• AI Time to Impact
  • Posts
  • . . AI: Two big OpenAI moves, and many tiny, local AIs (1.10.24)

. . AI: Two big OpenAI moves, and many tiny, local AIs (1.10.24)

Friends, two big announcements from OpenAI today, and maybe they will make a big impact on the world. But it's hard to not look past those and see the amazing work being done in small, local, open source AI models.

None of these alternatives can beat GPT-4 consistently yet, but will they be able to by the end of this year? Or will that even matter, if tiny local models can do all kinds of specific disruptive things?

Welcome to all our new readers today, not just to Robin, Dan, Anne, Casey, and Igor - but seeing you subscribe in particular made me real happy. I’d love to hear from the folks who subscribed today that I don’t know.

A special shout-out to Sean Ammirati, who published a short video today about how to use LLMs to get different perspectives on your startup idea. It’s a good video and it extends on some conversation Sean and I had on his podcast.

Ok, here are today’s top stories in AI, according to our weighted analysis of AI community engagement. And here’s a 3 minute video walk through if you prefer.

First impacted: GPT tinkerers
Time to impact: Short

OpenAI has launched the GPT Store, a platform for users to access Custom GPTs other people have made. The store, which features GPTs developed by partners and the community, plans to initiate a revenue program for US builders in the first quarter. They say over 3 million GPTs have been created in two months but I have found very few that work well for me yet. It seems promising though. [Introducing the GPT Store] Explore more of our coverage of: OpenAI, GPT Store, ChatGPT. Share this story by email

First impacted: Strategic businesses
Time to impact: Short

OpenAI has launched a new plan, ChatGPT Team, offering GPT-4 and DALL·E 3 for a monthly fee of $25 or $30, including a dedicated secure workspace. Highlighted in the announcement, a Harvard Business School study indicates that Boston Consulting Group employees using GPT-4 completed tasks 25% faster and improved work quality by 40%, while companies including Block, Canva, Carlyle, The Estée Lauder Companies, PwC, and Zapier are using ChatGPT Enterprise to improve their operations. [Introducing ChatGPT Team] Share this story by email

First impacted: AI app developers, Windows users
Time to impact: Medium to long

At CES, LlamaIndex and NVIDIA announced a collaboration that they claim will facilitate the development of AI apps using local LLMs on Windows, powered by NVIDIA's TensorRT-LLM backend. The partnership, they say, will leverage community models such as Phi-2 and Llama2, and integrate a chat interface. [Supercharging LLM Applications on Windows PCs with NVIDIA RTX Systems] Share this story by email

First impacted: OpenChat-3.5 users, Developers
Time to impact: Short

OpenChat has launched Update 0106 for its chatbot UI model, OpenChat-3.5, which it claims surpasses the performance of both Grok-0 and Grok-1. It's another small one, touted by its creators as the top open-source 7B LLM, it could provide world-class AI on local devices. You can test it out at the link here. [Chatbot UI] Share this story by email

First impacted: AI researchers
Time to impact: Short

JP Morgan ML scientist Maxime Labonne has released Phixtral-4x2_8, a Mixture of Experts (MoE) model, combining four Microsoft/phi-2 models, outperforming each individual model. Created with a custom version of the mergekit library and a specific configuration, the model can be operated on a free T4 GPU using a provided Colab notebook. Small mixture of experts! [mlabonne/phixtral-4x2_8 · Hugging Face] Share this story by email

That’s it! More AI news tomorrow! Please share with friends and co-workers! Thank you!