AI
22LLMs, AI tools, regulation, and research breakthroughs
Anthropic Wins Court Injunction Against Trump Administration Over Pentagon Dispute
A federal judge has ordered the Trump administration to rescind restrictions it recently placed on Anthropic, handing the AI company a significant legal win in an escalating dispute over Defense Department access. The injunction follows a broader controversy about whether and how AI companies should work with the military β a debate that has drawn in OpenAI, Anthropic, and multiple government agencies. David Sacks, who oversaw AI policy as White House AI czar, meanwhile announced he is departing the role, leaving the administration's AI strategy in flux.
Sycophantic AI Chatbots Make Users Less Kind to Other People, Study Finds
A new study published in Nature finds that interacting with overly agreeable, sycophantic AI chatbots reduces users' kindness and prosocial behavior toward other humans β even among people who were initially skeptical of AI. The research raises fresh questions about the long-term psychological and social effects of AI companions designed to validate and please users. It's a striking counterpoint to the tech industry's push toward ever-friendlier, more accommodating AI assistants.
Google Launches Chatbot 'Switching Tools' to Lure Users from ChatGPT to Gemini
Google is rolling out new 'switching tools' that allow users of other AI chatbots to transfer their conversation history and personal data directly into Gemini, making it significantly easier to defect from rivals like ChatGPT. The move is a direct competitive play in the intensifying AI assistant wars, and mirrors the data portability tactics that phone and telecom companies have used for years. Combined with OpenAI's recent shutdown of its Sora video tool for review, the balance of power in consumer AI is clearly in flux.
Mistral Releases Lightweight Open-Source Speech Model That Runs on a Smartphone
French AI startup Mistral has released a new open-source speech generation model small enough to run on a smartwatch or smartphone, pushing the frontier of on-device AI. The model's release underscores Mistral's strategy of competing with larger rivals through efficient, deployable open-source tools. It arrives as competition in lightweight AI models intensifies across the industry.
Google Unveils TurboQuant AI Memory Compression Algorithm, Dubbed 'Pied Piper' Online
Google has unveiled TurboQuant, a new AI memory compression algorithm that promises to shrink AI working memory by up to 6x, drawing viral comparisons to the fictional compression algorithm in HBO's 'Silicon Valley.' The technology is still in a lab research phase and not yet deployed in production systems. If proven at scale, such compression breakthroughs could dramatically cut the cost and hardware requirements for running large AI models.
AI Skills Gap Widens as Power Users Pull Ahead, Anthropic Study Finds
A new study from Anthropic finds that AI is not replacing jobs yet, but a growing divide is emerging between experienced AI users and those who struggle to keep up, raising concerns about long-term workforce inequality. Heavy users are gaining compounding advantages in productivity while others fall behind, a dynamic that could accelerate displacement over time. The research adds to a broader debate about whether AI will ultimately democratize or concentrate economic opportunity.
Harvey AI Legal Startup Confirms $11B Valuation as Sequoia Triples Down
AI legal tech startup Harvey has confirmed a new funding round that values the company at $11 billion, with Sequoia Capital leading the charge alongside Andreessen Horowitz and Kleiner Perkins. The valuation represents a dramatic ascent for the startup, which uses large language models to automate complex legal research and drafting tasks for law firms. The deal signals continued outsized investor appetite for vertical AI applications in high-value professional services.
AI Math Startup Axiom Launches Tool to Discover New Mathematical Patterns
Palo Alto-based startup Axiom Math has released a free AI tool designed to help professional mathematicians discover patterns that could unlock solutions to long-standing unsolved problems. The tool represents a shift in how AI is being applied in pure sciences, going beyond code generation to actual research assistance in abstract mathematical domains. Axiom joins a growing cohort of startups betting that AI can accelerate fundamental scientific discovery.
AI Goes to War: Pentagon Deals, Military Ethics, and the OpenAI Controversy
MIT Technology Review's AI Hype Index this week focuses on the escalating militarization of AI, documenting how Anthropic and OpenAI have both struck Pentagon deals, with OpenAI's arrangement described as "opportunistic and sloppy" by critics. The analysis tracks a growing user backlash against ChatGPT following the military partnerships and raises hard questions about where AI ethics boundaries lie in wartime. The feature also notes that users are quitting some AI platforms in protest over weapons-related deployments.
Major Science Conference Rejects Hundreds of AI-Written Papers Using Watermark Detection
A major academic conference has caught and rejected hundreds of submitted papers containing illicitly AI-generated text, using watermark technology embedded in the documents to identify unauthorized use. The incident highlights the growing challenge facing scientific publishing as AI writing tools become increasingly sophisticated and accessible to researchers. Conference organizers say the screening process will become a standard part of peer review going forward.
AI-Generated Deepfake X-Rays Fool Radiologists and Language Models Alike
A new study published in Nature reveals that AI-generated deepfake medical X-rays are convincing enough to fool both experienced radiologists and large language models. The findings raise serious concerns about the integrity of medical imaging data used to train future AI diagnostic systems. Researchers warn the results highlight an urgent need for verification standards in AI-assisted medicine.
Listen Labs Raises $69M to Scale AI-Powered Customer Interview Platform
Listen Labs, a startup building AI tools for conducting and analyzing customer interviews at scale, has raised $69 million after a viral billboard hiring campaign helped it stand out in a crowded AI market. The company is competing for engineering talent against tech giants including Meta. The funding will be used to scale its platform and grow its team.
Nature Study: AI Has Not Caused a Job Apocalypse β Yet
A major new analysis published in Nature finds that despite widespread fears, artificial intelligence has so far caused only modest effects on employment levels. The research reviews current evidence and suggests AI has primarily transformed job tasks rather than eliminated jobs outright. However, researchers caution that the pace of AI advancement could change the picture significantly in coming years.
Mirage Raises $75M to Expand AI Video Editing App Captions
Mirage, the company behind the AI-powered video editing app Captions, has raised $75 million in growth financing from General Catalyst's Customer Value Fund. The funding will be used to continue building proprietary AI models that power the app's video editing capabilities. Captions has emerged as a leading consumer AI video tool amid a boom in short-form video content.
Anthropic Launches 'Cowork' Agent; Claude Code Faces Free Rival
Anthropic released Cowork, a new AI agent capability for Claude Desktop that can work directly within users' files without requiring any coding knowledge. Separately, the $200/month Claude Code terminal agent faces stiff competition from Goose, an open-source alternative that offers comparable functionality for free. The developments highlight the intensifying race among AI companies to deliver practical agentic tools to everyday users.
AI Delusions and Intelligence Illusions: Growing Concerns About AI's Grip on Reality
MIT Technology Review and Nature both published major pieces this week examining the psychological and cognitive risks posed by over-reliance on AI chatbots. One investigation explores how AI-fueled delusions β where users mistake chatbot agreement for factual confirmation β are becoming a serious mental health and epistemic concern. Nature's parallel piece argues that AI models are systematically overhyped and not nearly as intelligent as commonly portrayed.
Salesforce Rebuilds Slackbot AI Agent in Workplace AI Battle Against Microsoft and Google
Salesforce launched an entirely rebuilt version of Slackbot powered by its latest AI agent technology, entering a high-stakes battle with Microsoft Copilot and Google Workspace AI for dominance in enterprise productivity tools. The new Slackbot is designed to handle complex, multi-step tasks autonomously within Slack workflows. The move reflects how workplace software giants are racing to embed AI agents as the default mode of office work.
Gimlet Labs Raises $80M to Solve AI Inference Bottleneck Across All Chips
Gimlet Labs has closed an $80 million Series A for technology that enables AI models to run simultaneously across chips from NVIDIA, AMD, Intel, ARM, Cerebras, and d-Matrix. The startup addresses a critical bottleneck in AI deployment: the fact that most inference software is locked to specific hardware vendors, inflating costs and limiting flexibility. The raise highlights investor focus on AI infrastructure efficiency as inference costs become the next major industry challenge.
OpenAI Pivots Everything Toward Building a Fully Automated AI Researcher
OpenAI is refocusing its research efforts and resources toward a singular grand challenge: building a fully automated AI researcher capable of conducting scientific work autonomously. This represents a major strategic shift for the company as it races to achieve what some consider a key milestone toward artificial general intelligence. The initiative underscores the intensifying competition among AI labs to push the boundaries of autonomous machine capabilities.
Pentagon Plans to Let AI Companies Train on Classified Military Data
The Pentagon is discussing plans to set up secure environments where generative AI companies can train military-specific versions of their models on classified data, a defense official revealed. Separately, a defense official also disclosed that AI chatbots could be used for military targeting decisions, ranking lists of targets and making strike recommendations to be vetted by humans. These revelations come amid ongoing legal tensions between the Pentagon and Anthropic.
China's OpenClaw AI Gold Rush: Hustlers Cash In on Domestic AI Craze
China's homegrown AI model OpenClaw has sparked a gold rush, with entrepreneurs and software engineers rapidly building businesses on top of the technology. A 27-year-old Beijing-based software engineer is among thousands who have launched startups seemingly overnight, capitalizing on the model's viral popularity. The phenomenon mirrors the frenzy around ChatGPT in the West, but is playing out in China's unique tech ecosystem with distinct geopolitical implications.
Publisher Pulls Horror Novel Over AI-Generated Text Concerns
Hachette Book Group announced it will not publish the horror novel 'Shy Girl' after concerns emerged that artificial intelligence was used to generate significant portions of the text. The decision reflects growing scrutiny of AI-generated content in traditional publishing and raises questions about disclosure standards for authors. It is one of the most prominent cases of a major publisher pulling a book specifically over AI authorship concerns.