Wednesday, May 6, 2026

Ads Inside ChatGPT Arrive

Ads Inside ChatGPT Arrive

Today’s Overview

Good morning, ChatGPT ads are suddenly on the table, while governments are getting a closer look at frontier models before release. On top of that, home-based mini data centers and new multimodal developer tools show how fast the infrastructure story is moving. Let's dive in.

Top Stories

Ads Inside ChatGPT

OpenAI is introducing ads inside ChatGPT, with campaign creation, management, and measurement built around the product. If it rolls out publicly as described, it would add a major new monetization surface to the ChatGPT experience.

  • OpenAI is building campaign creation around ChatGPT ads.
  • The product also includes management tools for those campaigns.
  • It adds measurement to round out the ad stack.

U.S. Government Will Inspect Frontier AI Before Release

Google DeepMind, Microsoft, and xAI have agreed to give the U.S. government early access to unreleased AI models for national-security testing. The work will run through the Commerce Department’s Center for AI Standards and Innovation, signaling a shift toward pre-release government evaluation of frontier systems.

  • Google DeepMind, Microsoft, and xAI agreed to give the government early access to unreleased models.
  • The testing runs through the Center for AI Standards and Innovation at the Commerce Department.
  • Officials say the focus includes cybersecurity and national security uses.

Mini AI Data Centers Move Into Homes

Span is teaming up with Nvidia to install mini AI data centers on residential homes and small businesses, using unused electrical capacity on local grids. The company says the approach can be faster and cheaper than building centralized data centers, and it is already testing the idea with PulteGroup in new communities.

  • Span’s XFRA boxes mount on exterior walls of homes and businesses.
  • Nvidia is supplying liquid-cooled RTX PRO 6000 GPUs to power each unit.
  • Span says the rollout could be 6x faster and one-fifth the cost of a comparable 100MW data center.

Research & Analysis

Google Reframes Hallucinations as Uncertainty

Google published a paper arguing that hallucinations are better understood as failures to express uncertainty than as simple knowledge gaps. The work proposes faithful uncertainty as a way to align model confidence with actual reliability.

  • The paper treats hallucinations as a failure to express uncertainty rather than missing knowledge.
  • It proposes faithful uncertainty as the fix.
  • The goal is to align confidence and reliability more closely.

AI2 Releases MolmoAct 2

MolmoAct 2 is an upgraded action reasoning model aimed at better real-world robot task performance. AI2 paired it with a large open bimanual manipulation dataset.

  • MolmoAct 2 is an upgraded action reasoning model for robotics tasks.
  • It is paired with a large open bimanual manipulation dataset.
  • The model is meant to improve real-world robot performance.

End-To-End Tokenizer Training For Images

A new end-to-end pipeline jointly optimizes image tokenization and generation, letting generation quality feed back into the training process. The work is presented as a technical result on autoregressive image modeling.

  • The pipeline jointly optimizes image tokenization and generation.
  • It adds direct feedback from generation quality into training.
  • The work targets autoregressive image modeling.

Trending AI Tools

  • Gemini API File Search Adds multimodal support, metadata filtering, and page-level citations for RAG.

  • Subquadratic Launches a 12-million-token context model and says a 50-million-token version is coming.

  • Unity AI Beta Brings project-aware agentic grounding to Unity workflows.

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.