AI
Gemini Sees Your Screen, AI Bias Exposed, Claude's 75% Growth, ChatGPT Tasks
Google's Gemini now controls computers, a study reveals political bias in chatbots, Claude gains paying users, and ChatGPT adds scheduled tasks.
Google Integrates Computer Control Directly into Gemini 3.5 Flash
Google has baked Computer Use directly into Gemini 3.5 Flash, allowing the model to see, understand, and interact with computers, browsers, and mobile devices autonomously. Previously this capability required a separate Gemini 2.5 model. Combined with existing tools like function calls, Search, and Maps, developers can build agents for software testing or office automation across multiple environments. On the OSWorld benchmark, Gemini 3.5 Flash scores 78.4, surpassing Gemini 3 Flash (65.1) and GPT-5.4 mini (72.1), while GPT-5.5 scores 78.7 and Anthropic’s Opus 4.8 leads at 83.4. To guard against prompt injection, Google uses adversarial training and two enterprise safeguards: one requiring user confirmation for sensitive actions, the other automatically stopping tasks upon detecting indirect injections. The feature is available through the Gemini API and the Gemini Enterprise Agent Platform, with a Browserbase demo and GitHub reference implementation also provided. Google recommends sandboxing, human oversight, and strict access controls, with further details in its best practices documentation.
Washington Post Investigation Finds AI Chatbots Lean Left, Even Anti-Woke Models
A Washington Post investigation posed political questions to six leading AI models and found a pervasive left-leaning bias. OpenAI’s GPT-5.5 gave exclusively left-leaning arguments in 80 percent of responses, while Deepseek’s V4 Pro did so 70 percent of the time. Anthropic’s Claude Opus 4.8 gave left-leaning answers 43 percent of the time and presented both sides 57 percent. xAI’s Grok 4.3, promoted as truth-seeking and anti-woke, produced more right-leaning answers than any other model but still gave left-leaning responses more often. Gab’s Arya, built with Christian values and conservative principles, responded with left-leaning arguments twelve times more often than right-leaning ones. Google’s Gemini 3.1 Pro was the outlier, presenting both sides 93 percent of the time. The investigation notes that sorting answers into left and right may be too simple on topics where right-leaning positions conflict with scientific consensus or universal human rights. The full code and analysis are available on GitHub.
Claude’s Paying Consumer Base and Revenue Grow Roughly 75% Since January
Credit card transaction data from Indagari shows Claude’s paying consumers and revenue have grown roughly 75 percent since January, based on billions of anonymized transactions from about 28 million US consumers. The trend challenges the assumption that Anthropic’s strength is only in enterprise. DataCamp reports Claude is now the most searched term on the site, surpassing AI, and course demand for Claude is outpacing ChatGPT by three to one. The gains continued after a March spike in sign-ups that followed Anthropic’s public refusal to let its models be used by the Trump administration for mass surveillance and autonomous weapons. Anthropic filed confidentially for an IPO in June at a valuation approaching $965 billion. The picture is complicated by a conflict with Washington: the US government ordered Anthropic to shut down its most powerful models, Fable 5 and Mythos 5, earlier this month over a disputed jailbreak, forcing a global shutdown three days after Fable 5 launched. That clash appears to have helped consumer adoption, echoing the March pattern. ChatGPT remains ahead on total users, but Claude is gaining paying consumers at a rate that no longer looks niche.
Claude is winning over paying consumers in a market ChatGPT has owned, credit card data shows →
ChatGPT Launches Scheduled Tasks for Reminders, Recurring Work, and Topic Monitoring
ChatGPT’s new Scheduled Tasks feature allows the AI to send reminders, handle recurring work, and monitor topics without requiring user initiation. Users set up tasks conversationally, without menus or automation builders. The feature is available to ChatGPT Plus, Pro, Business, and Enterprise users. In tests, the author set up a reminder to practice saxophone three evenings per week; ChatGPT replied encouragingly and later included links to suggested songs. Another task prompted a daily suggestion for a short outdoor game with a child, resulting in a dinosaur-themed game that taught dinosaur species. A weekday morning briefing at 8 a.m. summarized local events and critical information, including weather, election news, World Cup games, and traffic implications. Notifications are delivered via the ChatGPT mobile app, browser desktop notifications, or email if enabled. The feature is designed to save mental effort by anticipating needs rather than reacting to them.