This Week in AI: July 22–28

During the past week, model launches and updates to AI-powered assistants and features dominated the headlines. Meta and Mistral AI undoubtedly stole the show, each releasing large open models aimed at competing with state-of-the-art foundation models like GPT-4o or Claude 3.5 Sonnet. Meta released Llama 3.1 405B, one of the largest models lately released under a permissive license. At the same time, the company updated its 70B and 8B models.

Meta made the Llama 3.1 450B available for immediate preview on its AI-powered assistant Meta AI, which is now powered by Llama 3.1 70B by default. Meta AI also received some updates expanding its multilingual capabilities, its geographic availability, and a new feature called "Imagine me", which lets users generate images based on their photo, and how they complete the prompt. It was not long before Mistral AI matched Meta's announcement by introducing Mistral Large 2, a powerful new 123 billion parameter language model with multilingual support, advanced coding and reasoning capabilities, and wide availability through major cloud platforms.

Google joined the party by announcing updates to its Gemini assistant, including the availability of Gemini 1.5 Flash powering the free tier, an expanded context window, new features to combat hallucinations, and wider availability across platforms and regions. Google seems focused on building up the Gemini Assistant, increasing the ways and places where users can access Gemini, including the standalone mobile app and access through Google Assistant. The company is also expanding its audience, launching access for teens on their Workspace for Education and private accounts.

Meanwhile, Bing and OpenAI are experimenting with combining web search with generative AI. Bing has introduced a new generative search experience that combines AI-generated content with traditional search results, offering users more comprehensive and dynamic responses to their queries. Similarly, OpenAI launched SearchGPT, a prototype assisting users who want to enhance their web search experience with ChatGPT's conversational and summarization capabilities.

Other notable headlines include:

Cohere raised $500M in its Series D round at a $5.5B valuation: Toronto-based AI startup Cohere has raised $500 million in a Series D funding round led by PSP Investments, valuing the company at $5.5 billion and highlighting its rapid growth in providing customized large language models for enterprise customers.

Cohere's Rerank 3 Nimble supports fast and accurate enterprise search applications: Cohere has launched Rerank 3 Nimble, a new AI model that enhances enterprise search and RAG systems by offering 3x faster performance, multilingual support, and improved accuracy across various tasks, available on Amazon SageMaker and for on-premise deployments.

Harvey is the latest legal tech startup to announce a funding raise: AI-powered legal tech startup Harvey has secured $100 million in Series C funding led by GV, reaching a $1.5 billion valuation, and plans to use the investment to enhance its platform and expand globally in the professional services sector.

Adobe introduced new Firefly AI-powered features for Photoshop and Illustrator: Adobe introduced a Text-to-Image feature for Photoshop powered by its Firefly Image 3 Model, and it is also beta-testing a Generative Shape Fill feature powered by its Firefly Vector Model. Adobe recently attributed its reported record growth to its subscription model and approach to generative AI.

The FTC is gathering information on surveillance pricing products and services: The FTC has ordered eight companies to provide information about their products and services, focusing on their technical implementations, data sources, customer base, and potential impact on consumer privacy, to understand their effects on consumers.

Stable Video 4D showcases Stability AI's research into multi-angle video generation: Stability AI has launched Stable Video 4D, a revolutionary AI model that transforms single-view videos into eight novel-view videos from different angles, offering potential applications in game development, video editing, and virtual reality.

Lakera raised $20M in Series A funding to continue securing generative AI applications: Lakera, a company focused on securing generative AI applications, has raised $20 million in Series A funding to develop dynamic security solutions to address the growing need for AI security as enterprise adoption of generative AI is expected to reach 80% by 2026.

Fractile, the latest NVIDIA contender, emerged from stealth with $15M in seed funding: Fractile is the most recent startup to enter the AI chipmaking competition. The startup plans to develop chips that combine memory and processing into a single component, drastically boosting performance while slashing costs and energy consumption.

The US National Institute for Standards and Technology released new guidance and a tool to test AI models for risk: The National Institute for Standards and Technology (NIST) has released five products, including three finalized guidance documents, a draft on risk mitigation for AI models, and an open-source testing platform, to promote safe, secure, and trustworthy AI development.

European regulators are looking into X's most recent user data harvesting practices: Elon Musk's X has quietly updated its data collection policies to opt all users into sharing their data to train Grok, xAI's model, raising concerns about compliance with EU and UK GDPR laws and potentially facing regulatory scrutiny similar to that experienced by Meta.