Models · 2023
LLaMA
Meta's LLaMA was meant for researchers only, but within a week its weights had leaked across the internet.
In February 2023, Meta released LLaMA, a family of capable language models offered to researchers under a restricted license. The goal was to show that smaller, efficiently trained models could rival far larger ones, making serious AI research cheaper and more accessible.
Within days, the model's weights leaked and spread freely online. What might have been a setback instead lit a fire: a global community began fine-tuning, shrinking and improving the models on ordinary hardware.
This sparked an open-weight movement. Enthusiasts showed you could adapt a LLaMA model on a single consumer GPU, and a wave of open alternatives to closed systems like ChatGPT followed within months.
Meta leaned in, releasing Llama 2 and later versions with far more permissive licenses, cementing its position at the center of the open model ecosystem. For many companies, an open model they could run and control was more valuable than the most powerful closed one.
LLaMA reframed a central debate in AI: whether the most powerful models should be tightly controlled or openly shared. That argument is still very much alive today.
Related stories
From history to production
We turn these ideas into working systems
The same techniques, shipped into your stack with evals, observability, and measurable ROI.