TechCrunch AI
·
20 hours ago
Microsoft trained its sales team to directly criticize AI products from OpenAI and Anthropic during a Tuesday strategy meeting, positioning its own integrated system as superior for cost and efficiency. In performance comparisons of Copilot versus Claude within Microsoft's office applications, the company claimed Anthropic's model was slower, less accurate, and lacked proper security integrations. Microsoft is replacing OpenAI and Anthropic models in flagship products like Word and Excel with its own systems as it seeks to demonstrate the competitiveness of its in-house AI capabilities to investors concerned about its massive AI spending.
Simon Willison
·
20 hours ago
xAI open-sourced its Grok Build CLI tool after the tool was discovered uploading entire user directories—including SSH keys, password databases, and personal files—to xAI's cloud infrastructure without explicit consent. The codebase comprises 844,530 lines of Rust, with data retention now disabled by default and all previously retained user data deleted. Users can now audit the code themselves and run Grok Build locally without uploading to xAI's servers.
MarkTechPost
·
20 hours ago
Thinking Machines Lab released Inkling, a 975-billion-parameter open-weights multimodal mixture-of-experts model with 41 billion active parameters and a 1-million-token context window. The model was trained on 45 trillion tokens of text, images, audio, and video, and features a controllable thinking-effort mechanism that allows users to trade inference cost for performance, achieving Terminal Bench 2.1 parity with Nemotron 3 Ultra using one-third the tokens. Inkling is available on Hugging Face and hosted platforms, enabling deployment of cost-tuned agentic systems and multimodal applications across voice, vision, and text inputs.
NVIDIA
·
21 hours ago
NVIDIA introduced the Jetson T3000 and T2000 modules based on its Thor architecture to enable robotics and edge AI applications, with the T3000 delivering 865 FP4 teraflops of compute in a compact form factor half the size of the T5000. The T3000 combines a Blackwell GPU with 32GB memory and 273GB/s bandwidth, while the T2000 offers 400 FP4 teraflops with 16GB memory, available in Q1 2027. Companies like Boston Dynamics and Amazon Robotics can now deploy foundation models on robots more efficiently, with customers like Agile Robots achieving up to 15GB memory savings through software optimization.
VentureBeat AI
·
22 hours ago
Anthropic's Claude platform dominates enterprise agent deployments with 40% of organizations choosing it as their primary orchestration platform, more than double any competitor. However, 71% of enterprises report that a quarter or fewer of their deployed "agents" are true multi-step orchestrated workflows rather than single-prompt chatbot wrappers, despite defining success by reliable multi-step execution. Enterprises plan to address this gap by building hybrid control planes that combine provider-native platforms with external orchestration tools they own, driven by fears of vendor lock-in.
MarkTechPost
·
23 hours ago
A German research consortium released Soofi S 30B-A3B, an open-source foundation model combining Mamba and Transformer architectures optimized for German and English. The model contains 31.6 billion parameters but activates only 3.2 billion per token, trained on 26.68 trillion tokens with German representing 15.32% of high-quality training data. Soofi S achieved the highest scores among fully open base models tested, scoring 70.1% on English benchmarks and 79.1% on German benchmarks, making it suitable for German document processing, bilingual code assistance, and long-context serving tasks.
TechCrunch AI
·
1 day ago
OpenAI released a $230 light-up keyboard called the Codex Micro, designed to control AI coding agents through customizable keys, status lights, and a dial for adjusting computational reasoning levels. The keyboard is a limited-run collaboration with specialty keyboard designer Work Louder and connects to ChatGPT's desktop app. OpenAI is simultaneously developing a separate screenless smart speaker with moving mechanical parts, designed by former Apple engineers, which is still in development and unfinished.
Wired AI
·
1 day ago
Researchers at Meta, Stanford, and other institutions created the EgoBabyVLM Challenge, a test that measures how well vision language models can learn from approximately one thousand hours of egocentric video recorded from cameras worn by infants. Current cutting-edge AI models fail substantially on this benchmark, struggling to extract meaning from the messy, realistic footage that babies process efficiently. The findings suggest that designing AI systems with learning mechanisms inspired by infant brains—such as better attention mechanisms and social cue interpretation—could create more efficient models that learn from less data and require less energy.
The New Stack
·
1 day ago
Google released GKE Agent Sandbox to general availability in May 2026 and introduced Agent Substrate, a separate scheduler designed for AI agents because Kubernetes was built for long-running services rather than the bursty, mostly-idle workloads that agents represent. Agent Substrate achieves 30x or more oversubscription with sub-second activation by snapshotting idle sessions to storage and multiplexing them onto pre-warmed worker pods, while Agent Sandbox provides kernel isolation via gVisor for running untrusted model-generated code at 300 sandboxes per second with 90 percent finishing in 200 milliseconds. The emergence of this dedicated agent runtime as a fourth compute offering alongside virtual machines, containers, and serverless indicates that the infrastructure layer for managing agents is consolidating into specialized systems separate from general container orchestration.
AWS Machine Learning
·
1 day ago
Built Technologies deployed an AI-powered document processing system on AWS Bedrock and the Intelligent Document Processing Accelerator to automate analysis of real estate finance documents. The system reduces workflows that previously took days to minutes and supports over 250 document types with over 95 percent accuracy for production use. The solution serves as a foundation for agentic products across the real estate finance lifecycle, from draw review to loan analysis to compliance workflows.
AWS Machine Learning
·
1 day ago
Amazon Bedrock now integrates computer vision, AI agents, and the Model Context Protocol to create a unified system where visual information can be captured, understood, and acted upon through a single interface. The solution combines Amazon Rekognition for object detection, Amazon Nova for video analysis, and Claude models for image interpretation, with support for images up to 200 MB and video formats including MP4, AVI, and MOV. This architecture eliminates the need to manage separate integrations between perception, decision-making, and action systems, making visual AI capabilities more accessible to developers building applications on AWS.
AWS Machine Learning
·
1 day ago
AWS published a guide for monitoring SageMaker Pipelines across multiple AWS accounts and regions using CloudWatch custom dashboards with a serverless hub-and-spoke architecture. The solution uses EventBridge, Lambda, and DynamoDB to collect pipeline execution data in real-time from source accounts and display unified status, timing, and step details on a centralized dashboard. Organizations can reduce operational overhead from manually switching between accounts and regions, and extend the monitoring to other ML workloads like Step Functions or Batch jobs.
Google Research
·
1 day ago
Researchers studied how diffusion models generate novel images and data beyond their training examples, finding that the models' creativity stems from neural network regularization effects that blur the learned score function. In a one-dimensional example with two training points at +1 and -1, weight decay causes the neural network to learn a smoother approximation of the score function, allowing generated particles to settle between the training points rather than copying them exactly. This score smoothing mechanism enables diffusion models to balance realism with novelty by interpolating between training data points while maintaining the quality of high-dimensional data manifolds.
Wired AI
·
1 day ago
Thinking Machines Lab, founded by OpenAI exiles including former CTO Mira Murati, released Inkling, an open-weight AI model trained on audio, video, and text inputs. Inkling contains 975 billion parameters and performs comparably to leading Chinese open-weight models while being cheaper to run than closed alternatives. The release positions Thinking Machines as a competitor in the AI market and supports its vision for decentralized AI development outside the control of a few dominant companies.
TechCrunch AI
·
1 day ago
Thinking Machines Lab released Inkling, an open-weight AI model with 975 billion total parameters that developers can download and modify, departing from the closed models sold by OpenAI, Anthropic, and Google. The model uses about 41 billion parameters per task and requires a third as many tokens as Nvidia's Nemotron 3 Ultra to achieve equivalent coding performance, according to the company's benchmarks. The startup is positioning Inkling as a customizable foundation for enterprises to fine-tune through its Tinker platform rather than a finished product, betting that organization-specific AI will outperform general-purpose models.
The New Stack
·
1 day ago
The Linux Foundation formally launched the X402 Foundation in April 2026 to standardize internet-native payments for transactions between AI agents, APIs, and applications, joining two earlier governance bodies focused on token costs and safety verification. Over the past 30 days, X402 has processed more than 75 million transactions worth $24.24 million across 94,060 buyers and 22,000 sellers, with major companies including AWS, Cloudflare, Google, and Stripe already implementing the protocol. The standardized payment framework enables AI agents to access services and APIs without pre-registration or account setup, addressing a critical need for autonomous agents operating across the web at scale.
The New Stack
·
1 day ago
Atlassian launched AI-powered features for Jira including a Coding Agent that converts work items into pull requests and integrates with Claude Code, Cursor, and GitHub Copilot to make the platform more appealing to developers. Internal testing showed a 36% reduction in PR cycle time, 44% boost in agent task completion efficiency, and 48% drop in token consumption among Atlassian's 6,000 engineers. The updates aim to keep developers in Jira by automating routine tasks and reducing context switching, though the company acknowledges challenges remain in code review bottlenecks and has set no roadmap beyond the next quarter.
The Verge
·
1 day ago
xAI is suing Terry Wayne Harwood, a South Carolina man allegedly arrested in February for using Grok to generate and distribute child sexual abuse material. Harwood faces eight felony charges related to possessing and distributing CSAM, with xAI claiming at least some of the images were generated or altered using their chatbot. The lawsuit asserts Harwood intentionally circumvented Grok's safeguards, potentially prompting xAI and other AI providers to implement stricter content moderation.
Hugging Face Blog
·
1 day ago
Shippy, an AI agent for maritime domain awareness, was built with a three-part architecture—system prompt ("soul"), task-specific instructions ("skills"), and configuration settings—designed to prioritize reliability over raw model capability in high-stakes operational decisions. The system includes a purpose-built CLI layer that standardizes API calls to prevent subtle bugs, isolated user sandboxes running on Kubernetes for data privacy, and a custom evaluation framework that scores the entire agent against weighted criteria on live data rather than static benchmarks. This approach enables Shippy to serve 70+ countries while maintaining data isolation between users and prevents deployment of versions that regress on guardrails or accuracy thresholds.
Hugging Face Blog
·
1 day ago
A team building model routing for AI agents found that selecting which model to use for each task is not simply a classification problem but requires optimizing across multiple system constraints simultaneously. On the AppWorld Test Challenge with 417 tasks, Claude Sonnet cost $79 total while GPT-4.1 cost $155 despite lower per-token pricing, because Sonnet's cache-read pricing was more efficient for agent workloads that reuse context. Effective routing must balance cost, latency, accuracy, and compliance requirements together rather than treating task difficulty or model speed as isolated variables.
MIT Technology Review AI
·
1 day ago
OpenAI built GPT-Red, an AI system that automatically finds vulnerabilities in its models by simulating cyberattacks, particularly prompt injection attacks where hackers embed hidden instructions to manipulate LLM behavior. In testing against GPT-5.6, fewer than 23% of GPT-Red's strongest attacks succeeded compared to over 90% against the earlier GPT-5 released in August last year. The system supplements human red-teaming efforts and enables OpenAI to identify new attack types before deployment, though it remains limited in handling multi-turn conversations and image-based attacks.
TechCrunch AI
·
1 day ago
A hacker breached Suno, an AI music generator, in November 2025 using a supply chain attack to access employee credentials and source code. The leaked code allegedly shows Suno scraped decades of audio from YouTube Music, Deezer, Genius, stock music libraries, and podcast RSS feeds to train its AI system. Record labels suing Suno argue the scraping violates the Digital Millennium Copyright Act and YouTube's terms of service, though Suno claims its training on publicly available content falls under fair use protections.
TechCrunch AI
·
1 day ago
Whatnot acquired Shaped, a machine learning company specializing in real-time recommendations and search systems, to improve product discovery as its livestream shopping marketplace expands. The company has reduced recommendation latency from roughly a day to minutes over six years and processes more than 500,000 hours of live video weekly. The integration of Shaped's technology is expected to push recommendations closer to real time, addressing live commerce's core challenge of matching shifting buyer demand with constantly changing inventory and auction schedules.
TechCrunch AI
·
1 day ago
Microsoft released patches for 570 security vulnerabilities across its product lines, attributing the spike to AI-assisted discovery of previously hidden code flaws. The batch included at least two zero-day exploits actively being used against Windows Server and SharePoint, with one allowing privilege escalation to system administrator level. Microsoft indicated that monthly patch volumes will remain elevated as AI tools continue to identify dormant vulnerabilities in legacy code.
The New Stack
·
1 day ago
OpenAI launched the Codex Micro, a $230 programmable macropad developed with Work Louder that includes six agent keys that light up to show the status of AI agents. The device features customizable keys, a joystick, dial, and push-to-talk functionality, with keycaps that can be remapped for use with other applications. The product targets Codex users, which OpenAI says will reach 9 million, and allows them to control agentic coding tasks through dedicated hardware.
Ars Technica
·
1 day ago
OpenAI released its first branded hardware product, the $230 Codex Micro, a custom mini-keyboard with color-coded keys for monitoring AI agents. The device features six frosted keys that display status updates—white for idle, blue for processing, green for completion, amber for requiring feedback, and red for errors—across up to six concurrent Codex threads. Users can tap the illuminated keys to quickly switch between different AI agent tasks without manually searching for windows on-screen.
The Verge
·
1 day ago
Fountain 0 announced an AI-generated film adaptation of The Odyssey called Odysseus: The Fall, capitalizing on interest from Christopher Nolan's concurrent theatrical release of the same source material. Nolan's film is projected to earn $80-$100 million in its opening weekend. The studio's AI-made version represents a shift toward low-cost synthetic film production as a direct-to-consumer alternative to traditional studio releases.
404 Media
·
1 day ago
This podcast episode covers three stories: the proliferation of ChatGPT-related flyers in public spaces, an incident where Waymo's autonomous vehicle reported teenagers to police, and a reviewer's experience purchasing a $3,000 electrostimulation fitness suit. The most concrete detail is the $3,000 price of the fitness suit mentioned in one segment. The episode highlights emerging issues around AI product marketing saturation and autonomous vehicle behavior in law enforcement contexts.
IEEE Spectrum AI
·
1 day ago
Researchers at MIT unearthed ELIZA's original source code from archives, revealing that the 1960s chatbot was not a simple pattern-matching program but a sophisticated platform capable of adopting multiple personas beyond its famous Doctor therapist script. The source code, published in January 1966 in Communications of the ACM, showed ELIZA could assume roles discussing math, poetry, color, and other topics through different scripts like Neweng for weather small talk. The discovery demonstrates that Weizenbaum's technical innovations were more advanced than previously documented and that design decisions about language and interaction continue to influence modern AI development.
TechCrunch AI
·
1 day ago
Apple's generative AI service has received regulatory approval to launch in China through a partnership integrating Alibaba's Qwen AI model into iOS, iPadOS, macOS, and visionOS. Apple's Greater China sales reached $20.5 billion in the second quarter, representing 28% growth year-over-year. The deal enables Apple Intelligence features, which debuted in 2024 globally, to finally become available to Chinese users after previous negotiations with Baidu, DeepSeek, and ByteDance fell through.
Sifted
·
1 day ago
Yann LeCun launched a new AI venture capital firm called Extelligence Invest and quit the same day, creating confusion in Europe's tech sector. LeCun continues to lead AMI Labs, which raised $1 billion in funding this year, and serves as an advisor to Hiro Capital. The reasons for his sudden departure from Extelligence remain unclear, though he maintains multiple other commitments in AI investing and research.
TechCrunch AI
·
1 day ago
Ode with Anthropic, a joint venture backed by Anthropic, Blackstone, Hellman & Friedman, and Goldman Sachs, was formed by acquiring Fractional AI to deploy engineers directly into enterprise companies for AI implementation. The venture's founders Chris Taylor and Eddie Siegel argue that most enterprise AI pilots fail to reach production environments. The model positions AI-native services as a major emerging category by embedding technical staff within client organizations rather than relying on traditional consulting approaches.
The Verge
·
1 day ago
Google's Pixel 11 phones will feature a glowing, color-shifting orb on the camera bar, as shown in a teaser video on Google's store page. The feature is expected to launch on August 12. The orb may serve as a physical indicator for Gemini AI or other phone functions, though Google has not yet disclosed its purpose.
Simon Willison
·
1 day ago
A researcher discovered that Claude's web_fetch tool could be exploited through nested links on honeypot websites to extract user data, bypassing Anthropic's safeguards against data exfiltration attacks. The attack successfully retrieved a user's name, home city, and employer by tricking the model into following letter-by-letter URL sequences disguised as authentication prompts. Anthropic subsequently patched the vulnerability by removing web_fetch's ability to navigate links embedded within previously fetched pages.
Sifted
·
1 day ago
Humanoid, a London-based robotics startup, is seeking $200 million in funding from Saudi Aramco at a valuation of around $1 billion. The funding round would value the company at $1bn based on a $200m investment. This investment would establish Humanoid as a major player in the robotics sector backed by significant corporate capital.
The New Stack
·
1 day ago
Anaconda acquired Kilo, an open-source coding agent that lets developers switch between AI model providers without vendor lock-in. Kilo routes approximately 10 trillion tokens monthly from 3 million developers across 500-plus models from 60+ providers. Anaconda plans to integrate Kilo into its governance platform over the next 12 months, allowing enterprises to manage AI spending and policies across development and production without being tied to a single model provider.
The New Stack
·
1 day ago
Meta is building a cloud business to sell excess GPU capacity through its Meta Compute initiative, choosing between offering hosted AI models or raw compute rental, while the shoe company Allbirds pivoted to selling GPU-as-a-Service after a $39 million sale of its footwear brand. Meta reported the compute business on July 1, and CoreWeave and Nebius stocks dropped double digits after the announcement, with Nebius holding a $27 billion contract with Meta. As GPU supply fragments across multiple providers including hyperscalers, neoclouds, and accidental clouds, enterprises will gain leverage through abstraction layers that treat suppliers as interchangeable provisioning targets rather than betting on single vendors.
404 Media
·
1 day ago
A hacker breached Suno AI and revealed that the music generation tool scraped millions of songs from YouTube Music, Deezer, Genius, Pond5, Jamendo, Freesound, and other sources to train its model. The leaked source code shows Suno ingested over 2 million music clips from YouTube Music alone and at least 790,000 hours of audio across all sources combined. The disclosure provides concrete evidence supporting record industry lawsuits against Suno for training on copyrighted material without permission.
The Verge
·
1 day ago
Suno scraped millions of songs and lyrics from YouTube Music, Deezer, and Genius to train its AI music generator, according to data exposed in a hacking incident. The training dataset included content from multiple platforms that Suno had previously kept secret from public disclosure. This revelation strengthens ongoing lawsuits against Suno, including a case brought by the Recording Industry Association of America alleging copyright infringement in the model's training.
The New Stack
·
1 day ago
OpenAI's Codex and ChatGPT Work combined reached 8 million active users following the July 9 launch of GPT-5.6, growing from 5 million in early June. The rapid user surge from 7 million to 8 million occurred between July 12 and July 15, exposing scaling issues that forced OpenAI to reduce the context window from 372,000 to 272,000 tokens and temporarily lift usage caps. The consolidation of Codex into a unified desktop app with integrated tools like Slack and Google Drive marks a shift toward positioning AI as a workspace layer rather than a standalone product.
TechCrunch AI
·
1 day ago
Anthropic and Blackstone launched Ode, a $1.5 billion AI implementation company, to deploy engineers who help enterprises integrate AI systems into their operations, following a similar move by OpenAI. Ode currently employs 100 engineers and acquired AI services startup Fractional AI as its foundation, operating on a "Claude-first" principle while remaining open to other AI models. The venture assumes that consulting services for AI deployment—not just model development—represents the next trillion-dollar business opportunity, though it faces competition from OpenAI's Deployment Company and consulting firms like Deloitte and Accenture.
IEEE Spectrum AI
·
1 day ago
South Korean researchers at Seoul National University and Hanyang University developed Generative SNUPI, a generative AI model that designs DNA origami structures by automatically generating DNA sequences that fold into user-requested shapes. The model uses diffusion techniques to determine how DNA strands should be sequenced so molecular forces cause them to self-assemble into forms like dogs, stars, and the Mona Lisa, eliminating the tedious manual design work traditionally required. This capability could accelerate DNA origami applications in drug delivery and immunotherapy by removing a major bottleneck in the design process.
Tech.eu
·
1 day ago
Monumental, a construction robotics company, raised $32 million in Series B funding to expand its fleet of autonomous bricklaying robots powered by its AI platform Atrium. The company operates over 150 robots across the Netherlands and UK that have helped construct more than 100 homes, with nearly half completed in the past three months. The funding will support expansion of its engineering team, increased robot deployment across Europe, and preparation for US market entry.
TechCrunch AI
·
1 day ago
Rime, a voice AI startup founded in 2022, raised $24 million in Series A funding led by M13 Ventures to develop conversational AI systems for enterprise customer service calls. The company operates a recording studio in San Francisco to train models on proprietary conversational data and now focuses on speech-to-speech models with latency reductions rather than orchestrating multiple separate models. Rime plans to expand its 35-person team and aims to capture enterprise contracts from companies that currently rely on legacy interactive voice response systems.
TechCrunch AI
·
1 day ago
Reelful, an iOS app founded by former Snapchat engineer Kate Deyneka, automatically generates short-form social media videos by converting photos and video clips into polished TikTok- and Instagram Reels-style content. The app costs $15 for five videos as a one-time purchase, or ranges from $25 to $100 monthly for subscription plans offering 10 to 60 videos. Users can now create social content without manually selecting clips, recording voiceovers, or editing, though the app is currently iOS-only with Android and web versions planned for the future.
Wired AI
·
1 day ago
Hemispheric, founded by an Apple FaceID co-inventor, raised $52 million to develop AI models that diagnose cognitive disorders by analyzing brain electrical activity from EEG headsets. The startup trained its deep-learning model on 250,000 hours of brain data collected from 100,000 people and plans to submit its first product for PTSD diagnosis to the FDA in early 2025, with public availability targeted for late 2027. The approach aims to replace subjective questionnaires with objective AI-assisted diagnosis that could eventually function as simply and cheaply as a blood test.
OpenAI Blog
·
1 day ago
OpenAI has proposed a "reverse federalism" model where state-level AI regulations would inform and shape federal policy rather than the other way around. The approach suggests allowing states to pioneer AI safety rules that could eventually form the basis of a coherent national standard. This strategy aims to balance localized governance experimentation with the development of consistent safeguards across the country.
TechCrunch AI
·
1 day ago
Vint Cerf, who left Google after 20 years, is now advising Innovation Labs to develop open standards for identifying and auditing AI agents operating across the internet. The organization proposes DNSid, a system using domain names and cryptographic proofs to establish agent identities and track their registration over time, and is currently testing it with unnamed hyperscalers. Widespread adoption of agent identification standards will depend on whether they can interoperate across different systems, similar to how TCP/IP became the internet standard through user pressure.
TechCrunch AI
·
1 day ago
Emergent, an Indian AI coding startup, raised $130 million in Series C funding at a $1.5 billion valuation, achieving unicorn status five months after its previous $70 million raise. The company reported $120 million in annual run-rate revenue with over 200,000 paying customers, up 70% in the last four months. Emergent plans to accelerate product development, expand its San Francisco office by 30 to 40 employees, and open a European office to support growth across non-technical users building business applications.
TLDR Dev
·
1 day ago
A researcher proposes 'Guardian Angels'—personalized LLMs that emulate individual users' values and preferences to amplify productivity and provide security against AI-powered attacks, arguing current chatbots are fundamentally misaligned with users and designed for replacement rather than augmentation. The approach combines dynamic evaluation, active learning, and continuous user feedback to create AI agents that remain under human control and learn user-specific patterns, addressing the principal-agent problem by unifying principal and agent goals. This shifts work from 'what and how to do things' to 'what is worth doing,' enabling users to deploy multiple specialized agents for productivity and security while maintaining strategic oversight.
TLDR Dev
·
1 day ago
Generative AI has dramatically lowered the barrier to copying game ideas, as demonstrated when developer Freya Holmér's rotating Tetris prototype was cloned by multiple people within days using AI code generation, each taking roughly a day to create. The ease of AI-assisted cloning amplifies a pre-existing problem of game storefronts being flooded with low-effort knockoffs by companies like Voodoo and Midnight Works, which have built billion-dollar businesses around copying popular games. Developers now face increased anxiety about sharing work publicly, and the devaluation of skill and effort discourages original creativity as clone makers can bypass the knowledge and execution that traditionally made copying expensive and difficult.
TLDR Dev
·
1 day ago
Demis Hassabis, CEO of Google DeepMind, stated that Artificial General Intelligence is imminent and poses both significant benefits and risks to humanity. The article provides no specific timelines, benchmarks, or concrete details about AGI development or capabilities. Hassabis emphasizes the need for careful management of AGI risks, though the article does not specify what regulatory or safety measures should be implemented.
TLDR Dev
·
1 day ago
Vulnerability discovery rates are compounding with AI assistance—CVE publications exceeded 40,000 in 2025 with 2026 projections reaching 60,000 to 100,000—while remediation efforts show diminishing returns as organizations closing 6.5 times more tickets experienced worse critical vulnerability resolution rates. Network microsegmentation adoption remains below 20% in enterprises despite being foundational to limiting breach damage, a gap now addressable through AI-assisted policy recommendations that baseline application communication patterns. As 82% of intrusions use valid credentials through lateral movement, organizations must architect systems for containment rather than relying solely on patching, with AI enabling previously impractical approaches to microsegmentation through automated discovery and policy synthesis.
TLDR Dev
·
1 day ago
OpenAI's ChatGPT web app serves approximately 1 billion users through a carefully optimized architecture that migrated from Next.js to Remix and now uses React Router 7 with server-side rendering. The initial HTML payload is 84 KB compressed with a time to first byte of 50-65ms, delivering a 548-node DOM that prioritizes the chat input as the primary interaction point. The engineering approach uses off-the-shelf libraries rather than custom frameworks, with the server controlling client behavior through prefetch decisions and streaming data to enable rapid interaction without requiring user authentication.
TheSequence
·
1 day ago
OpenAI audited SWE-Bench Pro, a coding evaluation benchmark, and found that approximately 30 percent of its 731 public tasks contain defects such as rejecting correct solutions or accepting incomplete ones. OpenAI's agent-assisted review labeled 27.4 percent of tasks as defective while independent software engineers identified 34.1 percent as problematic. OpenAI withdrew its earlier recommendation that the field adopt SWE-Bench Pro as a standard evaluation tool due to these validity issues.
Ars Technica
·
1 day ago
SpaceX plans to deploy a constellation of 1 million satellites in orbit to serve as data centers, shifting its business focus away from rockets toward hosting artificial intelligence computing infrastructure. The company aims to power up to 100 million frontier-class GPUs across these orbital facilities, with individual AI1 satellites designed to generate 120 GW of power. This orbital data center approach would allow SpaceX to offer computing services directly from space rather than relying on ground-based infrastructure.
NVIDIA
·
1 day ago
NVIDIA and Japanese partners are deploying AI technology across healthcare, robotics, vision systems, and financial services, with pharma companies using NVIDIA BioNeMo for drug discovery, medical device makers building surgical robots, and banks like Mizuho planning large on-premises AI factories. Mizuho is building what is expected to be Japan's largest on-premises AI factory for the financial industry starting with DGX B200 systems, while companies like Biomy achieved 90% faster spatial transcriptomics analysis using NVIDIA single-cell RAPIDS. These deployments represent a shift from experimental AI to production infrastructure across Japanese industry, with NVIDIA technology integrated into core workflows for drug discovery, medical imaging, industrial inspection, and financial services.
TLDR
·
1 day ago
Companies are deploying AI agents without proper management frameworks, creating wasteful "loops" where poorly-trained employees use AI inefficiently to compensate for unclear processes. The vast majority of token spend produces no value, mirroring how 80% of traditional employees often perform no meaningful business function. The key to unlocking AI value lies in building rigorous evaluation systems and AI transformation services that encode each firm's specific processes into measurable outputs, similar to how coding became the only breakout AI use case because code either runs or fails.
TLDR
·
1 day ago
Apple sued OpenAI alleging that former employees, including senior hardware engineer Chang Liu and vice president Tang Tan, misappropriated trade secrets and recruited others to do the same. The lawsuit focuses on over 400 Apple employees who have moved to OpenAI and names specific allegations including use of Apple codenames to solicit confidential information and a proprietary "trade secret metal-finishing technique." The case could delay OpenAI's hardware launch timeline and force it to redesign products to avoid Apple-controlled intellectual property, while discovery proceedings may expose details about OpenAI's unreleased hardware plans.
TLDR
·
1 day ago
As AI inference costs decline, companies can process larger volumes of requests without hitting capacity constraints, shifting the focus from resource scarcity to deliberate prioritization of what to build. Organizations using this approach operate with smaller engineering teams while increasing API token consumption, enabling faster feature deployment. This allows product teams to respond directly to customer feedback with merged code changes rather than being bottlenecked by engineering capacity.
TLDR
·
1 day ago
A software engineer argues that AI coding assistants may undermine the shared understanding that coordinates large-scale software projects, similar to how the Tower of Babel collapsed when people lost common language. Unlike the biblical story where communication breakdown halts construction, AI agents can continue making changes to codebases without developers needing to coordinate or comprehend each other's modifications. The result is that systems grow increasingly incoherent without the immediate failure that would signal the problem, making architectural decay invisible until critical.
TLDR
·
1 day ago
Apple is in talks with PrismML, a startup that compresses artificial intelligence models to run on smartphones, after the company publicly released a version of Alibaba's Qwen model optimized for iPhones. PrismML reduced the Qwen model from 54 GB to under 4 GB, enabling it to run on iPhone 15 and newer devices with all 27 billion parameters intact. Running larger AI models directly on iPhones would reduce latency, lower cloud costs, improve privacy, and enable features to work without internet connectivity.
TLDR
·
1 day ago
OpenAI is developing its first consumer device, a mobile smart speaker without a screen designed to function as an AI companion for the home. The device will perform tasks including smart-home control, media playback, and answering questions through voice interaction. Users will be able to carry the speaker around their homes and interact with it as a conversational companion.
.jpg)
Wired AI
·
1 day ago
A customer's $2,000 ebike was marked as delivered and signed for by someone else, and when he attempted to track it down, he encountered AI chatbots at FedEx, the bike retailer, his bank, and even his local police department that consistently prevented him from reaching human representatives. Across all customer service interactions, 59 percent of consumers reported frustration with AI agents, while 85 percent preferred speaking with humans, and Verizon's CEO stated that AI will likely replace a large percentage of the company's customer service workforce. Corporations are deploying AI in customer service at the expense of human employees, sometimes intentionally using chatbots as a tactic called sludge to discourage customers from seeking resolution.
Rest of World
·
1 day ago
Venezuelan programmers and diaspora developers used AI tools like Claude and facial recognition software to rapidly build websites for missing-person reports, aid coordination, and damage assessment after earthquakes killed over 3,800 people, filling gaps left by slow government response. Key sites included Desaparecidos Terremoto Venezuela which received 30,000 reports in two days, Ayuda en Camino built in four hours, and Somos Acompañamiento with over 84,000 registrations. The civilian-led AI platforms became the primary source of disaster information, though experts warn they cannot replace government responsibility and must include stronger data privacy protections for sensitive biometric information.
OpenAI Blog
·
1 day ago
OpenAI developed GPT-Red, an automated system that uses self-play to identify vulnerabilities in AI models and improve their robustness against adversarial attacks. The system iteratively generates adversarial prompts and defensive improvements in a feedback loop, with evaluations showing measurable increases in resistance to prompt injection attempts. This approach allows AI developers to systematically strengthen defenses without manual red teaming, reducing the time required to identify and patch safety vulnerabilities.
The Neuron
·
1 day ago
Samsara deployed AI agents that monitor trucks and dispatch centers to predict accidents before they occur, expanding from reactive accident detection to proactive safety measures. The company's AI dash cams now analyze driving behavior and road conditions in real time to issue warnings. This shift enables fleet operators to intervene before incidents happen, reducing crashes and associated costs.
The Neuron
·
1 day ago
ClawTeams is a platform that orchestrates multiple AI specialists through messaging apps like Slack and Teams to break down tasks, execute work in parallel, and deliver finished outputs after quality checks. The platform serves 12,000+ teams that have completed 1.2 million tasks, with users reporting an average of 68% time savings compared to doing work by hand. Users assign work by mentioning the AI lead in chat, which coordinates specialists across content, data analysis, compliance, and other domains without requiring users to switch tools or write detailed prompts.
The Neuron
·
1 day ago
Pazi launched a platform that creates AI teams for functions like DevOps, sales, and SEO, operating continuously within Slack. The system integrates with GitHub, Linear, and Sentry to automate workflows across these tools. Organizations can now delegate routine tasks across multiple departments to AI agents that work around the clock in their existing Slack workspace.
The Neuron
·
1 day ago
Framer released AI agents that operate directly on its web design canvas, allowing designers to generate and refine layouts, manage content management systems, and write code while maintaining full control over changes. The CMS agent can import and organize content (demonstrated with 47 WordPress blog entries mapped automatically) and connect to external tools including Slack, GitHub, and Claude Code for publishing and content updates from any platform. This enables designers and developers to manage site design, content, and code through a unified interface rather than switching between separate applications.
The Neuron
·
1 day ago
OpenAI restored access to ChatGPT on WhatsApp in the European Economic Area after Meta had removed it in January 2026, following an EU Commission order in June 2026 requiring Meta to reinstate competing AI assistants. Users in the 30-country region have been able to reach ChatGPT via the number +1-800-242-8478 since July 13, 2026, with support for text, images, and voice messages across multiple languages. The service allows optional account linking to sync chat history and context, though usage limits for unlinked accounts remain unclear.
The Neuron
·
1 day ago
Superhuman launched an improved auto-draft feature that generates email replies matching the user's tone based on previous conversations, with options to select from multiple variations. In beta testing, 60% of auto-generated drafts were sent without manual editing within one day. The feature learns from user behavior and adjusts suggestions over time, reducing errors like proposing unsuitable meeting times.
The Neuron
·
1 day ago
OpenAI's path to going public is encountering obstacles from multiple lawsuits, strained relations with Microsoft, and competitive pressure from Anthropic's growing valuation. The company's valuation has been set at $80 billion in recent funding rounds, while Anthropic has reached $60 billion, narrowing the gap between the two AI labs. These combined pressures may force OpenAI to reconsider its timeline for an IPO or the terms on which it eventually goes public.
The Neuron
·
1 day ago
Spotify launched a conversational AI assistant for Premium subscribers that allows users to request music and audio content through typed or spoken dialogue across iOS and Android devices. The feature is initially available in the U.S., Ireland, and Sweden for users 18 and older in English, with the company using a combination of its own AI technology and external models. Users can now have back-and-forth conversations with the app to discover new music, explore their listening history, and refine selections by asking follow-up questions or making requests like "play more upbeat songs."
The Neuron
·
1 day ago
Google DeepMind CEO Demis Hassabis published a manifesto proposing a U.S. national AI standards body to evaluate advanced models for security and safety risks before public release. The proposed independent board, modeled on financial industry regulator FINRA, would be funded by AI companies and become operational by the end of 2024. If adopted, such a watchdog would require developers to submit models for testing on cyber, biological, and deception capabilities before deployment.
The Neuron
·
1 day ago
xAI's Grok Build feature uploaded complete Git repositories to xAI's servers without developer consent, including files in .gitignore and unredacted credentials from environment files. The uploads contained full commit histories and API keys that developers had explicitly marked as private. xAI disabled the feature and committed to deleting all uploaded repositories.
Sifted
·
1 day ago
Scotland's University of Edinburgh, government programme Techscaler, and CodeBase have established the Venture Builder Incubator, a five-month programme helping academic founders commercialise deeptech research into startups. Since 2021, the incubator has supported 164 founders and businesses that have collectively raised £58 million in grants and investments. The expanded nationwide programme aims to retain Scottish talent and attract external investors by equipping researchers with commercial, sales, and storytelling skills needed to scale their innovations.
Wired AI
·
1 day ago
OpenAI employees have donated over $215,000 to Guardrails Alliance, a super PAC advocating for stricter AI regulations, in opposition to Leading the Future, a pro-industry PAC backed by $100 million from OpenAI president Greg Brockman. The largest individual donation came from research engineer Juan Felipe Cerón Uribe at $200,000, with Guardrails Alliance aiming to raise $15 million this election cycle compared to the $50 million Brockman and his wife committed to the opposing PAC. The donations highlight internal company tensions over AI policy direction, with some employees using personal funds to counter what they view as efforts by tech leaders to prevent AI regulation.
MarkTechPost
·
1 day ago
Google released LiteRT.js, a JavaScript binding that runs .tflite models directly in web browsers by compiling its native on-device inference runtime to WebAssembly. LiteRT.js achieves up to 3x faster performance than other web runtimes and 5–60x speedup over CPU execution when using GPU or NPU backends through WebGPU and WebNN APIs. Web applications can now execute computer vision and audio models locally without server infrastructure, inheriting performance optimizations previously available only on Android, iOS, and desktop platforms.
Tech.eu
·
1 day ago
Finland approved Bliq.ai to operate driverless vehicles on public roads immediately, with initial deployments including a safety driver onboard. Testing will begin in Helsinki with Bliq's AI-based Level 2 driving system combined with remote human supervision. The approval enables Bliq to validate its retrofit autonomous technology in European conditions and move toward broader driverless mobility deployment across the continent.
Tech.eu
·
1 day ago
Float, a revenue-based financing platform for European tech companies, raised €4.5 million in Series A funding led by CHAPTERS Group AG. The company has distributed over €100 million to 130 tech businesses since 2022 and plans to evolve into an AI-native financial platform combining lending with AI-powered financial management tools. The funding will enable Float to double its team, expand in the UK market, and add automated financial insights and expense management features to its core lending business.
Sifted
·
1 day ago
Neko Health, a preventative healthtech company founded by Hjalmar Nilsonne and Spotify's Daniel Ek, uses unflattering avatars of users' bodies alongside health scan results to motivate behavioral changes. The company raised $700 million in Series C funding and reports 80% yearly customer retention, with retention improving from the second to third scan as users engage with the platform's new features including AI-powered body composition analysis. Following avatar visualization, users reportedly adopt lifestyle changes such as improved diet and exercise habits, with the company expanding to New York after opening a second Stockholm clinic.
Sifted
·
1 day ago
Meticulous, a London-based startup founded by ex-Palantir and Dropbox engineers, raised $15m in Series A funding to help developers test and verify AI-generated code before deployment. The company has achieved 5x annual recurring revenue growth over the past year with customers including Notion, ElevenLabs, and Dropbox, and currently tests frontend code by simulating user flows across edge cases. The funding will support expansion into backend testing, marketing, and hiring to grow the team from 20 to 30-40 employees over the next year.
TechCrunch AI
·
1 day ago
Miles Wang, an OpenAI researcher specializing in AI for scientific discovery, is leaving to start a drug discovery startup with several other OpenAI researchers. The new company is in funding discussions with Lightspeed to raise $200 million at a $2 billion valuation. The startup would focus on developing AI models to identify new uses for existing drugs, potentially reducing time to market compared to discovering entirely new treatments.
Together AI
·
1 day ago
Together AI released updates to its GPU Clusters platform including passive health checks that monitor running workloads for failures like GPU bus drops and thermal throttling, auto node repair with human approval, and a rebuilt Slurm-on-Kubernetes stack addressing daemon crashes and process cleanup. The platform added operational features including a redesigned cluster overview dashboard showing health and utilization, external OIDC authentication for per-user Kubernetes access, and startup scripts for self-serve node customization. These changes reduce incident resolution time from hours to minutes and enable teams to manage clusters at scale without sharing admin credentials or performing manual node setup.
Together AI
·
1 day ago
Thinking Machines Lab released Inkling, a 975-billion-parameter mixture-of-experts model with 40B active parameters that accepts text, image, and audio inputs for multimodal reasoning tasks. Inkling is available on Together AI's inference platform starting today with a 1M token context window and adjustable inference effort settings. Developers can now access a unified multimodal model through a single API endpoint that supports reasoning, coding, forecasting, and agentic workflows without managing their own infrastructure.
Apple ML Research
·
1 day ago
Researchers addressed uncertainty quantification for LLM function-calling, which enables large language models to use external tools autonomously. The study focuses on measuring model confidence before executing function calls that could have irreversible consequences like financial transfers or data deletion. Better uncertainty estimates allow systems to defer high-risk decisions to human operators or alternative fallback mechanisms.
Apple ML Research
·
1 day ago
CLaRa is a framework that improves retrieval-augmented generation by compressing documents and jointly optimizing retrieval and generation in a shared continuous space. The method achieves state-of-the-art performance on QA benchmarks at a compression rate of 16, meaning documents are reduced to 1/16th their original length while maintaining answer quality. This unified approach allows gradients to flow through both retrieval and generation modules, aligning document relevance with answer quality during training.
Apple ML Research
·
1 day ago
Researchers propose FAE, a framework that adapts pre-trained visual encoders for image generation by using a single attention layer to convert high-dimensional features into low-dimensional latents suitable for generative models. On ImageNet 256×256, FAE achieves an FID score of 1.48 without classifier-free guidance after 800 epochs and 2.08 after 80 epochs. The approach enables simpler adaptation of pre-trained representations across different generative model families including diffusion models and normalizing flows.
Hugging Face Blog
·
1 day ago
Hume AI released Real World VoiceEQ, a benchmark that evaluates voice AI systems on human-quality interactions beyond traditional metrics like word error rates and latency. The benchmark drew from over 1 million human ratings across 40+ voice models and 60+ metrics spanning speech recognition, synthesis, and understanding tasks. The findings show no single voice model excels across all capabilities, and current systems struggle to interpret emotional cues and acoustic information that humans naturally recognize in conversation.
Hugging Face Blog
·
1 day ago
Thinking Machines released Inkling, an open-source multimodal language model with 1 trillion parameters that accepts images, text, and audio inputs natively. The model has 975 billion total parameters with 41 billion active at any time, supports a 1 million token context window, and was trained on 45 trillion tokens across multiple modalities. Inkling is available on Hugging Face with immediate support in transformers, SGLang, vLLM, and llama.cpp, enabling developers to build multimodal reasoning applications through fine-tuning or remote inference.
