Pointers-
-
OpenAI announces a one-week break to help staff recover from burnout.
-
Meta tries to poach top OpenAI researchers, offering massive compensation (up to $100M).
-
Seven key researchers leave OpenAI to join Meta’s new AI superintelligence team.
-
OpenAI leaders urge focus on core mission — building safe Artificial General Intelligence (AGI).
-
Meta expands its AI portfolio with two major models:
-
Llama 3.2 for language and vision tasks
-
V-JEPA 2 for understanding the physical world (ideal for robots, self-driving tech)
-
-
The rivalry now goes beyond chatbots — it’s a battle for talent, purpose, and AI leadership.
OpenAI has announced a one-week company-wide break to help its employees rest and recover. Many team members have reportedly been working 80 hours a week, and the company wants to avoid burnout. But the break comes at a tricky time — Meta is trying hard to poach OpenAI’s best researchers, offering some of them huge salary packages of up to $100 million.
According to a report by Wired, OpenAI’s Chief Research Officer Mark Chen warned employees in a Slack message:
“Meta knows we’re taking this week to recharge and will take advantage of it to try and pressure you to make decisions fast and in isolation.”
Meta Hires Away Top OpenAI Researchers
In recent weeks, seven researchers have left OpenAI to join Meta, including Lucas Beyer, Alexander Kolesnikov, Xiaohua Zhai, and Trapit Bansal. All of whom were part of key AI model development at OpenAI’s Zurich office.
Meta is building a new AI Superintelligence team, and these hires are part of its efforts to become a stronger player in the AI space.
In response, OpenAI leaders like Chen and CEO Sam Altman are promising to review salaries, improve rewards, and focus on keeping the team united around their mission — building safe, advanced AI.
Chen admitted the recent exits were painful:
“I feel a visceral feeling right now, as if someone has broken into our home and stolen something.”
“Please trust that we haven’t been sitting idly by,” he added. He said he and Altman are available to talk “around the clock” with anyone thinking about leaving.
Still, he made one thing clear:
“While I’ll fight to keep every one of you, I won’t do so at the price of fairness to others.”
Staying Focused on the Bigger Goal
Mark Chen reminded the team that OpenAI’s main goal is to build Artificial General Intelligence (AGI), not to get caught up in competition:
“This is the main quest, and it’s important to remember that skirmishes with Meta are the side quest.”
Sam Altman supported Chen’s message, replying internally:
“It’s been really amazing to watch Mark’s leadership and integrity through this process, especially when he has had to make tough decisions. Very grateful we have him as our leader!”
Meta’s Bigger AI Game: From Language to Real-World Understanding
Meta is not just working on text and chat-based AI; it’s building powerful models to handle both language and the physical world.
Its latest open-source model suite, Llama 3.2, includes smaller versions like 1B and 3B for phones or on-device use, and larger ones like 11B and 90B for complex tasks like image captioning and document reasoning. These models are highly customizable and ideal for real-time, privacy-focused applications. Llama 3.2 is also performing well in vision-language benchmarks like VQAv2 and ChartQA.
But Meta is also going a step further with V-JEPA 2 — a new type of AI model that goes beyond language and learns how objects move, interact, and behave in the real world.
Unveiled at VivaTech in Paris, V-JEPA 2 is designed to help machines like self-driving cars and delivery robots understand their surroundings and predict what will happen next.
“A world model is like an abstract digital twin of reality that an AI can reference to understand the world and predict consequences of its actions,” said Yann LeCun, Meta’s Chief AI Scientist.
Unlike language models that rely on lots of text or video, V-JEPA 2 uses a simplified “latent space” to reason about cause and effect, like understanding that a ball will fall if it rolls off a table.
With both Llama 3.2 and V-JEPA 2, Meta is positioning itself as a major player in the race to build more human-like AI systems that can not only think and talk, but also see, plan, and act.
Denials and Online Reactions
Lucas Beyer, one of the researchers who joined Meta, denied the rumor that he got a $100 million signing bonus. On X (formerly Twitter), he wrote:
“Yes, we will be joining Meta. No, we did not get $100M sign-on, that’s fake news.”
Meanwhile, OpenAI engineer Cheng Lu shared his feelings online (before deleting the post):
“Such a huge loss to OpenAI and I feel really disappointed that the leadership didn’t keep them.”
As OpenAI takes a break this week, it’s clear the company is not just fighting burnout — it’s also facing a serious test in keeping its top talent together, while staying focused on its bigger dream: building powerful, safe AI for everyone.
