Daily AI Newsletter: Breaking: ChatGPT gets a memory 🧠

In today’s email:

👀 Andrej Karpathy is leaving OpenAI again — but he says there was no drama
🕵🏻 Ex-Salesforce and Google Veterans Secure $110 Million for Groundbreaking AI 'Agents'
🤔 Early Adopters of Microsoft’s AI Bot Wonder if It’s Worth the Money
🧰 12 new AI-powered tools and resources. Make sure to check the online version for the full list of tools.

ChatGPT will now remember — and forget — things you tell it to

OpenAI is introducing new "memory" features to ChatGPT, allowing the AI to retain and recall information across conversations. This capability is designed to make ChatGPT's responses more relevant and personalized, improving over time as users interact more with the system. Users can command ChatGPT to remember specific details, such as personal preferences or contextual information, enhancing the AI's ability to provide tailored advice and directions. The feature is currently being tested with a subset of users, with plans for a broader rollout in the future.

Additionally, OpenAI has implemented memory controls within ChatGPT's settings, allowing users to manage the AI's memory capabilities. This includes options to disable memory, view and delete specific memories, or clear all remembered information. Despite the potential for ChatGPT to store sensitive personal details, OpenAI assures users that it is taking measures to limit the AI from proactively remembering sensitive information unless explicitly instructed by the user.

For those concerned about privacy, OpenAI is also introducing a Temporary Chat feature. This option enables users to interact with ChatGPT without the AI accessing previous conversations or memories, essentially starting with a blank slate for each new dialogue. However, OpenAI may retain copies of Temporary Chat conversations for up to 30 days for safety reasons, ensuring users have a more privacy-preserving experience while still benefiting from ChatGPT's capabilities.

Andrej Karpathy is leaving OpenAI again — but he says there was no drama

Andrej Karpathy, a founding member of OpenAI and a prominent figure in the AI community, has announced his departure from the organization for the second time, emphasizing that his exit is not linked to any specific event, issue, or drama. Karpathy, known for his significant contributions to Tesla's Autopilot team and his engaging social media and YouTube presence, expressed his intention to focus on personal projects. His decision to leave follows a year marked by changes at OpenAI, including the temporary removal and reinstatement of CEO Sam Altman.

During his tenure at OpenAI, Karpathy praised the team's strength, the positive work environment, and the exciting roadmap ahead. Despite leaving, he showed optimism about the future of AI and OpenAI's role in it. Karpathy's responsibilities have been handed over to a senior researcher, with the company expressing gratitude for his contributions and best wishes for his future endeavors.

Karpathy's work at OpenAI included efforts on developing AI assistants, highlighted by his playful reference to building a "J.A.R.V.I.S" type of AI. This project reflects his ambition to create helpful, conversational AI that can serve as an empowering tool against challenges, showcasing his ongoing interest in advancing AI technologies and applications.

Ex-Salesforce and Google Veterans Secure $110 Million for Groundbreaking AI 'Agents'

Bret Taylor, following his departure as co-CEO of Salesforce in January 2023, teamed up with Clay Bavor, a veteran Google executive, to launch Sierra, a new venture focusing on conversational AI for business customers. Their collaboration, sparked over a casual lunch, has quickly evolved into a promising startup. Based in San Francisco’s SoMa neighborhood, Sierra has grown to a team of 30 and secured $110 million in funding from leading Silicon Valley investors like Sequoia Capital and Benchmark. The company aims to carve out a niche in the competitive conversational AI landscape, offering AI "agents" for customer interactions to clients such as WeightWatchers and SiriusXM.

Sierra distinguishes itself by building customized AI agents for businesses, diverging from the typical model of selling tools for self-implementation. This approach is designed to cater to companies lacking the resources or desire to undertake complex AI projects. Sierra leverages a blend of proprietary and open-source large language models, optimizing them for each client's unique needs. This strategy allows Sierra to rapidly adapt to changes and onboard new clients efficiently, setting it apart in a field where traditional models may lag in flexibility.

Behind Sierra's innovation are Taylor and Bavor, who bring a wealth of experience from their days at Google and beyond. Taylor, also serving as the chairman of OpenAI, emphasizes the non-competitive nature of Sierra's business with OpenAI, highlighting his commitment to ethical governance and the potential for AI in enhancing business operations. With a vision that reaches beyond the conventional AI narrative, Sierra aims to demonstrate the pragmatic benefits of conversational AI in everyday business applications, striving to make sophisticated technology accessible to all companies.

Nvidia’s new tool lets you run GenAI models on a PC

Nvidia has introduced Chat with RTX, an AI-powered chatbot tool for GeForce RTX 30 Series and 40 Series GPU owners, enabling offline interaction on Windows PCs. This innovative tool allows users to integrate and query personal documents, files, and notes using a customizable GenAI model, similar to OpenAI's ChatGPT. It leverages Mistral’s open-source model by default but is also compatible with other text-based models, including Meta's Llama 2. With support for various file formats and the ability to incorporate YouTube playlist transcriptions, Chat with RTX requires substantial storage space, ranging from 50GB to 100GB, depending on the chosen model.

However, Chat with RTX has limitations, such as its inability to remember context between questions and the influence of multiple factors on the relevance of responses, including question phrasing and dataset size. Nvidia suggests that the tool's effectiveness improves with larger, more focused datasets, but acknowledges that it's more of a novelty than a tool for professional use. Despite these drawbacks, the application underscores a growing trend towards running AI models locally, offering benefits such as privacy, lower latency, and cost savings over cloud-hosted models.

The emergence of tools like Chat with RTX reflects a broader movement towards democratizing access to AI technologies, enabling users to run and potentially train models offline on personal devices. While this opens up possibilities for misuse, the push towards local AI processing highlights the balance between innovation and the need for responsible use. As offline GenAI models become more accessible, the debate over their benefits and potential risks continues, with proponents emphasizing privacy and efficiency gains.