. .AI: Replicate funding, NexusRaven beats GPT-4 at tool use, and more (12.5.23)

Friends, here's the news the AI world is talking about most today. Not yesterday's reports of a next Mistral funding at $2B valuation (Bloomberg) or today's reports of Musk's xAI filing to raise $1B (CNBC). Our lens is heavily influenced by the developers building tech technology, and these are the stories they're talking about most.

Welcome to our new readers, I see there are quite a few people joining this week from the world of finance. I hope this developer-inspired view of what’s hot in AI will prove useful to you, literally, as analogy and inspiration, or both.

Thanks to everyone spreading the word about this news letter.

Now on to today’s news…

Marshall Kirkpatrick, Editor

Replicate Secures $40M in Series B Funding

First impacted: Open-source devs
Time to impact: Medium

Replicate has raised $40 million in a Series B funding round led by a16z, with participation from NVentures, Heavybit, Sequoia, and Y Combinator. The company, which started with generative images and has experienced growth in open-source language models, plans to use the funds to enhance its business platform, accelerate predictions, and refine language models. This follows a rise in registrations to 2 million users, 30,000 of whom have become paying customers. [Businesses are building on open-source AI] Share by email

NexusFlow Launches NexusRaven V2, Outperforms GPT-4 at Tool Use

First impacted: AI Agent Developers
Time to impact: Medium

NexusFlow has introduced NexusRaven V2, a 13B parameter LLM. They say it outperforms GPT-4 in real-world zero-shot tool use by up to 7%. In this case, zero-shot tool use means the platforms were tested on cold user prompts to do things like pull data from a Google Maps API, a climate data API, and multiple complex cybersecurity APIs. [Nexus Function Calling Leaderboard - a Hugging Face Space by Nexusflow] Share by email

Magicoder Launches Low-Bias, High-Quality Code Instruction Models

First impacted: Developers
Time to impact: Medium

Developers at the Department of Industrial and Enterprise Systems Engineering at University of Illinois Urbana-Champaign has introduced Magicoder, a series of models designed to produce high-quality, low-bias instructional data for coding. Available in different sizes, they are trained on synthetic data and can generate code snippets. The Magicoder-S-DS-6.7B model, for instance, can demonstrate a Python implementation of a TODO list API. [GitHub - ise-uiuc/magicoder: Magicoder: Source Code Is All You Need] Share by email

Pika Demos Modify Region Video Editing By Text

First impacted: Video editors, Content creators
Time to impact: Medium

Pika, a high profile new text to video startup, shared a short demo of its new AI 'Modify Region' feature online. It's a cool feature that lets you highlight a part of a video and type how you want that region to be changed. How long until the next US Presidential election? [via @pika_labs] Share by email

Morph Labs' Self-Teaching Method Boosts Chat Language Models' Learning

First impacted: AI developers, AI researchers
Time to impact: Medium

Morph Labs: "We have developed a proprietary method called self-teaching which allows chat language models to learn new knowledge with better multistep reasoning and significantly better preservation of previous capabilities than typical finetuning methods. Self-taught models forget less off-topic information, remember more on-topic information, and are better at complex reasoning over learned facts without looking at the source material." [Language model self-teaching for domain adaptation] Share by email

AI Alliance Launched to Foster Open Innovation

First impacted: AI developers, Academics
Time to impact: Long

The AI Alliance, an international group initiated by IBM and Meta, has launched to promote open innovation and science in AI. The Alliance, which includes partners such as AMD, Dell Technologies, Intel, and Oracle, plans to create resources for responsible AI development, support global AI skills building, and engage the academic community. This will be achieved through member-driven working groups, a governing board, and a technical oversight committee. [AI Alliance Launches as an International Community of Leading Technology Developers, Researchers, and Adopters Collaborating Together to Advance Open, Safe, Responsible AI] Share by email

That’s it! More AI news tomorrow!