Google DeepMind Launches SIMA 2 Alongside Meta AI's DreamGym System

The artificial intelligence sector is currently experiencing a significant investment surge, with major tech companies like Google, Meta, and Microsoft collectively planning to invest $400 billion this year. This boom, however, is prompting serious warnings from financial experts. Michael Burry, the investor known from "The Big Short," cautioned on November 17, 2025, that the "Cloud then AI Buildout" has created a capital expenditure bubble even larger than the Dot-Com and Housing bubbles. WIRED's Michael Calore, Lauren Goode, and writer Brian Merchant also discussed signs of an AI tech bubble, noting that many businesses using AI report little to no return on investment, a sentiment echoed by figures like Sam Altman and Bill Gates. Burry, who recently closed his hedge fund, Scion Asset Management, revealed he was betting against companies such as Nvidia Corp. and Palantir Technologies Inc. Amidst these financial discussions, major players continue to push AI development and application. OpenAI is focusing on making its products more useful and profitable, with Fidji Simo, who joined in August, leading the Applications division. Her efforts include launching tools like Pulse for personalized information and a jobs platform for AI-certified roles, aiming to bridge the gap between AI capabilities and public adoption. Google DeepMind introduced SIMA 2, an advanced AI agent powered by Gemini, designed to navigate complex 3D virtual worlds. This agent boasts a 62% success rate on tasks, close to human performance, and can adapt to new games while understanding various instructions, from spoken words to emojis. Its self-improvement loop, where a Gemini model generates and scores new tasks, allows for continuous learning. Meta AI also unveiled DreamGym, a system that trains Reinforcement Learning (RL) agents more efficiently by simulating complex environments like WebShop and ALFWorld entirely through text. This approach helps RL agents learn and improve skills without the high costs and difficulties of real-world training, addressing challenges in scaling RL for large language model agents. The underlying infrastructure for AI is also evolving rapidly. Arm Holdings Plc announced on November 17, 2025, its plan to integrate Nvidia's NVLink technology into its chip designs for AI data centers. This move strengthens the partnership between the two semiconductor giants, with Arm CEO Rene Haas aiming to expand the company's offerings beyond smartphones. Nvidia has also made an open version of NVLink, called Fusion, available to other hardware makers. AI's impact extends to business models, as seen with consulting firm McKinsey. Approximately a quarter of McKinsey's global fees now come from outcomes-based pricing, a shift driven by complex, multi-year AI transformation projects. Michael Birshan and Kate Smaje from McKinsey explain that the firm now provides deep implementation expertise, acting as a genuine partner in clients' business transformations. However, the rapid integration of AI also brings new concerns. Security expert Marina Moore warned on November 17, 2025, about cloud security risks, particularly with containers due to insufficient isolation and shared host OS kernels, which can lead to Linux kernel attacks. She also highlighted that AI inference creates new attack vectors because GPUs often do not clear memory between processes. Separately, a concerning trend shows medical disclaimers quietly disappearing from AI systems. Companies are removing warnings like "We do not give medical advice" to boost user engagement, leading to AI becoming a default medical advisor, especially in areas with limited access to doctors. This raises serious questions about accountability and the spread of misinformation. Even popular culture is grappling with AI's implications, with Gordon Smith, executive producer and writer for Apple TV+'s "Pluribus," discussing how viewers perceive the show as being against AI, touching on themes of political division and addiction.

Key Takeaways

• Top tech companies like Google, Meta, and Microsoft plan to invest $400 billion in AI this year, despite warnings of a potential AI tech bubble.
• Investor Michael Burry warned on November 17, 2025, that the "Cloud then AI Buildout" has created a capital expenditure bubble larger than the Dot-Com and Housing bubbles, betting against Nvidia Corp. and Palantir Technologies Inc.
• OpenAI's Applications division, led by Fidji Simo, aims to increase ChatGPT's utility and profitability, launching tools like Pulse and an AI-certified jobs platform.
• Google DeepMind released SIMA 2, an advanced AI agent powered by Gemini, achieving a 62% task success rate in 3D virtual worlds and learning through self-play.
• Meta AI introduced DreamGym, a system that efficiently trains Reinforcement Learning (RL) agents by simulating complex environments entirely through text, reducing real-world training costs.
• Arm Holdings Plc announced on November 17, 2025, it will integrate Nvidia's NVLink technology into its chip designs for AI data centers, strengthening their partnership.
• McKinsey's business model is shifting, with about 25% of its global fees now coming from outcomes-based pricing for complex, multi-year AI transformation projects.
• Medical disclaimers are disappearing from AI systems to boost user engagement, raising concerns about accountability, misinformation, and AI becoming a default medical advisor.
• Cloud security expert Marina Moore highlighted new AI-era risks on November 17, 2025, particularly from containers due to insufficient isolation and AI inference creating new attack vectors on GPUs.
• Apple TV+'s show "Pluribus" explores themes related to AI, political division, and addiction, with viewers often interpreting it as being against AI.

OpenAI's Fidji Simo aims to make ChatGPT more useful and profitable

Fidji Simo, former Instacart CEO, now leads OpenAI's Applications division, focusing on making ChatGPT and other products more useful and profitable. She joined in August and aims to turn OpenAI's research breakthroughs into essential consumer products. Simo has already overseen the launch of Pulse, a tool that provides personalized information, and a jobs platform for AI-certified roles. Her main goal is to increase how much people use OpenAI's intelligent models, closing the gap between their capabilities and public adoption.

New AI era cloud security risks from containers and inference

Security expert Marina Moore warns about new cloud security risks in the age of AI, especially concerning containers. She explains that containers pose a major threat due to insufficient isolation and sharing the host OS kernel, leading to Linux kernel attacks. To improve security, companies should minimize their code base and consider Micro VMs for high-isolation needs. Moore also highlights that AI inference creates new attack vectors because GPUs often do not clear memory between different processes, complicating isolation efforts. This information was shared in a podcast on November 17, 2025.

Experts discuss four signs of an AI tech bubble

WIRED's Michael Calore and Lauren Goode discussed the possibility of an AI tech bubble with writer Brian Merchant. Top tech companies like Google, Meta, and Microsoft are greatly increasing their AI investments, with $400 billion planned for this year. However, many businesses using AI report little to no return on investment. Experts like Sam Altman and Bill Gates have also spoken about a potential AI bubble. Brian Merchant used a historical framework to identify signs of an economic bubble, comparing the current situation to past events like the Dot-Com and Housing bubbles.

Pluribus creator discusses AI themes and hidden meanings

Gordon Smith, an executive producer and writer for Apple TV+'s show "Pluribus", discussed the show's themes in an interview on November 17, 2025. He noted that many viewers see the show as being against AI, especially with how the group called "the Others" operates. Smith, who wrote and directed several episodes, prefers not to explicitly state the show's meaning, believing it makes the storytelling richer and encourages viewers to think. He also mentioned that "Pluribus" touches on political division and addiction, seen through the character Carol's struggles.

Meta AI unveils DreamGym for training RL agents with text

Meta AI has introduced DreamGym, a new system designed to train Reinforcement Learning (RL) agents more efficiently. DreamGym simulates complex environments like WebShop and ALFWorld entirely through text, avoiding the high costs and difficulties of real-world training. This system uses a reasoning-based experience model, a replay buffer for memory, and a task generator to create varied learning challenges. By simulating interactions in text, DreamGym helps RL agents learn and improve their skills without needing thousands of real-world actions. This innovation aims to overcome common problems in scaling RL for large language model agents.

Google DeepMind launches SIMA 2 a smart agent for virtual worlds

Google DeepMind has released SIMA 2, an advanced AI agent powered by Gemini, designed to operate in complex 3D virtual worlds. This new version improves upon SIMA 1 by reasoning about goals, explaining its plans, and learning through self-play. SIMA 2 achieves a 62% success rate on tasks, close to human performance, and shows strong ability to adapt to new games it has not seen before. It also understands various types of instructions, including spoken words, drawings, and emojis. A key feature is its self-improvement loop, where a Gemini model creates new tasks and scores the agent's attempts, allowing it to learn continuously.

AI transforms McKinsey's business model and client fees

AI is significantly changing how consulting firm McKinsey operates and charges its clients. About a quarter of McKinsey's global fees now come from outcomes-based pricing, where clients pay based on the results achieved rather than just the time spent. This shift is due to the complex, multi-year AI transformation projects clients are undertaking. Michael Birshan and Kate Smaje from McKinsey explained that the firm now provides deep implementation expertise instead of just strategy advice. This new approach makes McKinsey a genuine partner in clients' business transformations.

AI medical disclaimers vanish raising health concerns

Medical disclaimers, like "We do not give medical advice," are quietly disappearing from AI systems. This change is not accidental; AI companies are removing them to increase user engagement and adoption. Users tend to trust AI answers more without these warnings and even downvote responses that include them. As a result, AI is becoming a default medical advisor, especially in regions with limited access to doctors. This trend raises serious concerns about accountability, the spread of medical misinformation, and potential health inequalities.

Michael Burry warns AI investment boom mirrors past bubbles

Famous "Big Short" investor Michael Burry issued a strong warning on November 17, 2025, about the current AI investment boom. He believes the "Cloud then AI Buildout" has created a capital expenditure bubble that is even larger than the DotCom and Housing bubble peaks. Burry shared a chart showing this trend and used a meme to suggest investors are ignoring the clear risks. This warning comes shortly after he closed his hedge fund, Scion Asset Management, due to a disagreement with market behavior. His final investment filings showed he was betting against companies like Nvidia Corp. and Palantir Technologies Inc.

Arm to use Nvidia NVLink in AI data center chips

Arm Holdings Plc announced on November 17, 2025, that it will add Nvidia's NVLink technology to its chip designs for AI data centers. This move strengthens the relationship between the two major semiconductor companies. NVLink is a key technology from Nvidia that connects many AI chips together. Arm CEO Rene Haas is expanding the company's offerings beyond smartphones to boost revenue by providing more complete chip designs to customers. Nvidia also recently made an open version of NVLink called Fusion available for other hardware makers.

Sources

• OpenAI's Fidji Simo Plans to Make ChatGPT Way More Useful—and Have You Pay For It
• Cloud Security Challenges in the AI Era - How Running Containers and Inference Weaken Your System
• The 4 Things You Need for a Tech Bubble
• ‘Pluribus’ Boss Gordon Smith Addresses Anti-AI Subtext, Says It’s “Less Rich” to Spell Things Out
• Meta AI Introduces DreamGym: A Textual Experience Synthesizer For Reinforcement learning RL Agents
• Google DeepMind Introduces SIMA 2, A Gemini Powered Generalist Agent For Complex 3D Virtual Worlds
• AI is reshaping how McKinsey makes money
• The Silent Disappearance of Medical Disclaimers
• Michael Burry's Ominous Warning Hints That AI Investment Boom Is Rivaling Dot-Com, Housing Bubble Peaks
• Arm to Offer Nvidia’s NVLink Technology in AI Data Center Chips