The AI startup Dreamer is joining Meta Superintelligence Labs with its entire team, bringing co-founder Hugo Barra—a former Meta VP—back into Mark Zuckerberg’s orbit. The deal marks Meta’s second move in agent-based AI this year as the company tries to regain ground against competitors.
OpenAI is offering private equity firms a guaranteed minimum return of 17.5 percent to win them over for joint ventures in the enterprise market. Moreover, participating firms would get early access to new OpenAI models. Reuters broke the story, citing people familiar with the matter. The investment amounts involved are reportedly larger than usual.
The goal is to get private equity firms—investment companies that buy and resell entire businesses—to rapidly roll out OpenAI's AI tools across hundreds of companies in their portfolios. Big names like TPG, Advent, Blackstone, and Permira are reportedly in the mix.
Anthropic is pursuing a similar distribution strategy, but allegedly without offering a comparable return guarantee. That could change now that OpenAI has raised the stakes. The already thin margins of AI companies compared to SaaS peers are likely to take an even bigger hit from these kinds of commitments.
The entire effort appears aimed squarely at Anthropic, which has been gaining ground with enterprise customers recently and currently leads in coding with Claude Code. OpenAI recently announced a renewed focus on the coding business with Codex and a consolidation of its products into a single super app.
Mark Zuckerberg is building a personal AI agent to help him run Meta. The tool is still in development, but, according to the Wall Street Journal, already helps him pull up information faster, bypassing the multiple layers of employees he'd normally have to go through.
The project is reportedly part of a broader reorganization at Meta. The company, which currently has around 78,000 employees, wants to flatten its hierarchies, build leaner teams, and keep pace with AI-native startups. Zuckerberg's long-term vision: everyone inside and outside Meta gets their own AI agent, and the company operates as efficiently as an AI startup, the WSJ reports.
That connects to a bigger picture: According to Reuters, Meta is planning to cut up to 20 percent of its workforce. The layoffs are reportedly tied not to efficiency gains already realized but directly to the company's massive investments in AI infrastructure. A Meta spokesperson called the report speculation.
Karpathy spent months hand-tuning his GPT-2 training setup. Then he let an autonomous agent take over for a single night. The agent discovered fine-grained adjustments Karpathy had overlooked, tweaks that also interact with each other in ways that are easy for a human to miss but straightforward for a systematic search to catch.
Karpathy's takeaway is that researchers should remove themselves from the loop, at least in areas where objective metrics exist. "To get the most out of the tools that have become available now, you have to remove yourself as the bottleneck. You can't be there to prompt the next thing," Karpathy says. Researchers at major AI labs, he argues, place too much unfounded trust in their own intuition and are ultimately in the process of systematically automating themselves out of a job. Which, Karpathy notes, is also their stated goal.
While models keep getting better at coding and other easy-to-verify tasks, Karpathy doesn't think these gains will carry over smoothly to less measurable domains. "Anything that feels softer is, like, worse," he says.
In a new guide, OpenAI explains how front-end designers can get better results from GPT-5.4 when building websites and apps and how to stop the model from falling back on generic designs.
A German research team lets Transformer models decide for themselves how many times they think about a problem. Combined with additional memory, the approach outperforms larger models on math problems.
