OpenAI’s ex-CTO, Mira Murati, has recruited OpenAI co-founder John Schulman

OpenAI cofounder John Schulman is joining former OpenAI CTO Mira Murati’s stealth startup after departing Anthropic just five months after joining, according to sources familiar with the matter. Schulman’s reasons for leaving Anthropic so soon remain unclear, as does his role at Murati’s new company, which has not yet announced a name or specific product focus. However, the startup has already gained attention for recruiting talent from OpenAI, Character AI, and Google DeepMind. While Schulman’s exit from Anthropic was previously reported, his move to Murati’s venture had not been disclosed until now.

Schulman left OpenAI in August 2024, explaining on X that he wanted to focus more on AI alignment and return to hands-on technical work. He initially saw Anthropic as a promising environment for this goal but has now opted to join Murati’s initiative instead. Murati, who resigned from OpenAI in September 2024, had stated that she wanted the freedom to explore new opportunities. By October, she was reportedly in discussions to raise over $100 million in funding for her startup, signaling significant ambition for her next endeavor.

In addition to Schulman, Murati has brought in several high-profile hires, including Jonathan Lachman, former head of special projects at OpenAI. One source also revealed that she has recruited Christian Gibson from OpenAI’s supercomputing team and Mario Saltarelli, a longtime IT manager at OpenAI. These moves suggest a major organizational shift and indicate that Murati’s startup is rapidly assembling a team of top AI talent, setting the stage for a potentially influential new player in the field.

Get More From Every Word

Meet VoiceHub from Rev - the #1 platform for recording, transcribing, and analyzing speech. It’s revolutionizing how businesses handle their most valuable asset: conversations. 

What sets VoiceHub apart? Their AI accuracy beats Microsoft, Google, and other leading providers in enterprise environments.

But it's not just about accuracy—it's about changing the way you use those accurate transcripts. With VoiceHub, you get:

• Universal capture of audio and video across mobile, desktop, and meetings

• Best-in-class AI transcription and summaries in seconds

• Enterprise-grade security with HIPAA, SOC 2 Type II compliance, and SSO

• Seamless integration with 18 major tools like Zoom and Slack

Elon Musk’s DOGE is feeding sensitive federal data into AI to target cuts

Elon Musk’s Department of Government Efficiency (DOGE) has reportedly fed sensitive financial and personal data from the Education Department into AI software to analyze spending and identify budget cuts. According to sources, the team is using Microsoft Azure-based AI tools to scrutinize expenditures, including grants, contracts, and work trip expenses. The initiative aligns with the Trump administration’s goal of drastically reducing or eliminating the department. Lower-level staff were instructed to grant access to DOGE, raising concerns about privacy, security, and the use of AI in government operations, as many agencies have banned AI for handling sensitive information.

Beyond the Education Department, DOGE intends to expand AI-driven cost-cutting measures across multiple agencies, including the Department of Health and Human Services, the Labor Department, and the CDC. The Technology Transformation Services (TTS), which was originally created to modernize government platforms, has been repurposed as a hub for AI-driven contract analysis under former Tesla employee Thomas Shedd. The team has already flagged dozens of contracts for potential cuts, focusing on programs deemed non-essential or not legally required. Former White House AI policy expert Alondra Nelson voiced concerns about AI’s role in governance, warning of public skepticism and the potential for misuse.

One of the most immediate impacts of DOGE’s efforts was placing roughly 100 Education Department employees on paid administrative leave, citing a Trump executive order banning Diversity, Equity, and Inclusion (DEI) training. Many of those affected had attended a single training session, and most were women or people of color, according to union president Sheria Smith. This rapid implementation has alarmed agency staff, who describe DOGE’s approach as swift and aggressive, with access to critical government data spreading quickly. One employee summed up the concern: “Once they’re in, it’s already over.”

OpenAI now reveals more of its o3-mini model’s thought process

OpenAI has announced an update to its AI model, o3-mini, that enhances transparency in its reasoning process. Free and paid users of ChatGPT will now see a more detailed “chain of thought” explaining how the model arrives at its answers. This update is particularly beneficial for premium users utilizing the “high reasoning” configuration, as it provides greater clarity into the model’s decision-making process. According to OpenAI, the change aims to help users better understand the model’s thinking while maintaining response accuracy and confidence.

This update comes in response to competitive pressure from AI companies like DeepSeek, whose R1 model openly displays its full reasoning process. OpenAI had previously withheld full reasoning steps for its models, including o1 and o1-mini, opting instead to show condensed summaries, which were sometimes inaccurate. While OpenAI is still not revealing the complete reasoning process for o3-mini, it has introduced a balance where the model first generates its raw thoughts and then organizes them into more detailed summaries. This approach improves the user experience by making responses more structured and informative.

Additionally, OpenAI has implemented a post-processing step to refine the chain of thought, ensuring that unsafe content is removed and complex ideas are simplified. This step also enhances accessibility by translating reasoning steps into users' native languages. OpenAI’s Chief Product Officer, Kevin Weil, recently hinted at this update during a Reddit AMA, emphasizing the company’s ongoing efforts to balance transparency with competitive considerations. While OpenAI acknowledges the value of showing complete reasoning steps, it is taking a cautious approach to maintain both usability and proprietary advantages.

Related: You can now use ChatGPT web search without logging in

Other stuff

• Understanding Reasoning LLMs

• California’s AI Act Vetoed

• Ex-Palantir engineers think they can beat Salesforce, Microsoft, and other tech giants at the AI agent game

• How to Train an AI Image Model on Yourself

• Learn AI by Hand

• AI tools are starting to allow nonexperts to search for patterns across the globe

• A Gentle Intro to Running a Local LLM

• OpenAI Leaks Sales Agent in Tokyo

• Daisy the 'AI granny' driving scammer crazy

• Andrej Karpathy: Deep Dive into LLMs like ChatGPT

Hope you enjoyed today's newsletter

⚡️ Join over 200,000 people using the Superpower ChatGPT extension on Chrome and Firefox.