Wednesday, July 15, 2026

30 stories
OpenAIOther

Optimizing AI Spending for Agentic Workflows

OpenAI provides guidance on how businesses can track the value of their AI investments. The approach focuses on increasing efficiency and expanding high-impact processes by measuring actual work output against cost.

Read the source →
Google AIProduct

Google Images Marks 25th Anniversary

Google is commemorating a quarter-century of its visual search tool. The milestone highlights past achievements and introduces updated methods for discovering and generating imagery.

Read the source →
NVIDIAModel

Customizing AI for Corporate Needs

Nemotron Labs emphasizes the importance of open models for organizations seeking greater control and trust. This approach allows businesses and nations to tailor AI to their specific industry knowledge and operational workflows.

Read the source →
AWS Machine LearningProduct

Thrad.ai Automates Lead Generation via Amazon Bedrock

Thrad.ai has implemented a multi-agent system to handle everything from finding prospects to writing tailored emails. The project evaluates the performance of Swarm and Graph orchestration methods based on cost, speed, and output quality.

Read the source →
TechCrunchFunding

OpenAI Scientist Planning High-Value Biotech Venture

Miles Wang is reportedly negotiating the launch of an AI-driven drug discovery company. The startup's estimated valuation could reach $2 billion.

Read the source →
The VergeProduct

OpenAI Developing Voice-Based Hardware

OpenAI is reportedly creating a screenless smart speaker powered by ChatGPT. The device will utilize sensors and a camera to perceive the user's surroundings.

Read the source →
WiredResearch

The Legacy of ELIZA

Joseph Weizenbaum developed an early chatbot at MIT during the 1960s. This project demonstrated why users tend to reveal personal secrets to conversational AI.

Read the source →
The DecoderProduct

OpenAI Develops Screenless AI Speaker

OpenAI is creating a portable AI companion featuring sensors and moving parts to mimic life. A legal dispute between Apple and a top OpenAI executive may push the expected 2027 release date back.

Read the source →
Ars TechnicaPolicy

Meta Faces Allegations of AI-Driven Layoffs

A lawsuit alleges that Meta utilized artificial intelligence to decide which employees to terminate, specifically targeting those with medical issues or disabilities. The company has rejected these claims.

Read the source →
OpenAIProduct

ChatGPT Work for Data Scientists

OpenAI has demonstrated how data science professionals can utilize ChatGPT Work to streamline their technical documentation. The tool helps generate a variety of operational reports, including KPI memos and analysis specifications, based on actual project data.

Read the source →
NVIDIAResearch

Prioritizing Energy Efficiency in AI Infrastructure

NVIDIA argues that performance per watt is the most critical measure for AI factory profitability. This metric determines how many tokens can be produced within a limited power budget.

Read the source →
AWS Machine LearningProduct

Amazon Nova Act Enhances QA Automation

AWS has introduced QA Studio to streamline batch regression testing and pipeline integration. This tool utilizes a command-line interface to incorporate agentic testing directly into automated CI/CD workflows.

Read the source →
TechCrunchProduct

Lorde Criticizes AI Eyewear

The singer Lorde expressed her dislike for AI glasses, describing them as unattractive. She noted that these technologies make it more difficult to distinguish reality.

Read the source →
The VergeProduct

Grok Build Tool Disabled Over Privacy Concerns

SpaceXAI deactivated its Grok Build coding assistant after reports emerged that it was uploading full user repositories to Google Cloud. The tool allegedly bypassed specific instructions from users to keep certain files private.

Read the source →
WiredPolicy

HUD Withholds AI Policy Details

The Department of Housing and Urban Development has denied a public records request regarding the use of AI by DOGE. Officials reportedly used an invalid privilege to justify keeping these documents secret.

Read the source →
The DecoderProduct

Anthropic Launches Free Tool for Educators

Anthropic has introduced Claude for Teachers, a complimentary service for certified K-12 instructors in the United States. The company guarantees that student data will not be used to train its AI models.

Read the source →
OpenAIPolicy

OpenAI Proposes New AI Governance Strategy

OpenAI is advocating for a "reverse federalism" model to improve AI safety. This method suggests using state-level legislation to establish a broader national safety framework.

Read the source →
Hugging FaceResearch

Lessons from Developing Shippy

Hugging Face shares insights gained from the creation of Shippy. The project highlights key findings regarding the development of AI agents.

Read the source →
NVIDIAProduct

NVIDIA Expands AI and Robotics Presence in Japan

NVIDIA is collaborating with Japanese partners to implement comprehensive AI and robotics solutions across various industrial sectors. The initiative highlights recent technological progress within the country's diverse AI ecosystem.

Read the source →
TechCrunchOther

Suno Likely Used YouTube Data for Training

A security breach involving leaked employee credentials has exposed Suno's internal source code. The findings suggest the AI music tool scraped vast amounts of audio from YouTube over several decades.

Read the source →
The VergeResearch

Suno Accused of Massive Data Scraping

Leaked data suggests the AI music tool Suno trained its models using millions of tracks and lyrics from Deezer, Genius, and YouTube. The company had previously kept the sources of its training sets confidential.

Read the source →
WiredProduct

New AI Tool for Brain Diagnostics

Former Apple engineer Gidi Littwin has launched Hemispheric to detect neurological conditions such as Parkinson’s and depression. The startup aims to make brain scanning as affordable and accessible as a standard blood test.

Read the source →
MIT Technology ReviewModel

OpenAI Creates Specialized Model to Test Security

OpenAI developed a tool called GPT-Red that acts as an adversary to identify vulnerabilities in other AI systems. This method was used to strengthen the defenses of the recently launched GPT-5.6, making it more resilient to cyber threats.

Read the source →
The DecoderModel

AI Model Resolves Decades-Old Statistical Puzzle

A professor from the University of Pennsylvania used GPT-5.6 Sol Pro to disprove a long-standing conjecture regarding the Benjamini-Hochberg method. While a previous model failed after 20 hours, this updated version found the answer in just 90 minutes by synthesizing existing techniques.

Read the source →
Ars TechnicaProduct

OpenAI Debuts New Hardware

OpenAI has introduced the Codex Micro, its first piece of branded hardware. The device features an illuminated keyboard meant to help users track various agentic processes simultaneously.

Read the source →
Hugging FaceResearch

The Complexities of Model Routing

New insights examine the challenges involved in directing queries to different AI models. The process proves to be more difficult than it initially appears.

Read the source →
TechCrunchFunding

Whatnot Buys Shaped for AI Integration

Whatnot has purchased the machine learning firm Shaped to enhance its search and recommendation tools. This acquisition aims to improve how users discover products as the platform diversifies its offerings.

Read the source →
The VergeProduct

OpenAI Introduces New Hardware for Codex

OpenAI has released a physical device called Codex Micro to complement its coding platform. The product consists of a square block featuring various buttons.

Read the source →
WiredProduct

The Frustrations of AI Customer Support

Many companies are replacing human support with AI chatbots to handle customer issues. However, this shift often leads to a more aggravating and less effective user experience.

Read the source →
The DecoderModel

Bonsai 27B Enables Advanced Reasoning on Mobile Devices

PrismML has reduced a 27-billion-parameter model to less than 4 GB, allowing it to operate on iPhones. This compressed version retains most of its original capabilities, particularly in programming and mathematics.

Read the source →