AI and tech notes

#ai#openai#chatgpt#agents

OpenAI is still working on a ChatGPT super app

OpenAI is still developing a broader ChatGPT “super app,” with a senior employee arguing internally that simple chat is no longer the end goal.

#ai#deepseek#enterprise#models

DeepSeek tops Ramp trending vendors as firms seek cheaper AI

DeepSeek led Ramp’s trending software vendors in June 2026, suggesting U.S. companies are actively looking for lower-cost AI services despite security concerns.

#ai#security#policy#surveillance

Survivor sues AI gun detection firm after missed weapon

A school shooting survivor has sued an AI gun detection company after its system allegedly failed to identify a weapon before an attack.

#ai#anthropic#productivity#infrastructure

Notion restores Anthropic access after service disruption

Notion restored access to Anthropic after a service disruption, highlighting how productivity tools now depend on third-party model providers.

#ai#apple#siri#consumer-ai

TechCrunch previews Apple Intelligence and Siri expectations for WWDC 2026

Apple’s WWDC 2026 is expected to focus heavily on Siri’s long-awaited revamp and broader Apple Intelligence updates.

#ai#research#safety#models

Sakana AI launches lab focused on recursive self-improving AI

Sakana AI has launched a research lab for recursive self-improvement, exploring AI systems that can iteratively improve themselves as an alternative to frontier labs’ compute race.

#ai#nvidia#gpu#personal-ai

NVIDIA brings RTX Spark push to South Korea’s PC bangs

It is promoting RTX Spark in South Korea with gaming partners, positioning new Windows PC hardware around personal AI agents and accelerated local computing.

#ai#data-centers#infrastructure#policy

Indiana data center fight intensifies after mayor comment

A proposed $2 billion data center in Shelbyville, Indiana, has become a sharper local controversy after the city’s mayor was recorded criticizing opponents’ yard signs.

#technology#data-engineering#aws#databases

AWS releases ExtendDB adapter for DynamoDB-compatible workloads

AWS has announced ExtendDB, an open source DynamoDB-compatible adapter that can connect existing DynamoDB-style applications to pluggable storage backends such as PostgreSQL.

#xai#anthropic#claude#training#distillation

xAI Reportedly Trained Its Coding Models on Claude Outputs for Months Before Being Cut Off

Elon Musk xAI allegedly used outputs from Anthropic Claude models to train its own coding AI for months before Anthropic detected the activity and blocked access, The Decoder reports.

#openai#security#prompt-injection#enterprise#lockdown-mode

OpenAI Launches Lockdown Mode to Protect Sensitive Data from Prompt Injection Attacks

OpenAI introduced Lockdown Mode, a new security feature that restricts how models handle external data to prevent prompt injection attacks that could expose sensitive information.

#multi-agent#simulation#open-source#small-models#hackathon

Hugging Face Hackathon Project Builds a Multi-Model Financial Drama Simulation on Small Models

A new Hugging Face hackathon project demonstrates five AI labs collaborating in a financial drama simulation, with each lab running a different small open-source model to model competitive dynamics.

#policy#government#white-house#personnel#regulation

Sriram Krishnan Steps Down as White House AI Advisor

Sriram Krishnan is leaving his position as White House AI advisor, TechCrunch reports, marking a significant departure from the administration AI policy team.

#meta#agents#pricing#products#hatch

Meta First Paid AI Product, Hatch, Could Cost Up to $200 Per Month

Meta is preparing to launch Hatch, its first paid AI agent product, with pricing that could reach $200 per month, marking a major shift in the company AI monetization strategy.

#infrastructure#data-centers#policy#community#energy

Giant AI Data Center Plan Cut in Half Following Community Protests

A controversial data center project has been scaled back by 50 percent after organized community opposition, highlighting growing tensions between AI infrastructure expansion and local concerns.

#agents#linkedin#enterprise#mcp#platform

LinkedIn Shares Its Approach to Platform Teams Enabling Multi-Agent AI Tools

LinkedIn engineering presented how the company platform teams are enabling MCP-based multi-agent AI tools across the organization, offering insights into enterprise-scale agent deployment.

#design#typography#generative-ai#culture#creative

AI Has Come for Serif Fonts—and Typography May Never Look the Same

WIRED explores how AI-generated typography is reshaping font design, with machine learning models now capable of creating functional serif typefaces that blur the line between human and algorithmic creativity.

#reinforcement-learning#ml-engineering#environments#best-practices#training

How to Stop Shipping Low-Quality RL Environments, With Concrete Examples

Latent Space published a practitioner guide on avoiding common pitfalls when building reinforcement learning environments for AI training, with specific examples of what makes an environment high or low quality.

#benchmarks#enterprise#agents#tools#evaluation

EVA-Bench Data 2.0 Expands to 3 Domains, 121 Tools, and 213 Evaluation Scenarios

ServiceNow Research released EVA-Bench Data 2.0, a major expansion of the enterprise AI agent benchmark covering CRM, HR, and IT domains with 121 tools and 213 real-world scenarios.

#aws#nvidia#sagemaker#models#deployment

NVIDIA Nemotron 3 Ultra Now Available on Amazon SageMaker JumpStart

NVIDIA Nemotron 3 Ultra, a large-scale language model, is now deployable through Amazon SageMaker JumpStart, giving AWS customers one-click access to the model for inference and fine-tuning.

#microsoft#edge-ai#developers#tools#foundry

Microsoft Foundry Local Brings Edge AI Development to On-Premises Environments

Microsoft launched Foundry Local, a new capability that lets developers build and test AI applications on local or on-premises hardware before deploying to the cloud.

#inference#costs#economics#industry#infrastructure

The Token Bill Comes Due: Inside the Scramble to Manage AI Runaway Inference Costs

TechCrunch examines how AI companies are grappling with soaring inference costs as models grow larger and usage scales, forcing a reckoning over token economics across the industry.

#safety#disinformation#evaluation#llm#research

Study Ranks LLMs by Their Ability to Resist Russian Propaganda Narratives

Researchers tested leading language models against Russian disinformation narratives and found significant differences in how well each model resists propagating propaganda when prompted.

#neuroscience#agi#funding#research#bezos

Jeff Bezos Is Funding an Ambitious Hunt for the Brain 'Core Algorithm'

Jeff Bezos is reportedly funding a scientific effort to uncover what researchers call the brain 'core algorithm'—a unified computational principle that could explain biological intelligence and inform AI development.

#openai#government#policy#regulation#national-security

OpenAI and Trump Administration Negotiating Government Stake in the AI Company

OpenAI and the Trump administration are in active negotiations over a potential US government equity stake in the AI startup, marking an unprecedented step in public-private AI governance.

#agents#coding#enterprise#dropbox#platform

Dropbox Built Nova, an Internal Platform for Running AI Coding Agents at Scale

Dropbox revealed Nova, an internal platform designed to orchestrate AI coding agents at scale across the company, handling hundreds of concurrent agent tasks for software engineering workflows.

#google#health#computer-vision#ml#on-device

Google Research Explores Passive Heart Health Monitoring Through Smartphone Cameras

Google Research demonstrated a technique for passive heart health monitoring using smartphone cameras, enabling continuous cardiovascular tracking without dedicated wearables.

#nvidia#safety#enterprise#content-moderation#multimodal

NVIDIA Nemotron 3.5 Content Safety Brings Customizable Multimodal Guardrails to Enterprise AI

NVIDIA released Nemotron 3.5 Content Safety, a customizable family of guardrail models for filtering harmful content across text and images in enterprise AI deployments.

#google#spacex#compute#infrastructure#cloud

Google to Pay SpaceX $920 Million Per Month for Compute—Over $11 Billion Annually

Google has signed a deal to pay SpaceX $920 million per month for compute capacity, totaling over $11 billion per year, as AI infrastructure demand reshapes cloud economics.

#spacex#openai#anthropic#sp500#markets

S&P 500 Blocks SpaceX Fast-Track Entry, Shutting Out OpenAI and Anthropic as Well

The S&P 500 index committee rejected SpaceX bid for fast-track inclusion, a decision that also blocks OpenAI and Anthropic from entering the benchmark index under current profitability rules.

#microsoft#ethics#agents#governance#ai-safety

Satya Nadella Publicly Rejects Proposal to Make Microsoft AI Agent Deliberately Addictive

Microsoft CEO Satya Nadella publicly shut down an internal proposal to design the company AI agent to be deliberately addictive, calling the idea antithetical to the company mission.

#openai#anthropic#investment#venture-capital#industry

OpenAI and Anthropic Share Overlapping Investors Who Are Not Picking Sides

Despite being fierce competitors, OpenAI and Anthropic share a growing number of mutual investors who are hedging their bets across both AI labs rather than choosing one over the other.

#ollama#gguf#local-ai#open-source#inference

Ollama Improves GGUF Model Performance and Broadens Hardware Compatibility

Ollama released updates improving GGUF model loading performance, memory efficiency, and hardware support across a wider range of GPU and CPU configurations.

#nvidia#infrastructure#robotics#korea#vera-rubin

NVIDIA and South Korea Partner to Build Sovereign AI Infrastructure

NVIDIA CEO Jensen Huang visited Seoul to meet with South Korean partners building sovereign AI infrastructure, with a focus on robotics, memory manufacturing, and the Vera Rubin platform entering full production.

#agents#multi-agent#open-source#simulation#small-models

Thousand Token Wood: A Multi-Agent Economy Running on a 3B Parameter Model

A hackathon project demonstrates a multi-agent economic simulation where autonomous agents trade, build, and cooperate, all powered by a compact 3-billion-parameter language model.

#google#rag#agents#enterprise#gemini

Google Introduces Agentic RAG for Dependable Enterprise AI Responses

Google Research detailed a new Agentic RAG approach for the Gemini Enterprise Agent Platform, designed to make AI-generated answers more reliable and auditable in business settings.

#gemini#google-deepmind#ai-models#google-io

Google DeepMind Releases Gemini 3.5 Series Combining Frontier Intelligence with Action Capabilities

At Google I/O 2026, Google DeepMind unveiled Gemini 3.5, a new series of AI models that integrate frontier-level reasoning with the ability to take actions on behalf of users.