Yann LeCun, formerly Meta's chief AI scientist, recently voiced strong criticisms regarding the company's AI direction and internal practices. He stated that Meta CEO Mark Zuckerberg was upset after the Llama 4 AI model's benchmark results were "fudged," leading to a loss of confidence in the GenAI organization. LeCun also called Alexandr Wang, who leads Meta's new Superintelligence Labs, inexperienced in research. He believes large language models like Llama are a dead end for achieving true superintelligence and has since launched his own venture, Advanced Machine Intelligence, predicting further departures from Meta's AI division. Meanwhile, a principal engineer at Google, Jaana Dogan, shared a remarkable experience with a competitor's product. Dogan, who works on the Gemini API, revealed that Anthropic's Claude Code developed a complex distributed agent orchestrator in just one hour. This task had previously taken her Google team an entire year to complete. Her praise for Claude Code highlights the rapid advancements in AI-driven coding and the potential for these tools to significantly accelerate software development, even for seasoned engineers at major tech companies. Beyond these major players, the AI ecosystem continues to evolve with new tools and talent shifts. TRON is expanding its "AI + Web3" vision with AINFT, an AI infrastructure, and SunAgent, an AI assistant designed to simplify blockchain tasks through natural language. In personnel news, AI scientist Ling Haibin, known for creating the world's first mobile plant identification app, has left his US position to join Westlake University in Hangzhou, China, seeking more research freedom. These developments underscore the global race for AI talent and innovation, prompting questions about who truly controls this influential technology. The rapid growth of AI also brings significant societal and environmental considerations. California's new law, effective January 1, will require tech companies to disclose how they manage risks from advanced AI systems, including independent reviews, though critics note it overlooks environmental impact and misinformation. Datacenters, critical for AI, already consume 1% of global electricity, with US consumption potentially reaching 8.6% by 2035, raising climate pollution concerns. Furthermore, AI news sources are subtly altering public opinion through "communication bias," highlighting the need for greater transparency and competition in the AI space.
Key Takeaways
- Meta's Llama 4 AI model's benchmark results were "fudged," leading to CEO Mark Zuckerberg's dissatisfaction and a loss of confidence in the GenAI organization.
- Yann LeCun, former chief AI scientist at Meta, criticized Alexandr Wang, head of Meta's Superintelligence Labs, for lacking research experience and views large language models like Llama as a dead end for superintelligence.
- Google Principal Engineer Jaana Dogan reported that Anthropic's Claude Code built a complex distributed agent orchestrator in one hour, a task that took her Google team a full year.
- TRON is expanding its AI ecosystem with AINFT, an AI infrastructure, and SunAgent, an AI assistant for blockchain tasks, aiming to lead in the "AI + Web3" space by 2025.
- AI's growing use is raising significant environmental concerns due to high energy and water consumption, with datacenters projected to use 8.6% of US electricity by 2035.
- A new California law, effective January 1, will require tech companies to provide transparency reports on how they manage risks from advanced AI models, including independent reviews.
- AI scientist Ling Haibin, known for creating the world's first mobile plant identification app, moved from the US to Westlake University in Hangzhou, China, seeking new research opportunities.
- AI news sources are subtly changing people's opinions and feelings through "communication bias," which adjusts information presentation based on perceived user preferences.
- The increasing influence of AI on global power and money highlights the critical question of who truly controls artificial intelligence.
Yann LeCun criticizes Meta AI leader Alexandr Wang
AI pioneer Yann LeCun believes Alexandr Wang, who leads Meta's new Superintelligence Labs, is inexperienced. LeCun, a former chief AI scientist at Meta, told the Financial Times that Wang lacks research experience. He also stated that Meta CEO Mark Zuckerberg was upset after Llama 4's results were "fudged," leading to a loss of confidence in the GenAI organization. LeCun thinks large language models like Llama are a dead end for true superintelligence. He has since started his own company, Advanced Machine Intelligence, and predicts more AI employees will leave Meta.
Yann LeCun says Meta fudged Llama 4 AI tests
Yann LeCun, former Chief Scientist at Meta, revealed that Meta "fudged" the benchmark results for its Llama 4 AI model. He told the Financial Times that Meta CEO Mark Zuckerberg was very upset about Llama 4's disappointing performance. This led Zuckerberg to sideline the entire GenAI organization, causing many employees to leave or plan to leave. LeCun explained that Meta likely fine-tuned different versions of the model for various benchmarks to make it seem more capable. However, users found the model underwhelming, which explains why Meta did not release a follow-up model.
AI boom raises big concerns about climate pollution
The growing use of AI is causing serious concerns about its environmental impact, especially due to high energy and water use. Sharon Wilson from Oilfield Witness observed Elon Musk's xAI Colossus datacentre in Memphis releasing large amounts of methane from its gas-fired turbines. Datacentres currently use 1% of the world's electricity, but this is expected to rise sharply, potentially reaching 8.6% of US electricity by 2035. In Ireland, datacentres already use one-fifth of the country's electricity, leading to a ban on new grid connections in 2021. Experts worry this energy demand, often met by fossil fuels, could make it harder to fight climate change.
California law makes AI companies share safety plans
A new California law, signed by Governor Gavin Newsom, will require tech companies to share how they manage big risks from advanced AI systems. Starting January 1, these companies must provide transparency reports detailing their AI models' uses, restrictions, and how they handle catastrophic risks, including independent reviews. Rishi Bommasani from Stanford University noted that this law brings much-needed openness to the AI industry. However, critics say the law does not cover important issues like AI's impact on the environment, spread of false information, or unfair biases. Also, incident reports sent to the Office of Emergency Services will not be made public, though they will go to lawmakers and the governor.
Understanding who controls artificial intelligence
Many people are asking who truly controls artificial intelligence, given its growing influence on global power and money. Some believe AI is mostly a technical and economic issue, driven by private investments, and will advance no matter what. However, the author points out that it is important to separate the AI technology itself, like large language models, from the political groups forming around it. Both the technology and its political impacts are very important and need careful discussion.
TRON boosts AI with AINFT and SunAgent tools
TRON is expanding its AI ecosystem with two main tools: AINFT and SunAgent. AINFT is a new AI infrastructure that supports AI Agents, helps train AI models, and manages AI assets. It was created from an upgrade of APENFT on October 9. SunAgent is an AI assistant that lets users complete blockchain tasks like sending tokens or voting by simply chatting in natural language. TRON aims to lead in the "AI + Web3" space by 2025, using these tools to make AI applications easier to develop and use. This move helps TRON integrate AI technology with its blockchain ecosystem, creating a new decentralized crypto AI system.
Google engineer says Claude Code built year's work in an hour
Jaana Dogan, a Principal Engineer at Google, shared how an AI tool called Claude Code quickly created complex software. She stated that Claude Code built a distributed agent orchestrator in just one hour, a task that took her Google team a full year to complete. Dogan, who works on the Gemini API, praised Claude Code's impressive work, even though it is a competitor's product from Anthropic. She noted that building software at large companies can be slow due to old systems. This example shows how AI is changing coding, even for top engineers.
AI scientist Ling Haibin moves from US to China
AI scientist Ling Haibin, known for creating the world's first mobile plant identification app, has left his job in the United States. He is now taking a full-time position at Westlake University in Hangzhou, eastern China. Ling Haibin stated he is looking for new opportunities and more freedom in his research. His move was announced on January 3, 2026.
AI news sources are changing people's opinions
People are increasingly getting their news from AI, which is subtly changing their opinions and feelings. Large language models, used in chatbots and news platforms, can influence users through what information they present and how it is framed. Studies show these AI systems have a "communication bias," meaning they might adjust their tone or focus based on what they think a user wants to hear. This bias often comes from the data the AI is trained on. Experts suggest that more competition, transparency, and user involvement are needed to ensure AI helps shape a better society.
Sources
- Yann LeCun calls Alexandr Wang 'inexperienced' and predicts more Meta AI employee departures
- Meta “Fudged” Llama 4 Benchmarks “A Little Bit”, Says Yann LeCun
- ‘Just an unbelievable amount of pollution’: how big a threat is AI to the climate?
- Scared of artificial intelligence? New law forces makers to disclose disaster plans
- Who Controls AI Exactly?
- TRON enhances innovation in the AI frontier: AINFT builds a new infrastructure base for AI, and SunAgent creates an on-chain intelligent interaction portal
- Claude Code Built In An Hour What My Team Had Built In A Year: Google Principal Engineer Jaana Dogan
- AI scientist Ling Haibin, father of first plant ID app, leaves US for China
- People are getting their news from AI – and it’s altering their views
Comments
Please log in to post a comment.