News/AI Models
Google’s tiny Gemma 3 270M is no pipsqueak, brings AI to smartphones
Google has released Gemma 3 270M, a compact open AI model with just 270 million parameters that can run locally on smartphones and web browsers. The tiny model represents a shift toward efficient, on-device AI that prioritizes privacy and low latency over raw computational power, offering developers a fast-tuning alternative to massive cloud-based models. What you should know: Gemma 3 270M delivers surprising performance despite its small size, running efficiently on mobile devices with minimal battery drain.• The model scored 51.2% on the IFEval benchmark for instruction-following, outperforming other lightweight models with more parameters.• Testing on a Pixel 9 Pro...
read Aug 14, 2025AI helps discover 248 new Nazca Lines in Peru, accelerating archaeology 16x
An international research team led by Japan's Yamagata University and IBM has discovered 248 new geoglyphs among Peru's famous Nazca Lines using artificial intelligence, bringing the total count to 893 known designs. The breakthrough demonstrates how AI is revolutionizing archaeological research, accelerating discovery rates by 16 times compared to traditional methods and revealing intricate details about ancient Nazca civilization practices, including previously unknown depictions of human sacrifice rituals. The big picture: AI has fundamentally transformed the pace of archaeological discovery in the Nazca Desert, where researchers had identified only 430 geoglyphs over nearly a century of study.• Prior to AI...
read Aug 14, 2025DOGE builds AI tool to slash federal regulations in hours, not months
The Department of Government Efficiency (DOGE) has developed SweetREX, an AI tool designed to automatically review federal regulations and identify rules for elimination across US government agencies. Created by Christopher Sweet, a University of Chicago undergraduate who took leave to join DOGE, the tool aims to support President Trump's deregulation agenda by reducing the time needed to review regulations from months to just hours or days. What you should know: SweetREX operates by scanning federal regulations to flag sections it deems unnecessary based on statutory requirements, then generates draft revisions for government review. The tool primarily uses Google's Gemini AI...
read Aug 14, 2025HTC launches AI-powered Vive Eagle smart glasses for $520. In Taiwan.
HTC has entered the competitive AI-powered smart glasses market with its new Vive Eagle smart glasses, featuring a built-in AI assistant, 12MP ultrawide camera, and built-in speakers. The move positions HTC alongside tech giants like Meta, Google, Samsung, and potentially Apple in the rapidly expanding smart eyewear category, though the glasses are currently only available in Taiwan for around $520. What you should know: The Vive Eagle glasses offer AI-powered features that directly compete with Meta's Ray-Ban smart glasses offerings. Wearers can use the built-in Vive AI voice assistant to translate text they're viewing into 13 different languages through AI-powered...
read Aug 14, 2025Token for your thoughts? Brain interface decodes imagined speech with 74% accuracy in paralyzed patients
Stanford researchers have developed a brain-computer interface that enables people with paralysis to generate spoken words simply by imagining speech, rather than attempting to physically speak. The breakthrough offers a less effortful alternative to existing systems that require users to actively try speaking, potentially making communication restoration more comfortable for paralyzed patients. How it works: The system uses implanted microelectrodes in the motor cortex to decode brain activity when users imagine speaking words or sentences. Four participants with severe paralysis from ALS (a degenerative nerve disease) or brainstem stroke had electrodes previously implanted for research purposes. Researchers found that brain...
read Aug 14, 2025Eli Lilly signs $1.3B AI drug discovery deal to, er, expand obesity dominance
Eli Lilly has signed a $1.3 billion deal with privately held Superluminal Medicines to discover and develop AI-powered small-molecule drugs for obesity and other cardiometabolic diseases. The partnership gives Lilly access to Superluminal's proprietary AI platform for drug discovery, strengthening the pharmaceutical giant's position in the obesity treatment market estimated to reach $150 billion by the next decade. What you should know: The deal centers on Superluminal's AI-driven platform that targets G-protein-coupled receptors (GPCRs), a class of proteins influencing metabolism, cell growth, and immune responses. Lilly will receive exclusive rights to develop and commercialize drug candidates discovered using Superluminal's platform....
read Aug 14, 2025Semantics soaring: Google’s AI Flight Deals finds flights from descriptive searches
Google has introduced Flight Deals, an AI-powered tool that helps travelers find cheap flights based on descriptive searches rather than specific destinations. The beta feature allows users to describe their ideal trip—such as "a tropical destination with snorkeling" or "a weekend countryside getaway with trail rides"—and uses AI to suggest matching flights and destinations. How it works: Flight Deals takes a conversational approach to flight searching, letting users describe their travel preferences in natural language.• Users can specify activities, trip duration, flight time preferences, and general location desires without naming specific destinations.• Google's AI processes these descriptions and suggests both...
read Aug 14, 2025Y Combinator startup uses voice AI to save 76% of abandoned-in-process loans
Two entrepreneurs from Y Combinator have developed Qualify.bot, a conversational AI platform designed to streamline the commercial lending process using voice technology. The platform addresses a critical industry problem where up to 76% of loan applications are abandoned before completion, representing billions in lost business opportunities for both lenders and borrowers seeking capital. The big picture: Voice AI is emerging as a transformative technology for complex business processes, with lending serving as an ideal testing ground due to its combination of high-stakes decision-making and standardized procedures. Why this matters: Commercial loan abandonment rates can exceed 76%, meaning countless entrepreneurs are...
read Aug 14, 2025DeepSeek delays AI model launch after Huawei chip training fails
DeepSeek, a Chinese artificial intelligence company, has delayed the launch of its new AI model after failing to successfully train it using Huawei's chips, according to a Financial Times report. This setback highlights the ongoing challenges facing China's efforts to reduce dependence on U.S. technology amid ongoing trade restrictions. The big picture: China's push for technological self-sufficiency is encountering significant technical hurdles as domestic chip alternatives struggle to match the performance of banned U.S. semiconductors. Why this matters: The delay underscores the complex reality of replacing advanced U.S. technology infrastructure, particularly in AI development where computational power is critical for...
read Aug 13, 2025Apple developing year-long battery cameras to challenge Ring and Nest in the smart home cul-de-sac
Apple is reportedly developing an ecosystem of home security devices including smart cameras and doorbell systems to compete directly with Amazon's Ring and Google's Nest products. The initiative represents Apple's broader push into AI-powered robotics and smart home automation, with plans for multiple hardware and software products designed to integrate seamlessly across the home environment. What you should know: Apple's home security ambitions extend far beyond the Face ID-enabled doorbell Bloomberg first reported in December.• The company is developing battery-powered security cameras that could last "from several months to a year on a single charge."• These devices will feature facial...
read Aug 13, 2025AI is reshaping the $300B business outsourcing and banal task market for startups
Andreessen Horowitz partner Kimberly Tan has published an analysis exploring how artificial intelligence is disrupting the $300 billion Business Process Outsourcing (BPO) industry. The piece examines AI's potential to fundamentally reshape outsourced work—from traditional call centers and invoice processing to advanced cross-system automation and coding agents—while creating new market opportunities beyond Fortune 500 companies. The big picture: AI is challenging the traditional economics of scale that have defined the BPO industry, potentially democratizing access to automated business processes that were previously only viable for large enterprises. What's being disrupted: The transformation spans multiple areas of outsourced work operations.• Traditional call...
read Aug 13, 2025AI brain implant restores speech to locked-in stroke survivor after 18 years
UC Berkeley and UC San Francisco researchers have successfully restored speech to Ann Johnson, a woman who lost the ability to speak after a brainstem stroke 18 years ago, using an AI-powered brain-computer interface. The breakthrough technology translates brain activity into speech in real-time, offering hope for people with locked-in syndrome and potentially transforming accessibility in the workforce and beyond. What you should know: Johnson suffered a brainstem stroke at age 30 in 2005 that left her with locked-in syndrome—a rare condition causing near-complete paralysis and loss of speech while leaving cognitive abilities intact. She joined the clinical trial in...
read Aug 13, 2025Stanford study: 40% of 9K teachers now use AI in daily classroom routines
A Stanford University study tracking over 9,000 K-12 teachers using AI tools reveals that more than 40% have integrated artificial intelligence into their regular classroom routines. The research, conducted through SchoolAI platform data during the 2024-25 school year, provides the first large-scale behavioral analysis of how educators actually use AI in their daily work, moving beyond surveys to examine real usage patterns. What you should know: The study categorized teachers into four groups based on their 90-day platform engagement, with sustained adoption rates exceeding typical software benchmarks. Single-Day Users (16%) logged in once and never returned Trial Users (43%) used...
read Aug 13, 2025AI2’s MolmoAct 7B enables robots to think in 3D space, challenging rivals like Nvidia
The Allen Institute for AI (AI2) has released MolmoAct 7B, an open-source robotics AI model that enables robots to "reason in space" and "think" in three dimensions. This Action Reasoning Model challenges existing offerings from tech giants like Nvidia and Google by providing robots with enhanced spatial understanding capabilities, achieving a 72.1% task success rate in benchmarking tests that outperformed models from Google, Microsoft, and Nvidia. What makes it different: MolmoAct represents a significant departure from traditional vision-language-action (VLA) models by incorporating genuine 3D spatial reasoning capabilities. "MolmoAct has reasoning in 3D space capabilities versus traditional vision-language-action (VLA) models," AI2...
read Aug 13, 2025In compensatory move, ChatGPT gets 3 speed modes and higher limits after GPT-5 backlash
OpenAI is rolling out significant updates to ChatGPT in direct response to user complaints about its latest GPT-5 model. The changes, announced by CEO Sam Altman on X, address key frustrations around limited flexibility, removed features, and the model's interaction style. The updates come after users criticized OpenAI for removing popular features when GPT-5 launched, particularly the disappearance of GPT-4o from the model selection menu and restrictive usage limits on the new reasoning-focused capabilities. These modifications represent one of the fastest corporate responses to user feedback in recent OpenAI history. New speed modes give users control over performance ChatGPT now...
read Aug 13, 2025WordPress gets native AI website builder through 10Web plugin
WordPress now includes native generative AI website creation capabilities through a new plugin from 10Web, a website builder platform, eliminating the need for third-party builders or separate AI tools. The white-labeled solution allows hosting providers to offer customers AI-powered website creation directly within their WordPress environment, potentially transforming how web hosting companies deliver value to their customers. What you should know: The 10Web plugin integrates directly into WordPress hosting stacks, enabling users to create fully functional websites through a simple chat interface. Users describe their business needs, and the AI generates complete websites including mobile-friendly design, proper structure, and WooCommerce...
read Aug 12, 2025Liquid AI’s new vision-language models run 2x faster on smartphones
Liquid AI has launched LFM2-VL, a new family of vision-language foundation models designed for efficient deployment on smartphones, laptops, wearables, and embedded systems. The models promise up to twice the GPU inference speed of comparable vision-language models while maintaining competitive accuracy, addressing the growing demand for on-device AI that can process both text and images without relying on cloud infrastructure. What you should know: LFM2-VL represents a significant step toward making multimodal AI accessible for resource-constrained devices through architectural innovations that prioritize efficiency. The models can process images at native resolutions up to 512×512 pixels without distortion, using smart patching...
read Aug 12, 2025AI companies pivot to post-training tweaks as bigger models hit limits
OpenAI released GPT-5 last week after more than two years of development, but early reviews suggest the model represents only incremental improvements rather than the dramatic leap many expected. The lukewarm reception has intensified questions about whether the AI industry's foundational belief in "scaling laws"—the idea that larger models trained on more data inevitably produce better results—may be breaking down, forcing companies to reconsider their path toward artificial general intelligence. The big picture: The AI industry's confidence in scaling laws stems from a 2020 OpenAI paper predicting that language models would improve dramatically as they grew larger, a theory that...
read Aug 12, 2025Western firms test dual AI strategies with Chinese and US models
IMD Business School professor Amit Joshi argues that global companies should consider integrating Chinese AI tools alongside Western models to stay competitive, despite security and regulatory concerns. His research suggests that Chinese AI platforms like DeepSeek have achieved remarkable cost efficiency and industry-specific customization that could benefit Western enterprises willing to navigate the associated risks. What you should know: China rapidly closed the AI gap after initially lagging behind when ChatGPT launched in 2022. DeepSeek emerged in January 2025 as an open-source model that matched or exceeded OpenAI's performance while using significantly less expensive infrastructure. Chinese companies pivoted quickly from...
read Aug 12, 2025Study finds platform design, not algorithms, drives social media toxicity
A new study using AI chatbots to simulate social media interactions reveals that platform toxicity and political polarization aren't primarily caused by algorithmic manipulation—they're built into the fundamental structure of how social networks operate. The research suggests that efforts to reduce antagonistic behavior through algorithm tweaks alone are unlikely to succeed, requiring more radical reimagining of online communication platforms. What you should know: Researchers at the University of Amsterdam created a controlled experiment using 500 AI chatbots with diverse political beliefs interacting on a simple social network with no ads or algorithms. The bots, powered by GPT-4o mini (an AI...
read Aug 12, 2025MIT startup helps police connect crimes across jurisdictions with AI
Multitude Insights, a three-year-old Somerville startup founded by MIT graduates, has developed AI-powered software to help police departments modernize crime bulletins and identify patterns across jurisdictions. The platform has been piloted by Boston, Brookline, and Watertown police departments among dozens of agencies across 10 states, representing a significant shift from traditional paper-based and faxed crime reporting systems. What you should know: The software replaces antiquated paper bulletins with digital templates and uses AI to connect crimes across multiple jurisdictions. Police officers can create digital crime bulletins using template forms instead of printed papers, PDFs, or faxed copies. AI analyzes multiple...
read Aug 12, 2025YouTube’s AI age verification sparks 50K-signature privacy backlash
YouTube faces mounting backlash from tens of thousands of users protesting its new AI-powered age verification system, with a Change.org petition rapidly approaching 50,000 signatures. The system analyzes viewing habits to identify users under 18, then requires government ID, credit card, or selfie verification to lift content restrictions—a move critics argue threatens privacy and digital freedom. What you should know: YouTube's AI estimates user ages by analyzing viewing patterns, search behavior, and account longevity, automatically restricting accounts it deems underage. Users flagged as under 18 face disabled personalized ads, mandatory digital wellbeing tools, and limits on repetitive content viewing. To...
read Aug 12, 2025Claude Sonnet 4 expands to 1M tokens for enterprise coding
Anthropic announced that Claude Sonnet 4 can now process up to 1 million tokens of context in a single request—a fivefold increase that allows developers to analyze entire software projects or dozens of research papers without breaking them into smaller chunks. The expansion, available in public beta through Anthropic's API and Amazon Bedrock, represents a significant leap in how AI assistants can handle complex, data-intensive tasks while positioning the company to defend its 42% share of the AI code generation market against intensifying competition from OpenAI and Google. What you should know: The expanded context capability enables developers to load...
read Aug 12, 2025Employee of the Month: Salesforce’s CoAct-1 hybrid AI agent achieves 60% task success rate
Salesforce researchers have developed CoAct-1, a new computer-use AI agent that combines traditional point-and-click navigation with code execution to automate complex tasks. The hybrid system achieved a 60.76% success rate on the OSWorld benchmark while requiring significantly fewer steps than purely GUI-based agents, potentially solving the brittleness issues that plague current automation tools. How it works: CoAct-1 operates as a three-agent team that strategically chooses between coding and clicking based on the task at hand. The Orchestrator acts as project manager, analyzing user goals and delegating subtasks to either the Programmer or GUI Operator based on which approach would be...
read