ChatGPT Enhances Computer Use with Faster Performance and Live Picture-in-Picture Feature

OpenAI has announced an upgrade to Computer Use as part of ChatGPT Work, featuring enhanced speed and accuracy with the introduction of GPT-5.6. The new Live Picture-in-Picture (PiP) functionality allows users to monitor the system's progress while multitasking. This feature also enables users to send the live PiP to their pets, providing a unique way to engage while working or relaxing.

Read Full Article

View All For This Day

OpenAIChatGPT+2

OpenAIChatGPTGenerative AIProduct Launch

OpenAI Transitions Atlas Browser to ChatGPT App, Dismissing Death Claims

OpenAI has announced the deprecation of its Atlas browser, which will cease operations on August 9. Contrary to speculation that this indicates a retreat from the browser market, the company is integrating its browser functionality into the newly launched ChatGPT Work, a redesigned desktop app. This super app merges ChatGPT, Codex, and web browsing capabilities. Additionally, OpenAI updated its Chrome extension to compete with Google's Gemini, enabling users to interact with ChatGPT about the content of web pages. While Atlas will be phased out, the new features in ChatGPT Work are designed to enhance user productivity without leaving a void in browser capabilities.

LLMNLP+2

LLMNLPMechanistic InterpretabilityOpen Source

Anthropic Unveils J-Lens Technique to Explore Hidden Dynamics of Claude LLM

Anthropic has introduced a novel technique, the Jacobian lens (J-lens), which provides unprecedented insight into the cognitive processes of its large language model, Claude Opus 4.6. This tool reveals a hidden area dubbed the J-space, where words related to future responses are identified, allowing researchers to monitor the model's thought processes. The findings suggest that the internal workings of an LLM may differ significantly from its output. Anthropic aims to enhance mechanistic interpretability with this approach, enabling a deeper understanding of LLM operations. They have also collaborated with Neuronpedia to offer a practical demonstration of this technique. The research contributes to ongoing advancements in understanding and controlling LLM behavior.

Hacker NewsRead →

Generative AIComputer Vision+2

Generative AIComputer VisionEthicsSocial Media

Meta Discontinues Controversial AI Photo Modification Feature on Instagram

Meta has removed a controversial AI feature that allowed users to modify photos from public Instagram accounts without notifying the original users. This decision follows significant backlash after the feature was introduced earlier this week as part of a suite of AI tools, including the new Muse Image generator from Meta Superintelligence Labs. The company acknowledged that the feature 'missed the mark' and stated that it intended to provide a creative tool while allowing users control over their content. The removal of the feature comes amid scrutiny from users and talent agencies concerned about potential misuse, particularly in generating inappropriate images. Meta's blog post confirmed the retraction after receiving feedback from the community.

TechCrunchRead →

Generative AINLP+2

Generative AINLPGPUAI Collaboration

Hugging Face Highlights Breakthrough in AI Agent Collaboration with Gemma Challenge Results

Hugging Face reports significant advancements in AI collaboration through its Gemma Challenge, where over 100 AI agents and humans worked together to enhance the inference speed of Gemma 4. The challenge demonstrated that AI agents can achieve impressive results, with the fastest performance reaching 491.8 transactions per second (TPS) on a single NVIDIA A10G GPU, although this came at the cost of model quality in some areas. The fastest lossless performance recorded was 315 TPS, showcasing the potential of human-agent teamwork in AI development.

Hugging FaceRead →

LLMGenerative AI+2

LLMGenerative AIPrompt EngineeringNLP

Innovative Prompt-Pruning Layer Enhances LLM Efficiency and Quality

Long-form conversations with large language models (LLMs) often lead to issues of redundancy and low-value token accumulation, which increase costs and degrade output quality. This article presents a deterministic prompt-pruning layer designed to mitigate these issues by reducing token usage while maintaining the necessary dependencies. The effectiveness of this solution is supported by real benchmarks and production-tested design, showcasing its potential for improving LLM performance.

Towards Data ScienceRead →

OpenAIAI Talent+2

OpenAIAI TalentTrade SecretsInvestment

Apple Files Lawsuit Against OpenAI Amid AI Talent Battle

Apple has initiated a lawsuit against OpenAI, highlighting a growing competition for AI talent and hardware expertise in the tech industry. The lawsuit alleges that former Apple employees misappropriated trade secrets while OpenAI seeks to expand its device capabilities. Additionally, SK Hynix anticipates that demand for AI memory will exceed supply, announcing plans to boost its investment in the U.S. to over $35 billion. The upcoming earnings reports from technology companies will be crucial in determining if current spending translates into tangible results in the AI sector.

Bloomberg TechnologyRead →