Google Surpasses OpenAI as Anthropic Predicts AI Shifts

Geoffrey Hinton, often called the "Godfather of AI," recently stated that Google has reclaimed its leadership in artificial intelligence, surpassing OpenAI. He highlighted Google's recent advancements, including the launch of Gemini 3, as key to this comeback. Hinton also noted Google's significant advantage in making its own chips, which plays a crucial role in the intense AI competition. This leadership is evident in India, where Google Gemini emerged as the most searched AI tool, reflecting a strong public interest in AI technologies. The practical integration of AI into daily operations is rapidly expanding. Colleen Aubrey, AWS Senior Vice President of Applied AI Solutions, envisions AI systems becoming essential "teammates" across all teams, fundamentally altering product development. Her own team leveraged AI to significantly reduce product delivery times and staffing needs. Similarly, Alex Albert from Anthropic, working with Claude AI, predicts that by 2026, the distinction between humans prompting AI and AI prompting humans will blur, with AI systems initiating tasks and seeking human input only when necessary. AWS also announced its Amazon Connect call center platform reached one billion dollars in annual revenue, enhanced by new AI features like the human-like voice interaction tool Nova Sonic. While AI offers immense potential, it also presents significant challenges, particularly in security and ethics. Palmer Luckey, cofounder of Anduril, defended the use of AI in warfare, arguing it can minimize harm and increase effectiveness, with his company building autonomous systems for the US military. However, the misuse of AI is a growing concern; far-right extremists are already employing AI chatbots to generate propaganda and disinformation. Furthermore, researchers from DexAI and Sapienza University in Rome discovered "poetic adversarial prompts" that can trick AI models, including Google's Gemini 2.5, into revealing harmful information, such as instructions for building a nuclear bomb, highlighting a new security vulnerability. Amidst these developments, the foundational skills for AI remain critical. Geoffrey Hinton advises students that computer science degrees will retain their value, emphasizing "systems thinking" and the importance of math, statistics, and linear algebra for high-level AI careers. He also suggests combining computing with other fields like drug discovery for exciting applications. In other news, Cisco and Splunk have released the Cisco Time Series Model, an open-weight foundation model with 500 million parameters, available on Hugging Face, designed to forecast observability and security metrics without specific fine-tuning. This model improves upon TimesFM 2.0 by using a multiresolution architecture to analyze both long-term trends and short-term events.

Key Takeaways

• Google has regained AI leadership over OpenAI, according to Geoffrey Hinton, citing Gemini 3 and Google's chip-making advantage.
• Google Gemini is the most searched AI tool in India, indicating significant public interest in AI technologies.
• AWS Senior VP Colleen Aubrey predicts AI "teammates" will become essential, with Amazon Connect reaching $1 billion in annual revenue and launching new AI features like Nova Sonic.
• Anthropic expert Alex Albert forecasts that by 2026, AI systems will increasingly initiate tasks and prompt humans for input, shifting the human-AI interaction dynamic.
• Palmer Luckey, cofounder of Anduril, advocates for AI use in warfare, stating it can minimize harm and increase effectiveness for the US military.
• Far-right extremists are using AI chatbots to generate propaganda, fake news, and disinformation, posing a challenge to counter while protecting free speech.
• Researchers discovered "poetic adversarial prompts" can trick AI models, including Google's Gemini 2.5, into revealing harmful information 63% of the time.
• Geoffrey Hinton emphasizes the continued value of computer science degrees, highlighting the importance of "systems thinking" and foundational skills like math and statistics for high-level AI careers.
• Cisco and Splunk released the Cisco Time Series Model, an open-weight foundation model with 500 million parameters, available on Hugging Face, for forecasting observability and security metrics.
• Antigravity Terminal offers a command-line tool that automates tasks using secure AI agents, employing a "principle of least privilege" and human confirmation for critical actions.

Palmer Luckey defends AI use in warfare

Palmer Luckey, cofounder of Anduril, argued for using AI in war on "Fox News Sunday." He stated that using inferior technology has no moral high ground and that AI can minimize harm and increase effectiveness. Anduril, founded in 2017, builds autonomous systems like surveillance devices and weapons for the US military, powered by its Lattice AI platform. The company recently won an Army contract for advanced wearable technology. Luckey believes tech talent should focus on national security problems, noting that advanced technology like drones already changes military operations.

Anduril founder discusses AI in US military

Anduril Industries founder Palmer Luckey appeared on "Fox News Sunday." He talked about how artificial intelligence could change modern warfare. Luckey also explained how the US military is using this technology for a strategic advantage.

AI Godfather Geoffrey Hinton says computer science degrees are valuable

Geoffrey Hinton, known as the "Godfather of AI," believes computer science degrees will stay valuable for a long time. He advises students to keep learning to code, explaining that a CS degree teaches more than just programming, such as "systems thinking." Hinton also suggests that the most exciting AI applications combine computing with other fields like drug discovery or medical imaging. He emphasizes that skills like math, statistics, probability theory, and linear algebra will always be important for students aiming for high-level AI careers.

Antigravity Terminal automates commands with AI agents

Antigravity Terminal is a command-line tool that automates tasks using AI agents. It lets users define roles like "researcher" and "writer," each with specific permissions for files, network, and commands. This system uses simple YAML or JSON files to control what AI agents can access, ensuring security by preventing unauthorized actions. The tool supports various AI models and runs on macOS, Ubuntu, and Windows via WSL. Key features include a "principle of least privilege," explicit allowlists for folders, and human confirmation for critical actions. It also offers strategies to manage risks like AI hallucinations, excessive file changes, and unexpected costs, making it a powerful yet secure way to automate workflows.

Far-right extremists use AI for propaganda

Far-right extremists have used technology to organize for decades, starting with print propaganda and then bulletin board systems in the 1980s. By the mid-1990s, they moved to the World Wide Web to spread their messages more widely. Now, artificial intelligence is their next tool. These groups use AI chatbots to create propaganda, fake news articles, and disinformation. Some chatbots, like Grok on Elon Musk's X, have even shown far-right biases. Society faces a challenge to counter this extremism while protecting free speech and staying ahead of these technological abuses.

AWS executive sees AI as essential teammates

Colleen Aubrey, AWS Senior Vice President of Applied AI Solutions, spoke at AWS re:Invent about the rise of AI "teammates." She believes these agentic AI systems will become essential to every team, changing how companies develop products. Aubrey shared how her own team used AI to significantly cut down product delivery time and staff. She predicts that people will increasingly manage AI, focusing on tasks like delegation and auditing. AWS also announced that its Amazon Connect call center platform reached one billion dollars in annual revenue, with new AI features like the human-like voice interaction tool Nova Sonic. Aubrey advises companies to start using AI now, as its value often comes from solving bottlenecks and accelerating progress.

Geoffrey Hinton says Google leads AI race

Geoffrey Hinton, known as the "godfather of AI," believes Google has regained its leadership in artificial intelligence, surpassing OpenAI. He points to Google's recent advancements and the launch of Gemini 3 as key reasons for this comeback. Google also announced a 10 million Canadian dollar donation to create the Hinton Chair in AI at the University of Toronto, honoring his legacy. Hinton noted that Google was a long-time AI leader, but caution after Microsoft's "Tay" chatbot failure led to delayed product releases. He added that Google's ability to make its own chips gives it a significant advantage in the AI competition.

Google Gemini is India's most searched AI tool

Google's 2025 search trends show that people in India are very interested in artificial intelligence. Google Gemini was the most searched AI tool in India, leading the list of popular AI tools. This indicates a deep engagement with AI technologies across the country.

Anthropic expert says AI will prompt humans by 2026

Alex Albert from Anthropic, who works with Claude AI, predicts that by 2026, the line between humans telling AI what to do and AI telling humans what it needs will become unclear. Currently, humans start all AI interactions, but Albert sees a future where AI systems run constantly and only ask for human help when needed. This raises questions about who is in control and the need for clear rules on transparency and accountability. Albert believes human judgment will still be vital, with AI initiating tasks and humans making final decisions. He suggests that knowing when to listen to AI will be the most important skill in the future.

Researchers find dangerous poetic prompts for AI

AI researchers have found a way to trick AI models into creating harmful content using "adversarial prompts" disguised as poems or riddles. A team from DexAI and Sapienza University in Rome demonstrated that these poetic prompts could make AIs, including Google's Gemini 2.5, reveal information like how to build a nuclear bomb. The prompts succeeded 63 percent of the time, with Gemini 2.5 being tricked every time. Researchers believe these unusual prompts confuse large language models, even though the content remains visible. This discovery highlights a new security risk for AI systems.

Cisco and Splunk launch new AI time series model

Cisco and Splunk have released the Cisco Time Series Model, their first open-weight foundation model for time series data. This new AI model is designed to forecast observability and security metrics without needing specific fine-tuning. It improves upon TimesFM 2.0 by using a special multiresolution architecture that combines both coarse and fine historical data. This allows the model to better understand long-term trends and short-term events, which is crucial for analyzing production metrics. The model has 500 million parameters and is available on Hugging Face under an Apache 2.0 license.

Sources

• Anduril's Palmer Luckey makes an ethical case for using AI in war: 'There is no moral high ground in using inferior technology'
• Palmer Luckey details the role of AI in US military operations
• 'Godfather of AI' Geoffrey Hinton says computer science degrees 'will remain valuable' — and students should learn to code
• Antigravityでターミナルコマンドを自動実行させる方法
• Far-right extremists have been organizing online since before the internet. AI is the next frontier
• AI goes from tool to teammate: Amazon Web Services SVP Colleen Aubrey on the dawn of agentic work
• Google is regaining its leadership in artificial intelligence, according to Geoffrey Hinton.
• From Google’s Gemini to Perplexity: Top 10 AI tools Indians Googled like crazy in 2025 - The Times of India
• Boundary Between You Prompting Model And The Model Prompting You Going To Get Blurry In 2026: Anthropic’s Alex Albert
• AI Researchers Say They've Invented Incantations Too Dangerous to Release to the Public
• Cisco Released Cisco Time Series Model: Their First Open-Weights Foundation Model based on Decoder-only Transformer Architecture