. .AI: VC comes to Japanese AI, OpenAI on Democracy, and Hermes on Hugging Chat (1.16.24)

Friends, today’s AI news is really interesting, if you’re into international affairs, new chatbots, and cost-effective new ways to classify very large sets of documents.

As always, these are the stories the AI community is talking about most today, and over the weekend, according to our weighted analysis of community engagement.

Thanks for joining us. Now here’s today’s news.

Marshall Kirkpatrick, Editor

Sakana AI Secures $30M in Seed Funding

First impacted: AI researchers, Venture capitalists
Time to impact: Medium to long

Tokyo-based Sakana AI says it has secured $30M in seed funding, with contributions from Lux Capital and Khosla Ventures, among others. The company aims to establish a leading AI lab in Japan to develop new foundational models inspired by nature, such as evolution and collective intelligence.

• Sakana AI, inspired by the Japanese term for fish, utilizes the concept of a school of fish as a metaphor for their AI development strategy. They say this strategy focuses on nature-inspired intelligence and group decision-making, where the red fish symbolizes their commitment to pursuing innovative trends.

• As they write in their blog post, "To our knowledge, Sakana AI is the first AI startup founded in Japan (or almost the first in any industry in Japan) backed by top Silicon Valley VCs at the seed stage."

[Sakana AI] Explore more of our coverage of:Funding,biology,Japan,AI Labs. Share this storyby email

OpenAI Selects 10 Teams for Democratic AI Development

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

Can AI be built democratically? OpenAI has concluded a program that saw nearly 1,000 applications from 113 countries, and ultimately selected 10 teams to develop prototypes for shaping AI system behavior through democratic input. The teams, from 12 countries and diverse fields, showcased their projects on a Demo Day and OpenAI now has now shared the code from their projects. You can see them all via the "reports" links in this article and they look pretty cool! [Democratic inputs to AI grant program: lessons learned and implementation plans] Share this story by email

OpenAI Prepares for 2024 Global Elections

First impacted: Voters, Secretaries of State
Time to impact: Short to medium

As the elections approach worldwide in 2024, OpenAI is focusing on preventing misuse of AI technologies, "misleading deepfakes, scaled influence operations, or chatbots impersonating candidates", and increasing transparency in AI-generated content. In partnership with the National Association of Secretaries of State (NASS), the company will steer US voters towards trustworthy voting information, while ensuring that its applications do not disrupt democratic procedures or misrepresent voting systems. Sure makes me wonder - what about the leading open source models?

• OpenAI is incorporating ChatGPT with live global news sources to increase transparency. This is said to give users instant news updates, complete with source attribution and links, aiming to provide diverse information sources and assist voters in evaluating the validity of the information they receive.

[How OpenAI is approaching 2024 worldwide elections] Explore more of our coverage of:OpenAI,Elections,Security. Share this storyby email

Now on HuggingChat: Nous-Hermes-2-Mixtral-8x7B-DPO

First impacted: AI researchers, general public
Time to impact: Short

Hugging Face has put NousResearch's Nous-Hermes-2-Mixtral-8x7B-DPO up on Hugging Chat for anyone to chat with. Give it a try! This is the top of the line open source stuff: Hermes 2 small locally optimized version of Mixtral's Mixture of Experts model. Somebody make a chef's kiss gesture! [HuggingChat] Explore more of our coverage of: Hugging Face, Hermes, Mixtral. Share this story by email

Anyscale and Pinecone Launch Cost-Efficient Vector Database Service

First impacted: Data Scientists, AI Engineers
Time to impact: Short

Anyscale, in collaboration with Pinecone, has launched a service for creating a distributed vector database to generate embeddings for Retrieval-Augmented Generation (RAG) applications. Pinecone's serverless vector database, which only charges users for what they search, is credited for reducing costs by up to 50 times, particularly when generating embeddings for the sizeable falcon-refinedweb dataset, containing over 600 billion tokens. [RAG at Scale: 10x Cheaper Embedding Computations with Anyscale and Pinecone] Explore more of our coverage of: Vector Database, Retrieval-Augmented Generation, Pinecone, Anyscale. Share this story by email

Meta Launches MAGNeT, Open-Source Text-to-Audio Model

First impacted: Music Producers, AI Developers
Time to impact: Short

Meta has launched a model named MAGNeT, available on the Hugging Face platform, which it says transforms text into music and audio. The company claims this model operates seven times faster than its counterparts while preserving quality, generating sequences by predicting masked token spans and utilizing a rescoring method. [MAGNeT - a facebook Collection] Explore more of our coverage of: Meta, Text-to-Audio . Share this story by email

Hugging Face Launches Machine Learning Contest Platform

First impacted: Data scientists, Machine learning engineers
Time to impact: Medium

Hugging Face has launched a new platform for hosting machine learning contests, according to the company. The platform, which charges for compute resources, facilitates two types of contests: one in which participants submit predictions as CSV files and another where code is provided to generate predictions, and it can host both public and private contests. [GitHub - huggingface/competitions] Explore more of our coverage of: Hugging Face, Machine Learning, Contests. Share this story by email

That’s it! More AI news tomorrow.