OPEN SOURCE LLM AI SNOWBALL ROLLS ON: Cited here, Facebook, Google, Amazon, Microsoft, Stability, and OpenAI.
WHAT’S OF NOTE:.
Open Source Large Language Model (LLM) AI Software Infrastructure continues to be a very rapidly rolling Snowball every week. And it’s speeding up.
As we discussed in the preceding post, Facebook/META has long played a significant role on the Open Source LLM AI and Software infrastructure front. As explained below, their business model allows that flexibility.
Notable this year has been the the roll out of its quasi-open source LLaMA models in March, which has been a key catalyst to open source initiatives since then. Their impact was noted of course, in the now infamous “Google (and OpenAI) have no Moat” leaked memo of a few weeks ago, by a senior Google engineer, Luke Sernau. As he notes,
“Open source companies, plainly put, are lapping us. Things we consider “major open problems” are solved and in people’s hands today. Just to name a few:
LLMs on a Phone: People are running foundation models on a Pixel 6 at 5 tokens / sec.
Scalable Personal AI: You can finetune a personalized AI on your laptop in an evening.
Responsible Release: This one isn’t “solved” so much as “obviated”. There are entire websites full of art models with no restrictions whatsoever, and text is not far behind.
Multimodality: The current multimodal ScienceQA SOTA was trained in an hour.
While our models still hold a slight edge in terms of quality, the gap is closing astonishingly quickly. Open-source models are faster, more customizable, more private, and pound-for-pound more capable.”
As noted in my last post, Facebook/Meta Founder/CEO continues to use Open Source LLM AI initiatives to throw sand in the business model gears for prime competitors OpenAI/Microsoft, Google, Amazon and others.
Meta’s business models aren’t geared around Search, Prompt Queries or API cloud usage fees which have been multi- billions fountains of revenues and free cash flow for the other major incumbents ($GOOG $AMZN $MSFT) and others. They benefit from the content and services (and Ads off them), on top of the AI Tech Value Stack chart, to be discussed in a later post.
This past week there were continued developments the Open Source LLM AI front. One notable element was Stability’s release of StableStudio, an open source version of its DreamStudio text to image product.
“Stability AI is releasing an open-source version of DreamStudio, a commercial interface for the company’s AI image generator model, Stable Diffusion. In a press statement on Wednesday, Stability AI said the new release — dubbed StableStudio — “marks a fresh chapter” for the platform and will serve as a showcase for the company’s “dedication to advancing open-source development.”
Making an open-source version of DreamStudio carries benefits for Stability AI. It allows community developers to improve and experiment with the interface, with the company potentially reaping the rewards conferred by these improvements. Stability AI stressed community building in its press release, noting how “from enabling local-first development, to experimenting with a new plugin system, we’ve tried hard to make things extensible for external developers.”
This is also notable since it’s a concurrent release of an open source product that competes with a for profit product, DreamStudio, by the same company. Will be a notable example of how both approaches can potentially drive innovation forward at the same time.
In other Open Source developments (besides of course Facebook’s latest Open Source LLM AI with new capabilities (aka modalities), Imagebind, “and the EU AI ACT’s anti-open source initiatives, discussed in earlier post, would be the following.
The Information reported that OpenAI may be mulling an Open Source offering of its own. As Reuters expands:
“OpenAI is preparing to release a new open-source language model to the public. OpenAI is unlikely to release a model that is competitive with GPT, the report added”.
This may be a trial balloon by the company, given the recent investor and media focus around open sourced models, but in theory the company’s earlier models, GPT2 and GPT3, could theoretically be open sourced with caveats, while GPT4 is available with commercial API terms and subscriptions.
This also allows the company to fend off developers standardizing around Facebook/Meta’s LLaMA hardware and software infrastructure. Also interesting applications are possible with open sourced LLM AI versions of the earlier models for Local inference and even training on user computer and mobile devices. It’s an area of exuberant developer open source activity, accelerating every week. Stay tuned.