Mistral open-sources new LLMs for edge agents

The company releases new lightweight models, a roundup of the tech giants making agent moves, and more

Author

Michael Cunningham
October 17, 2024

Welcome back to Building AI Agents, your biweekly guide to everything new in the AI agent field!

With agents beginning to replace human labor across a wide range of jobs, it was only a matter of time before the position “memecoin scam shill” met its match

In today’s issue…

  • Mistral’s new lightweight LLMs

  • A simple new framework by AgentOps

  • What stocks will be the winners in the agentic economy?

  • Evaluating software engineering agents…with agents

…and more

📰 NEWS

Source: Wikipedia

The French LLM provider has released a pair of lightweight LLMs specializing in reasoning and function calling, capable of running agentic workflows on edge compute.

Decagon, a startup which builds AI customer support agents for major enterprises such as Duolingo and Eventbrite, has raised a total of $100 million after recently completing a Series B funding round for $65 million led by Bain Capital.

If you find Building AI Agents valuable, forward this email to a friend or colleague!

🛠️ USEFUL STUFF

Source: GitHub

This repo, forked from OpenAI’s experimental new Swarm agent framework, provides support for Anthropic and Groq LLMs.

Agent orchestration startup AgentOps has entered the agent framework arena with this package which integrates other major providers such as LangChain, LlamaIndex, AutoGen, and CrewAI, and includes AgentOps’ technology for observability.

Google Cloud, Weights & Biases, Copilot Kit, and others are sponsoring a hackathon by meetup group AI Tinkerers on November 2-3, in which users will build human-in-the-loop agents.

💡 ANALYSIS

Source: Adapted from Wikipedia

This piece reviews many of the major tech players making moves into AI agents, predicting which may benefit from the field’s rapid growth.

This author of this article forecasts the impact that agents will have on the technology industry in the coming year, including a brutal shakeout in the GenAI space.

Cybersecurity experts are warning that, as agents rapidly integrate into enterprise IT systems, companies are not taking adequate steps to vet the autonomous decisions they make.

🧪 RESEARCH

Source: ArXiv

The authors of this paper, including pioneering AI researcher Jürgen Schmidhuber, propose a new method of using AI agents to judge and enhance the performance of their ML engineering-oriented fellows, laying the groundwork for self-improving AI.

This paper finds that LLMs trained to refuse potentially dangerous inputs when used as chatbots do not maintain this behavior when used as the backbone for agents, making them them potential vectors for attack.

Thanks for reading! Until next time, keep learning and building!

If you have any specific feedback, just reply to this email—we’d love to hear from you

Follow us on X (Twitter), LinkedIn, and Instagram