February Tech Digest: Ex-OpenAI Leaders’ Moves, DeepSeek’s Woes amp; AI Advances

February has been another eventful month for the AI world, with significant developments ranging from new ventures by former OpenAI executives to the latest breakthroughs in research and technology. As companies like Apple and Meta push forward with AI innovations, others face mounting challenges, such as DeepSeek’s growing security concerns. Let’s dive into the latest updates from the world of AI in February 2025.

Life after OpenAI: what are ex-employees up to in 2025?

As you know, in 2024 and earlier, many key figures left OpenAI for various reasons — some disagreed with the company's direction, while others wanted to pursue their own ambitions. Let’s take a look at what they’re up to now, in 2025.

Mira Murati, John Shulman, and Barret Zoph lead Thinking Machines Lab

So, former OpenAI CTO Mira Murati has launched a new AI startup called Thinking Machines Lab. The company aims to develop customizable, widely understood, and capable AI tools that address key gaps in current AI systems. Thinking Machines Lab focuses on multimodal AI systems that can collaborate with people across various domains, such as science and programming. AI safety is also a priority, and plans are in place to prevent misuse, share best practices, and support external AI alignment research.?

Here is the most interesting part: Mira Murati recruited John Schulman, OpenAI co-founder, who left OpenAI for Anthropic and then left Anthropic in February 2025 and has already joined Mira Murati’s new startup as a chief scientist. Next, she hired Barret Zoph as CTO, bringing in OpenAI’s former chief research officer to help lead the development of advanced AI systems.

Ilya Sutskever’s Safe Superintelligence

Another former OpenAI executive, Ilya Sutskever, has launched his own AI startup, Safe Superintelligence. He is currently in talks to raise funding at a valuation of at least $20 billion, a significant jump from the company’s $5 billion valuation last September.

Safe Superintelligence remains largely secretive about its work, but it has already secured $1 billion in funding from investors like Sequoia Capital, Andreessen Horowitz, and DST Global. Sutskever, known for his key role in AI breakthroughs at OpenAI, founded the company alongside ex-OpenAI researcher Daniel Levy and former Apple AI lead Daniel Gross.Dario and Daniela Amodei’s Anthropic

Dario and Daniela Amodei are now Anthropic’s leaders and former OpenAI researchers. After working at OpenAI on various AI safety initiatives, they left to address concerns about AI alignment and safety. They founded Anthropic, a company dedicated to building AI systems that are safe, interpretable, and aligned with human intentions. Anthropic has developed the Claude family of large language models and secured major investments from companies like Amazon and Google.

Apple’s February digest

Apple Intelligence is coming to Apple Vision Pro

Apple has confirmed that its generative AI platform, Apple Intelligence, is coming to the Vision Pro headset as part of the upcoming visionOS 2.4 update, set for public release in April. This update introduces AI-powered text and image generation tools, including Rewrite, Proofread, and Summarize, enhancing the headset’s productivity potential. While Vision Pro has been marketed as a "spatial computing" device, text input has remained a challenge—Apple now aims to improve this with AI-driven voice dictation and smarter interactions.

Additionally, the update includes Image Playground, an integrated image generation feature in the Photos app, and a new Vision Pro iPhone app arriving with iOS 18.4. The app allows users to browse visionOS content on their iPhone before transferring it to the headset and manages guest accounts. While no Vision Pro-specific AI features are debuting, Apple is betting on AI to make the headset a more seamless part of users' workflows.

Apple explores robots for the future smart home

According to analyst Ming-Chi Kuo, Apple is diving into robotics, researching both humanoid and non-humanoid designs. The company recently published a paper on human interactions with “non-anthropomorphic” robots, such as a Pixar-style lamp. While still in the early stages, Apple’s work could lead to consumer robots by 2028 — though Kuo warns the project could end up like the scrapped Apple Car.

Rather than focusing on humanoid looks, Apple is prioritizing how users interact with robots, hinting at innovations in sensing hardware and software. While a full-fledged humanoid helper is unlikely to be available anytime soon, Apple may introduce robotic smart home assistants. However, given the high costs and challenges in the robotics industry, Apple appears to be cautiously testing the waters before making any major moves.

How is doing DeepSeek after such a big hype?

Well, not really good. DeepSeek faces global bans over security concerns. DeepSeek is under fire as multiple governments and corporations ban its technology due to privacy and security fears. Countries like Italy and Taiwan have blocked DeepSeek’s AI tools, citing data protection violations and national security risks. The U.S. Congress, Pentagon, and NASA have also restricted its use, while Texas and the U.S. Navy issued outright bans, fearing potential data leaks to the Chinese government.

As regulatory scrutiny intensifies, DeepSeek’s expansion faces serious hurdles. Despite the bans, some U.S. agencies still allow limited access through approved platforms, keeping the debate over AI security and geopolitics alive.

However, Databricks CEO Ali Ghodsi views DeepSeek’s rise as a transformative moment for AI, signaling a shift from brute-force scaling to smarter, domain-specific models. He noted that DeepSeek’s rapid model releases prove that efficiency and specialized reasoning can rival even the largest AI systems. This disrupts the traditional AI hierarchy, where bigger models dominate, and challenges industry leaders like OpenAI. Ghodsi also emphasized that as AI becomes more cost-effective, the demand for hardware will shift from training to inference, potentially opening the door for new competitors in the AI chip market.

AI latest discoveries

Meta unveils LlamaCon: its first generative AI dev conference

Meta has announced its first-ever generative AI developer conference, LlamaCon, which will take place on April 29, 2025. The conference will focus on Meta's open-source AI developments, specifically its Llama family of models, and will offer tools to help developers create innovative apps and products. The event comes as Meta faces increasing competition from DeepSeek, a Chinese AI company that may have created more efficient models, leading Meta to invest heavily in AI, including hiring and new data centers. Meta is also gearing up to release new Llama models, including reasoning and multimodal capabilities while navigating legal and privacy challenges.

Hugging Face challenges OpenAI with open deep research

Researchers from Hugging Face, led by Clement Delangue, have developed Open Deep Research, an open alternative to OpenAI’s deep research tool. Using OpenAI’s o1 model and an open-source agentic framework, their system autonomously navigates the web, compiles research, and processes data. While it doesn’t yet match OpenAI’s proprietary o3 model in performance, the team has made the source code publicly available on GitHub, aiming for continuous improvement and wider accessibility. This move underscores the growing demand for open and transparent AI research tools, challenging the dominance of proprietary models and fostering innovation through community collaboration.

Google unveils new Gemini AI models to compete with DeepSeek

Google has launched its new flagship AI model, Gemini 2.0 Pro Experimental, along with Gemini 2.0 Flash Thinking, aimed at enhancing reasoning capabilities. The move comes as competition with the Chinese AI startup DeepSeek intensifies, particularly in the realm of cost-effective AI reasoning models. Google’s Gemini 2.0 Pro boasts improved coding skills, better reasoning, and a 2 million token context window, allowing it to process vast amounts of text in a single prompt. Additionally, Gemini 2.0 Flash-Lite, a cost-efficient model, has been introduced to rival DeepSeek’s affordable AI offerings.

Perplexity introduces freemium Deep Research tool

Perplexity has launched its own "Deep Research" feature, joining Google and OpenAI in offering AI-powered research tools that provide detailed, cited reports for expert-level tasks. Unlike OpenAI’s $200/month Pro-only access, Perplexity’s version is available for free with limited queries, while subscribers get unlimited access. The tool performs faster than OpenAI’s alternative and excels in accessibility, though OpenAI still leads in analytical depth. With AI-driven research becoming more common, questions remain about its impact on critical thinking and creativity in professional fields.

Elon Musk’s Grok 3 surges in popularity but faces controversy

In February 2025, Elon Musk’s xAI launched Grok 3, its latest AI model designed to rival OpenAI’s GPT-4o and Google’s Gemini. Trained using a massive data center with 200,000 GPUs, Grok 3 boasts enhanced reasoning abilities and outperforms competitors in math, science, and programming benchmarks. It also introduces DeepSearch, an AI-powered research tool, and upcoming voice features. Musk promises to open-source Grok 2 once Grok 3 stabilizes, aiming to make xAI’s AI more politically neutral.

Later, Grok 3 sparked a massive surge in users, with mobile downloads increasing 10x and global daily visits skyrocketing from 627K to 4.5M. This growth was partly driven by the app’s expansion into Europe, Latin America, and Southeast Asia. However, the launch was marred by controversy when Grok censored Musk’s criticism, raising concerns about bias and content moderation. While initial adoption is strong, questions remain about whether Grok can maintain user engagement and overcome these challenges.