Trump appointed Jay Clayton to lead a White House “superintelligence” task force, giving it 120 days to study AI’s risks and opportunities and make recommendations. OpenAI will test ads in ChatGPT image generation in the U.S. later this month; a16z data shows that, as of April 2026, about 2% of U.S. households paid for AI services. Aleph Alpha has released the weights for Kolibri, a model with 78B total parameters, 3.46B active parameters and support for a context window of up to 1M tokens. Sponsored content is coming to image generation, though OpenAI says it will be kept separate from generated content.
The task force has 120 days to study risks and opportunities and make recommendations.
Trump has formed a White House “superintelligence” task force led by Jay Clayton, directing it to study AI’s risks and opportunities and make recommendations within 120 days. The group will examine how to respond to AI incidents, cyberattacks and failures of advanced models. Trump had previously instructed the U.S. government to refer to AI as “superintelligence.” Clayton says the risk of the U.S. failing to lead in AI development is high, and he opposes pausing research and development.
Read original →OpenAI is testing sponsored content while the share of U.S. households paying for AI services remains low.
OpenAI will test ads in ChatGPT image generation in the U.S. later this month, showing sponsored products and experiences when users generate images. Image-generation scenarios such as home decorating, outfit planning and desk design offer places to display relevant products. Ads will be labeled as sponsored and kept separate from generated content; OpenAI says they will not affect ChatGPT’s answers.
Read original →a16z’s “State of Markets II” says that, as of April 2026, about 2% of U.S. households paid for some form of AI service. The report says that although AI applications have spread, actual use remains relatively shallow: nearly 30% of S&P 500 companies report a “quantifiable impact” from AI, but only about 2% report tracking metrics. a16z notes that the share of households paying has since risen, while demand for compute has continued to exceed supply.
Read original →🔍 Analysis: a16z’s estimate that about 2% of households pay for AI services, alongside OpenAI’s test of sponsored content in image generation, points to a revenue path beyond subscriptions. For AI app developers, monetization is not just about how many users will pay. It also depends on whether sponsored content can be clearly labeled and kept separate from generated content.
Kolibri’s weights are available now; Reflection’s first model is still in preparation.
Nvidia-backed Reflection is preparing to release its first open-weight model, aiming to compete with Chinese models such as Kimi and Qwen. Reflection wants enterprises to be able to download, customize and run the model themselves rather than rely entirely on OpenAI or Anthropic APIs. It also plans to combine enterprise private data, its own model and Nvidia GPUs to help companies build proprietary AI systems.
Read original →Aleph Alpha has introduced Kolibri, a model built in Europe with 78B total parameters, 3.46B active parameters and support for a context window of up to 1M tokens. Alongside the model specifications, Aleph Alpha has released the weights so users can run it on their own hardware. The weights are licensed under Apache 2.0.
Read original →🔍 Analysis: Kolibri’s weights are available under the Apache 2.0 license, making self-hosting an option now. Reflection, meanwhile, plans to combine an open-weight model with enterprise private data and Nvidia GPUs to build proprietary systems. Enterprises evaluating these options need to distinguish between getting access to model weights and building a system with private data: the two are related, but they are not the same deliverable.
The daily release plan for the next 28 days sets a clear improvement for most users as its standard.
OpenAI’s Thibault Sottiaux says that on each of the next 28 days, the team will release something that noticeably improves the experience for most Codex or Work users, or carry out a full reset. This is a day-by-day release plan, not a single feature update. He set a noticeable improvement for most users as the standard for daily changes, while leaving a full reset as another possibility.
Read original →A Granola co-founder discusses enterprise users completing tasks without going through the interface.
Granola co-founder Sam says users are bypassing the product interface and completing tasks directly through MCP—a disappointing development for him as an interface designer. The team had once insisted that users do everything inside the Granola interface. But he believes bypassing the interface becomes hard to avoid when an enterprise grows large enough.
Read original →