logo
Open Source
Audio
Applied AI

llama-recipes/recipes/quickstart/NotebookLlama at main · meta-llama/llama-recipes

10/26/2024 • github.com
llama-recipes/recipes/quickstart/NotebookLlama at main · meta-llama/llama-recipes

Scripts for fine-tuning Meta Llama with composable FSDP & PEFT methods to cover single/multi-node GPUs. Supports default & custom datasets for applications such as summarization and Q&A…

Read Full Article...

C4AIL Commentary

It’s all out war in the digital economy

The 100s of billions of dollars spent in AI are not strange when put in context of all out hot war for the shape and dominance over the future of the information economy between formerly stale-mated tech superpowers. .

Today’s frontline skirmish: NotebookLM

Meta is clearly on war footing, weaponizing Open Source to strike back at any charge the competition is mounting.

Google, surprised by the success of their NotebookLM science experiment quickly rallied to try to turn it into a product to be able to claim the hordes of users driven to the project by viral videos of AI hosts in existential crisis as AI MAU

So, like with any other product Google has been putting out, Meta is shooting arrows of price dumping +5 at the target to ensure it doesn’t turn into any kind of competitive advantage. Its proprietary context window technology is not enough to keep the competition at bay

With NotebookLlama [1], they release the recipe for anyone to build a product replicating NotebookLM while waving a giant sign reading “It’s so simple, anyone can do it”.

Product is not a moat in the age of AI

Only eyeballs, exclusive access to training data and compute infrastructure at scale can create moats in AI and Meta can match Google on all of these without having to spend a dime on product engineers - just dropping models and the occasional recipes is enough for them to keep the competition from establishing any significant lead.

Text models have of course, converged increasingly on each other. Gemini has no advantage over Claude, OpenAI or Llamas in podcast creation.

Audio from OpenAI, Elevenlabs and others easily matches Google’s abilities and beyond at present.

It is clear that if dreams of dominating the market exist, they can only be realized by driving the required compute and data to such heights that all other players fail … and none of the big players has the pockets to outlast the others.

And unlike in the 2010s, having more engineers or product resources is no longer a competitive advantage. On the contrary.

Open Source is a shield

OpenSource ensures that product alone is no longer capable of establishing technology moats and many of the magical projects people fawn over in the market are rather simple scaffolding wrappers around increasingly commoditized big tech models. All while allowing meta to shrink high paid product teams.

As long as Meta’s strategy is arming the rest of the internet, no company can turn product resources into an advantage over them.