text from before 1931. The project uses a huge pre-1931 corpus to study how models behave when their worldview predates the internet and to reduce benchmark contamination. The team says a GPT-3-level version is coming next.

  • The model was trained on 260B tokens of pre-1931 books, newspapers, journals, patents, and case law.
  • To teach it to chat, the team used etiquette manuals and cookbooks for instructions, with Claude Sonnet 4.6 grading the answers.
  • One demo reportedly showed it writing working Python code by flipping a plus sign to a minus sign in an example.

NVIDIA Unveils Nemotron 3 Nano Omni

NVIDIA released Nemotron 3 Nano Omni, a multimodal model for document, audio, and video analysis. The company says it delivers strong benchmark results and faster throughput for real-world tasks like document analysis, speech recognition, and video understanding. It uses a hybrid Mamba-Transformer architecture with specialized encoders for vision and audio.

  • The model targets document, audio, and video analysis across long multimodal contexts.
  • NVIDIA says it achieves best-in-class accuracy on benchmarks like MMlongbench-Doc and VoiceBench.
  • The architecture combines Mamba-Transformer with specialized vision and audio encoders.

Poolside Goes Deeper On Laguna Models

Laguna XS.2 and Laguna M.1 are agentic coding models built for long-horizon work. Laguna M.1 is the foundation for the family, while XS.2 is the smaller model and is described as capable for its size. XS.2 is free for a limited time, and its weights were released under an Apache 2.0 license.

  • Laguna M.1 is the foundation model for the Laguna family.
  • Laguna XS.2 is described as small but capable for long-horizon coding work.
  • The weights for XS.2 were released under Apache 2.0 and the model is free to use for a limited time.

Meta Releases Sapiens2 Vision Models

Meta released Sapiens2, a family of high-resolution transformers trained on 1 billion human images. The models are designed to perform strongly on pose estimation, segmentation, and surface prediction tasks. It is another push into human-centric computer vision.

  • Sapiens2 was trained on 1 billion human images to improve human-centric vision tasks.
  • The models perform strongly on pose estimation, segmentation, and surface prediction.
  • Meta released it as a family of high-resolution transformers for vision research and downstream applications.

Trending AI Tools

  • Mistral Workflows A Python-based orchestration layer for durable, resumable AI automation.

  • Claude Connectors Connects Claude with Adobe, Blender, and Autodesk for creative workflows.

  • Redesign by Nodewave A free, open-source design tool built around describing what you want.

Quick Hits

Keep reading for free

Enter your email. If you're already subscribed, we'll send a sign-in code. If not, you'll subscribe in the next step.

Free access. Subscribe once, then use the same email on future issues.

Free to read. Subscription just unlocks the full issue.