OpenAI Disrupts Global Disinformation

Today:

• OpenAI Disrupts Global Disinformation

• Google's AI Search Blunder

• Copilot Now on Telegram

• Mistral Launches Codestral AI

• Big Tech Unveils UALink Standard

• Perplexity Launches Pages Feature

• Meta Plans Premium AI Assistant

• Claude AI Gains Autonomous Tool Use

• Ex-OpenAI Researcher Joins Anthropic

• You.com Launches Custom AI Assistants

OPENAI

OpenAI has formed a Safety and Security Committee led by directors Bret Taylor, Adam D’Angelo, Nicole Seligman, and CEO Sam Altman to oversee safety measures in its projects. This committee, which includes technical and policy experts, will consult former cybersecurity officials and provide recommendations within 90 days.

Seven months after a brief ousting, CEO Sam Altman has strengthened his position by removing internal opponents and securing a significant deal with Apple. This partnership aims to integrate OpenAI’s AI technology into Apple products, potentially generating substantial revenue. Altman is also considering turning OpenAI into a for-profit company.

The Atlantic has partnered with OpenAI to make its articles available on OpenAI’s platforms, including ChatGPT, enhancing news discoverability and supporting high-quality journalism. Vox Media has also teamed up with OpenAI to integrate its content into ChatGPT, expanding reach and developing new products for its audience and advertisers.

OpenAI disrupted influence operations from Russia, China, Iran, and Israel using AI technology. These operations involved generating deceptive social media content, but OpenAI found that AI had minimal impact on their reach. This effort is part of a broader strategy to combat disinformation.

Google Admits Its AI Overviews Search Feature Screwed Up

Google's new AI Overview search feature made headlines for generating bizarre and incorrect answers, including suggesting rock eating and using glue in pizza sauce. Liz Reid, Google's head of search, acknowledged the errors and outlined improvements. The mistakes were partly due to misinterpreting satirical content and sarcastic forum posts. 

Google faced criticism from viral, often misleading screenshots. Reid emphasized that while the feature was extensively tested, real-world use revealed flaws. Google has since implemented technical improvements and will continue refining the feature based on user feedback.

WIRED

Telegram gets an in-app Copilot bot

Microsoft has introduced an official Copilot bot within the Telegram messaging app. This bot, currently in beta, allows users to chat with AI for tasks like searching the internet, suggesting movies, generating workout routines, coding assistance, translation, and finding quick facts. It's available for free on both mobile and desktop, but interactions are limited to 30 exchanges per day. 

To use it, users must share their Telegram phone number with Microsoft. This move is part of Microsoft's broader strategy to integrate AI into its various services, joining similar efforts by Meta and Google in their messaging platforms.

THE VERGE

Mistral announces Codestral, a code-generation LLM it says outperforms all others

Paris-based AI startup Mistral launched Codestral, a new code-generation AI model with 22 billion parameters, under a non-commercial license. Specializing in over 80 programming languages, Codestral aims to assist developers in generating and completing code, writing tests, and reducing errors. It has shown superior performance compared to other models, like CodeLlama 70B and Deepseek Coder 33B, especially in Python code tasks. 

Available on Hugging Face and via API endpoints, Codestral is currently in beta. Major tech firms like JetBrains and SourceGraph are already testing it for its potential to streamline software development.

VENTUREBEAT

Big Tech develops AI networking standard but without chip leader Nvidia

Major tech companies, including Meta, Microsoft, AMD, and Broadcom, announced a new networking standard for AI data centers called the "Ultra Accelerator Link" (UALink). This initiative aims to reduce dependency on Nvidia, the dominant player in the AI chip market. Notably absent from the consortium are Nvidia and Marvell Technologies. 

The UALink standard seeks to streamline communication between AI accelerators, supporting various applications like high-performance computing and cloud services. Specifications will be available in the third quarter of 2024 for companies joining the UALink Consortium. Nvidia declined to comment on this development.

REUTERS

Perplexity will research and write reports

Perplexity, an AI search platform, launched a new feature called Pages, which generates customizable webpages based on user prompts. This tool is designed to aid educators, researchers, and hobbyists by conducting research and compiling it into a structured report with sections, sources, and visuals. 

Users can choose the audience level but cannot edit the text directly. While Pages can simplify the creation of basic reports, its depth of research and accuracy are limited, making it more suitable for quick, surface-level information gathering rather than comprehensive research. Pages is being gradually rolled out to all Perplexity users.

THE VERGE

Meta Is Working on a Paid Version of Its AI Assistant

Meta Platforms is considering introducing a premium version of its AI assistant, Meta AI. This paid service would follow similar offerings from Google, Microsoft, OpenAI, and Anthropic, which charge $20 per month for enhanced chatbot capabilities. These subscriptions typically include integration with workplace applications and priority access during high-usage periods. 

The specific features and pricing for Meta's premium tier are yet to be disclosed and may change. This move aligns with Meta's broader strategy to enhance its AI offerings and compete with other tech giants in the chatbot subscription market.

THE INFORMATION

Anthropic’s Claude AI now autonomously interacts with external data and tools

Anthropic has launched Tool Use for its AI assistant, Claude, enabling it to autonomously interact with external data sources, APIs, and tools. This feature is now available across the Claude 3 model family for developers using the Anthropic Messages API, Amazon Bedrock, and Google Vertex AI. Tool Use allows businesses to automate tasks, personalize recommendations, and streamline data analysis, enhancing efficiency and productivity. 

Early adopters like StudyFetch and Hebbia are already utilizing this capability. Emphasizing security and data privacy, Anthropic ensures robust protection against AI vulnerabilities. This development positions Anthropic as a leader in enterprise AI, advocating for AI to augment human capabilities.

VENTUREBEAT

OpenAI researcher who resigned over safety concerns joins Anthropic

Jan Leike, a former OpenAI researcher who resigned over safety concerns, has joined competitor Anthropic to focus on AI oversight and alignment research. Leike’s departure aligns with Anthropic's mission of developing safer AI, contrasting with OpenAI's commercial focus post-Microsoft deal. 

Anthropic, founded by ex-OpenAI employees, emphasizes "constitutional AI," incorporating ethical principles in AI training. Leike criticized OpenAI for insufficient resources for its "Superalignment" team. His move follows another notable resignation, co-founder Ilya Sutskever, who also shares values with Anthropic. Leike aims to ensure AGI (Artificial General Intelligence) benefits humanity.

THE VERGE

You.com empowers users to create personalized AI assistants with top language models

You.com, a Palo Alto-based startup, launched Custom Assistants, a feature enabling users to create personalized AI assistants using top language models like GPT-4o, Llama 3, and Claude 3. This allows users to customize AI behavior through custom instructions, enhancing productivity in complex work-related tasks. The platform’s live web access ensures up-to-date information, setting it apart in accuracy. 

Early adopters use Custom Assistants for generating interview questions, brainstorming marketing copy, and more. Creating a Custom Assistant is easy, and the service is free, with unlimited queries for YouPro subscribers. This innovation makes AI a personalized companion, tailored to individual needs.

VENTUREBEAT