Breaking AI & Tech News Analyzed

Lead: A Columnist’s Warning About AI’s Empty VoiceNesrine Malik contends that AI‑generated text is fundamentally meaningless, a fact that makes it dangerously suited to today’s political climate of repetitive, low‑emotion rhetoric. She describes a personal “nightmare scenario” where AI research tools introduce misquotes and dilute the writer’s own intellectual labor.The Column’s Core Claim: AI Lacks Humanity and Fuels Empty Political RhetoricMalik frames AI as a “tinny chant” that pervades everything from customer‑service bots to social‑media posts, stripping language of its personal alchemy. She argues that while AI can mimic styles, it cannot generate truly original voices, leaving writers dependent on a chorus of existing tones.Lack of Quantitative Data – Qualitative Observations OnlyNo financial or usage statistics are cited in the piece.The argument relies on anecdotal evidence: misattributed quotes, a Commonwealth short‑story controversy, and personal writing habits.References to external research (e.g., a Time study) suggest AI may reduce brain engagement, but no specific figures are provided.Implications for Journalism, Politics, and Public DiscourseThe column warns that AI’s bland, repeatable tone amplifies disinformation and enables political actors to hide behind “empty slogans.”Keir Starmer‑like voices are cited as examples of how AI‑styled language can mute genuine ideological expression, allowing extremist narratives to surface unchecked.Future Trajectory of Human Authorship in an AI‑Saturated LandscapeMalik predicts a growing cultural atrophy if writers continue to outsource research and prose to LLMs. She urges a conscious resistance to preserve the “social contract” of trust and authenticity, suggesting that the battle for credible, human‑crafted content will define the next era of public communication.

The Rise of Recursive Self-Improvement in AIThe word "recursion" is the latest buzzword in AI circles. Two separate startups have taken on the name, and many more have started referencing recursive self-improvement (RSI) in their roadmaps. Like AGI before it, RSI has become a three-letter byword for a cataclysmic AI takeoff – even if there's still a little disagreement about what it exactly means.In basic terms, RSI refers to an AI system that can continuously upgrade itself. Once AI systems can manage the upgrade cycle better than humans, the process can become a closed loop, limited only by the compute power they can access, and humans are no longer necessary or even helpful.Scary or not, that's a vision that a lot of AI labs are eager to chase.Key Players Pursuing Recursive SystemsEarlier this month, well-known AI researcher Richard Socher launched the aptly named Recursive Superintelligence with RSI as an explicit goal. "Our main focus is to build truly recursive, self-improving superintelligence at scale," Socher told TechCrunch at launch, "which means that the entire process of ideation, implementation, and validation of research ideas would be automatic."A number of other prominent researchers are already chasing that same goal, hoping for a breakthrough that will make recursive self-improvement possible.One of the most prominent is Andrej Karpathy, a legendary figure from Tesla and OpenAI, who is using agent swarms to train LLMs on simple tasks for a project he calls Auto-Research. Karpathy has been unusually open about the project, tweeting about milestones regularly and making the building blocks available through a public GitHub repo. So far, the work has mostly been confined to making minor improvements on a GPT-2 scale model — as Karpathy noted in March, "It's not novel, ground-breaking 'research' (yet)" — but it's been enough to convince lots of other researchers to follow the RSI dream. And with Karpathy now working on pre-training at Anthropic, he will have plenty of opportunity to apply the idea at a larger scale.Adaption — founded by Cohere and Google alum Sara Hooker — recently launched a similar tool called AutoScientist in an effort to automate frontier training. Like Karpathy's auto-researchers, the system trains agents to make incremental improvements — but for Adaption, the goal is to make it easier to train a full-scale frontier model. If those same researchers start to push the frontier forward, the system could quickly spiral into something very much like RSI.Disarray founder Doris Xin drew more specific RSI interest when her self-trained machine learning agent took home 28 medals in a recent Kaggle competition, beating out many human-trained agents. As she sees it, the major challenge is reliability."I would argue, given infinite compute and infinite time horizon, we are already there," Xin told me. "I want to make an argument that this is not a creative endeavor, really. It's just a lot of meat-and-potatoes engineering."The Current State of Self-Improving AIThere's also plenty of evidence that the AI industry isn't very close to recursive systems in any meaningful way — and is still grappling with talking to a wary public about its progress. So Google CEO Sundar Pichai basically admitted in a recent podcast interview."It's a continuum, and we are all definitely making progress," Pichai said. "But in the way people describe RSI, that would represent a next level of acceleration and would have a lot of implications, but we aren't quite there yet."But the continuum includes an awful lot of self-improving AI systems.In January, one of Anthropic's lead programmers for Claude Code estimated that "close to 100%" of his team's code was written by the tool — a frank admission that Claude Code was literally writing itself.Just because engineers are using an AI tool doesn't mean the tool can replace them — but Anthropic seems to be getting close to replacing engineers too. In a recent survey tied to the Mythos preview, five out of 18 Anthropic engineers believed that, with harness improvements, this version of Mythos could soon substitute for an L4 engineer — a midlevel programmer who can take on involved projects without supervision.Still, there were some of the same weaknesses you might expect."Some of Claude's major reported weaknesses compared to an L4 include: self-managing week-long ambiguous tasks, understanding org priorities, taste, verification, instruction-following, and epistemics," the report reads.In other words, its weaknesses are everything involved with self-direction, which is the cornerstone for RSI. But sure, for everything else, Claude is ready to step right in.Expert Perspectives on RSI TimelinesJust like the AGI term before it, the AI industry also can't tell us how far away it is from showcasing a meaningful recursive system. When Georgetown's Center for Security and Emerging Technology assembled a group of experts to study RSI last year, the group found a major split in assessments — some expecting an imminent "superintelligence" style explosion while others expected slower progress and an eventual plateau. But all agreed that recursion made the future especially difficult to predict.Helen Toner, director of CSET and a former board member at OpenAI, told TechCrunch that simply using AI tools to do AI research isn't enough to qualify as RSI. "They're just using AI for as much as they can," Toner told TechCrunch. "And I think that is different from the classic definition of RSI, which is really that there are no humans needed."Toner pointed to a recent post by METR's Ajeya Cotra, which distinguishes different milestones on the path to the AI research takeover. One step, which Cotra calls "adequacy," would come when the system can still perform research after all humans are removed — even if the resulting research isn't as valuable or efficient. "Parity" comes when an AI-only system is as good at research as a human-only system. "Supremacy," the final stage, comes when an AI-only system outperforms a collaborative system between humans and AI.Ultimately, Cotra concludes that AI is very close to the adequacy threshold of being able to produce some work on its own — similar to the incremental changes made by Karpathy's Auto-Research system. "I wouldn't be totally shocked if you told me this milestone had already passed, and I expect it to happen in the next couple years," Cotra wrote.She was less clear on when parity will come, but once it does, she thinks it would "massively accelerate the pace of AI progress, leading to AI research supremacy within another year."The Challenges Ahead for Recursive AIWith so much of AI built on scaling laws, there's a strong tendency to think RSI will follow the same curve. Toner thinks that many of those pursuing AI research and development via RSI "think of it as a pretty smooth ladder, where you can just keep scaling up."But even if AI researchers are able to make incremental improvements like Karpathy's auto-researchers, there will be larger challenges in handing off the whole process of research. Toner put it in terms of the history of computing, which has seen human beings handing off more and more of the process while still directing things from the top."We went from machine languages to assembly language and compiled languages; you're getting further and further from the guts of the computer," Toner said. "But the human is still, in some intuitive sense, running the show."Moving beyond that paradigm will take significant challenges, both in engineering and alignment. But even with the massive investments happening, there's no infinite compute available — and the basic trade-off between human labor and machine intelligence will be hard to overcome.The Future of Recursive Self-ImprovementAs for a total recursive AI system of apocalyptic visions? The only thing researchers essentially agree on is that, like AGI, it's not here yet.

Google’s AI Overview Stumbles on Simple Letter Counting Google’s newly rolled‑out AI Overview feature in Search incorrectly counted letters in everyday words – claiming there are two “P”s in “Google”, one “r” in “poop”, and even misspelling “journalism”. The blunders highlight a long‑standing weakness of large language models (LLMs) when it comes to exact spelling. The Miscounted Letters Behind the New Search AI “Google” – AI said 2 Ps (actual: 0) “poop” – AI said 1 r (actual: 0) “journalism” – AI said 2 d’s (actual: 0) U.S. President’s last name – AI reported 1 P but rendered “t‑r‑p‑u‑m” Quantifying the Miscounts: Numbers Behind the Errors Beyond the anecdotal examples, the AI also produced a faulty definition for the word “disregard”, responding with “Understood. Let me know whenever you have a new prompt or question!” This illustrates that token‑based encoding can produce nonsensical outputs even when the input is a single word. Implications for Search Trust and AI Adoption Google’s AI‑driven overhaul aims to make generative responses the centerpiece of its 29‑year‑old search product. Repeated factual and spelling errors risk eroding user confidence, especially after earlier AI Overviews cited satirical sources and gave absurd advice such as “eat rocks”. Trust in AI‑generated answers remains a critical hurdle. What’s Next for Google’s Generative Search? Google told TechCrunch it is “working to fix this particular issue” and will likely refine its tokenizer and post‑processing pipelines. Industry observers expect incremental improvements rather than a complete architectural shift, meaning users may continue to see occasional glitches while the broader AI‑search strategy matures.

The Architecture of Agentic FinanceRobinhood is fundamentally redefining the user experience by launching support for AI agentic trading and a new agentic credit card. This initiative allows users to create separate accounts for their AI agents, connecting them to a dedicated wallet. While these agents can analyze portfolios and suggest strategies, they are restricted to executing trades using only pre-loaded balances. The platform ensures safety through a mandatory approval workflow for trade previews and employs a dedicated fraud detection team to review suspicious activities.Protocol Integration: Agents connect via the Model Context Protocol (MCP) to analyze concentration risk and sector exposure.Control Mechanism: Users receive real-time notifications and can monitor all agent activities within the app.Current Scope: The beta feature is currently limited to stock trading.Expanding the Agentic EcosystemThe rollout of these tools represents a significant expansion of Robinhood's capabilities. The company is not only enabling autonomous trading but also introducing a virtual credit card for AI agents to facilitate payments. Currently, this card is exclusive to Robinhood Gold Card holders, who can link their accounts to set monthly limits and approval preferences. The platform has also outlined a clear roadmap for future asset classes.Upcoming Assets: Support for options, crypto, event contracts, futures, and prediction markets is planned for the near future.Platinum Access: The Robinhood Platinum Card will receive similar agentic card features later this year.Redefining the Role of the TraderThis development marks a pivotal shift in the financial services industry, moving from active manual trading to agentic finance. By adopting the Model Context Protocol (MCP), Robinhood allows users to integrate third-party Large Language Models (LLMs) directly into their investment workflow. This reduces the friction of manual data analysis and positions Robinhood as a central node in the growing network of autonomous financial agents.The Future of Autonomous FinanceAs major players like Stripe, Amazon, and Google race to build similar capabilities, the barrier to entry for AI-driven financial management is rapidly dropping. We predict that by the end of the year, the distinction between a traditional trading account and a managed portfolio will blur, with AI agents becoming the primary interface for routine financial transactions and payments.

The Lead When the founders of a mental health app for men called Mental discovered that one feature — AI interactive audio — was resonating strongly with users, they recognized a significant opportunity. This insight led to the creation of The Path, a new AI therapy application co-founded by renowned motivational speaker Tony Robbins and former Calm employees, which has now secured $14.3 million in seed funding. The Birth of a Safer AI Therapy Platform The Path emerged from observations made by co-founder and CEO Anson Whitmer and co-founder Tyler Sheaffer, who previously worked together at meditation app Calm. Whitmer's personal experiences with suicide in his family inspired him to pursue mental health technology. After working at Calm until 2021, he felt he could make a greater impact by addressing the unique, personal nature of people's mental health challenges. Whitmer sees large language models (LLMs) and AI as the bridge to providing personalized mental health care to everyone, especially given the shortage of therapists worldwide. "What's exciting and game-changing is that, for the first time in my career, I've seen that there's actually this possibility for every single person to have the personalized sort of access and care that they need to really get the help," he said. Funding and Celebrity Endorsement The Path has successfully raised $14.3 million in seed funding, led by Prime Movers Lab where Tony Robbins is a partner. Other notable investors include Olympic speed skater Apolo Anton Ohno, boxer Deontay Wilder, and Designer Fund. After Prime Movers invested, Robbins initially consulted on branding but his enthusiasm grew, leading to him becoming a co-founder. The author has since helped shape The Path into a therapy-plus-coaching app that incorporates his popular self-improvement methods. The app currently offers 11 virtual AI therapists that users can customize based on their preferences for directness and other details. While it's currently free to gain users, The Path plans to eventually charge $40 per month for the service. Superior Safety Benchmarks A key differentiator for The Path is its specially trained AI model, which has scored a 95 on the Vera-MH mental health safety AI benchmark. This significantly outperforms consumer chatbots, which top out at 65 on the same benchmark. According to Whitmer, consumer chatbots are "optimized for engagement," which is counterproductive to effective therapy and coaching that should focus on deep understanding rather than quick solutions. "It's meant to challenge you. It's not just meant to agree with you," Whitmer explains. The Path's AI is designed to help users dig out their assumptions and discover their own solutions rather than simply reinforcing ideas to keep users engaged. The startup's model is post-trained from open source models and doesn't use major consumer LLMs, positioning it as a specialized therapeutic tool rather than a generic chatbot wrapper. Market Potential and Future Outlook The mental health tech market is experiencing significant growth, with OpenAI reporting that at least 900 people use ChatGPT for mental health-related queries every week. This demonstrates the clear demand for AI-powered mental health solutions. However, The Path aims to capture a specific segment of this market by focusing on therapeutic rigor and safety. As mental health awareness continues to grow and technology becomes more sophisticated, AI therapy platforms like The Path could play an increasingly important role in addressing global mental health challenges. The combination of Tony Robbins' brand recognition, the technical expertise of the Calm alumni team, and the specialized focus on therapeutic safety positions The Path as a notable contender in the emerging field of AI-powered mental health care.

The Leadership Shift at Anthropic Andrej Karpathy, the AI researcher who co-founded and formerly worked at OpenAI and previously led AI at Tesla, has joined Anthropic. Karpathy announced his move on X, stating that he is excited to join the team and get back to R&D.; Karpathy's Role in Pre-Training Karpathy started this week at Anthropic, where he is working on pre-training under team lead Nick Joseph. Pre-training is responsible for the large-scale training runs that give Claude its core knowledge and capabilities. Karpathy will start a team focused on using Claude to accelerate pre-training research. The Significance of Karpathy's Move Karpathy is one of the few researchers who can bridge the gap between LLM theory and large-scale training practice. Tapping him to build such a team is a clear sign from Anthropic that it believes AI-assisted research, rather than pure compute, is how it stays competitive with OpenAI and Google. Karpathy's Background Co-founded OpenAI and worked on deep learning and computer vision until 2017 Led Tesla's Full Self-Driving (FSD) and Autopilot programs from 2017 to 2022 Returned to OpenAI for one year before leaving in 2024 to start Eureka Labs, a startup dedicated to applying AI assistants to education Anthropic's Recent Hires Anthropic has also brought on Chris Rohlf to its frontier red team, which stress-tests advanced AI models against severe threats. Rohlf is a veteran of the cybersecurity industry with more than 20 years of experience. The Future of AI Research Karpathy's move to Anthropic and the company's focus on AI-assisted research signal a new direction in the AI landscape. As Karpathy stated, "I think the next few years at the frontier of LLMs will be especially formative."

The Rise of AI-Generated Research Papers ArXiv, a widely used open repository for preprint research, is taking steps to crack down on the careless use of large language models in scientific papers. Although papers are posted to the site before they are peer-reviewed, ArXiv has become one of the main ways that research circulates in fields like computer science and math. ArXiv's New Policy The organization has already taken steps to combat a growing number of low-quality, AI-generated papers, such as requiring first-time posters to get an endorsement from an established author. In its latest move, ArXiv will ban authors for a year if they submit papers that contain "incontrovertible evidence" that the authors did not check the results of LLM generation. The Data Analysis Authors who submit AI-generated papers without proper oversight will face a 1-year ban from ArXiv. After the ban, authors will be required to have subsequent ArXiv submissions accepted by a reputable peer-reviewed venue. The Impact Analysis Recent peer-reviewed research has found that fabricated citations are on the rise in biomedical research, likely due to LLMs. ArXiv's new policy aims to ensure that authors take "full responsibility" for the content, "irrespective of how the contents are generated." The Prediction The move is expected to have a significant impact on the scientific community, as researchers will need to be more transparent about their use of AI-generated content. ArXiv's new policy is a "one-strike" rule, but moderators must flag the issue and section chairs must confirm the evidence before imposing the penalty. Authors will also be able to appeal the decision.

The LeadOsaurus has introduced an innovative open-source, Apple-only LLM server that allows Mac users to seamlessly switch between local and cloud AI models while maintaining data privacy on their own hardware. This development addresses growing concerns about AI token costs and security by providing a user-friendly interface that runs AI in a hardware-isolated virtual sandbox.The Evolution from Dinoki to OsaurusOsaurus evolved from the idea for a desktop AI companion called Dinoki, which Osaurus co-founder Terence Pae described as a sort of "AI-powered Clippy." Dinoki's customers had questioned why they should buy the app if they still had to pay for tokens—the usage units AI companies charge for processing prompts and generating responses. This concern led Pae to develop Osaurus as a solution that allows users to run AI locally on their Macs, accessing files, browsers, and system configurations without relying on cloud services.Technical Capabilities and Model SupportOsaurus can flexibly connect with locally hosted AI models or cloud providers like OpenAI and Anthropic, allowing users to choose which AI models best fit their needs. The platform supports various models including MiniMax M2.5, Gemma 4, Qwen3.6, GPT-OSS, Llama, and DeepSeek V4. It also supports Apple's on-device foundation models, Liquid AI's LFM family of on-device models, and cloud connections to OpenAI, Anthropic, Gemini, xAI/Grok, Venice AI, OpenRouter, Ollama, and LM Studio. As a full MCP (Model Context Protocol) server, it provides access to tools for MCP-compatible clients and ships with over 20 native plugins for Mail, Calendar, Vision, macOS Use, XLSX, PPTX, Browser, Music, Git, Filesystem, Search, Fetch, and more. Recent updates have also added voice capabilities.User Adoption and Market PositionSince launching nearly a year ago, Osaurus has been downloaded over 112,000 times according to its website. The platform distinguishes itself from similar tools like OpenClaw or Hermes by offering an easy-to-use interface for consumers rather than developers, while addressing security concerns through a hardware-isolated, virtual sandbox that limits the AI's scope and keeps users' computers and data safe. Currently, Osaurus' founders, including co-founder Sam Yoo, are participating in the New York-based startup accelerator Alliance.The Future of Local AI and Business ApplicationsOsaurus' founders are exploring potential business applications, particularly in sectors like legal services and healthcare where running local LLMs could address privacy concerns. The team believes that as local AI models become more powerful, they could reduce demand for AI data centers. Pae noted that "the intelligence per wattage—which is like the metric for local AI—has been going up significantly," with local AI evolving from barely being able to finish sentences last year to now being able to run tools, write code, access browsers, and perform various tasks. The vision is for businesses to deploy Mac Studios on-premise, using substantially less power than traditional data centers while maintaining cloud-like capabilities.

The Rise of Legal Tech: Clio's $500M Milestone Clio, a Canadian law firm management software company, has reached a significant milestone: $500 million in annual recurring revenue (ARR). This achievement is a testament to the growing demand for legal tech solutions, particularly those powered by artificial intelligence (AI). AI-Driven Growth in Legal Tech Clio's growth has accelerated sharply since integrating AI into its offering in 2023. The company's ARR surpassed $200 million in mid-2024, doubled that figure by late last year, and now has reached $500 million. According to Jack Newton, co-founder and CEO of Clio, LLMs (Large Language Models) are poised to revolutionize the legal tech industry. The Potential of LLMs in Legal Tech Newton believes that LLMs can leverage the vast repository of existing legal documents, such as contracts and agreements, to automate time-consuming tasks like document review and drafting. This potential is not limited to Clio; other legal tech companies, like Harvey and Legora, are also experiencing significant revenue surges driven by AI. The Competitive Landscape: Anthropic's Move Anthropic's recent announcement of new legal-specific features for its AI model, Claude, has added a new layer of complexity to the competitive landscape. Both Harvey and Legora rely on Claude as a core model, making the dynamic an uncomfortable one: a key supplier is now also a competitor. The Future Outlook Despite these challenges, Newton remains optimistic about the vast potential of the legal AI market. Clio's valuation of $5 billion and its recent $1 billion acquisition of data intelligence platform vLex have positioned the company for continued growth and innovation in the legal tech sector.