. . AI: Humane's mixed reviews, a new #1 open LLM, make-your-own GPT and more (11.9.23)

Friends, today's AI news is wild!  Hugging Face updated their open LLM leaderboard and Kai-Fu Lee's new unicorn startup is #1! Humane launched its AI pin - do you love it or do you hate it? Even the LLMs I asked couldn't agree.

Here are the top 7 things people are talking about in AI today.  All 7 of them are very interesting today. In some day’s news, I would have led with today’s 7th story!

For those who are new here, each day we track the news stories rising to the top of a weighted analysis of AI community engagement, then offer short summaries and time to impact hypotheses, with the help of an ensemble of LLMs.

Let me know what you think! Here’s today’s news…

Marshall Kirkpatrick, Editor

Humane Launches AI Pin to Mixed Reactions

First impacted: AI technology enthusiasts, Apple product users
Time to impact: Medium to long

Ex Apple leaders launched Humane's AI Pin today and though I shrugged it off yesterday, many people in AI were talking about it today. The 10 minute demo video on the home page and Om Malik's interview with co-founder Imran Chaudhri are both worth a look. Claude.ai, when I asked, said: "Interesting concept but limitations in functionality and niche appeal limit near term impact." Mistral 7B, though, says "the AI Pin is likely to be adopted by many AI technology enthusiasts and Apple product users. The potential impact on the lives of these users could be significant." [Hu.ma.ne ] Share by email

Stanford's NOIR System Enables Brain-Controlled Robots

First impacted: Neuroscientists, Robotics Engineers
Time to impact: Long

Stanford researchers have developed a new system, Neural Signal Operated Intelligent Robots (NOIR), that employs EEG to convert human brain signals into robotic actions. The NOIR system, capable of interpreting complex intentions like "pick up the mug from the handle", has been successfully trialed on "20 challenging, everyday household activities" from "CollectToy" to "PetDog" to "Sandwich," "OpenGift," and "CovidCare." [NOIR: Neural Signal Operated Intelligent Robots for Everyday Activities] Share by email

Hugging Face Updates Open LLM Model Leaderboards; 01.AI is #1

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

Hugging Face says it has improved its open-source LLM leaderboards by incorporating three new metrics, a step the company says was required due to the models' growing power and the demand for a more rigorous, real-world applicable assessment system. The evaluation tests models on 4 benchmarks against 7 tests of knowledge, truthfulness, reasoning, and math. Now leading the charts: Yi-34B from Kai-Fu Lee's new company 01.AI, Godzilla 2-70B from Philippine financial services company Maya, and Stellar Bright, an "upgrade to Llama 2" from Canadian AI firm Valiant Labs. Pretty international! [Open LLM Leaderboard] Share by email

OpenAI Launches GPTs for ChatGPT+ Subscribers

First impacted: ChatGPT+ subscribers
Time to impact: Short

OpenAI has launched the ability to create your own chatbots (GPTs) for all ChatGPT+ subscribers. Every one I tested was buggy and when I tried creating two myself, they were unable to perform the tasks I asked them to effectively or at all. [GPT Editor] Share by email

OpenAI Launches Data Partnerships Initiative

First impacted: AI researchers, Government organizations
Time to impact:

OpenAI has initiated a new project, OpenAI Data Partnerships, designed to encourage cooperation between organizations to exchange public and private datasets, thus advancing AI research and development. This project, already in collaboration with the Icelandic Government and the Free Law Project, provides two methods of cooperation: the Open-Source Archive for generating open-source datasets for LLMs training, and Private Datasets for training exclusive AI models using private data. [OpenAI Data Partnerships] Share by email

Latent Consistency Models Speed Up Image Generation, A Lot

First impacted: Digital Artists, AI Researchers
Time to impact: Short

A new model for rapid generation of images called Latent Consistency Models, led by a team from China's Tsinghua University, has had a collection of resources released today. From the Hugging Face blog: "To gauge the speed difference we are talking about, generating a single 1024x1024 image on an M1 Mac with SDXL (base) takes about a minute. Using the LCM LoRA, we get great results in just ~6s (4 steps). This is an order of magnitude faster, and not having to wait for results is a game-changer. Using a 4090 [Nvidia's market leading RTX 4090 GPU], we get almost instant response (less than 1s). This unlocks the use of SDXL in applications where real-time events are a requirement." [SDXL in 4 steps with Latent Consistency LoRAs] Share by email

Adept AI Labs Launches Experiment Platform, Demoing AI Agent Workflows

First impacted: Businesses, Industry experts
Time to impact: Medium

Much-anticipated Adept AI Labs has introduced Adept Experiments, a new platform that includes self-contained mini-tools and demonstrations of its multimodal autonomous AI agent framework for businesses. The experiments take the form of a browser sidebar users interact with, then an AI agent performs complex multi-step tasks like filing expenses or creating discounts in a Shopify store, automatically. [Introducing Adept Experiments] Share by email

That’s it! More AI news tomorrow