Meta's LLaMA Leaked - GPT Caliber Large Language Model Now In The Wild π€―
Who expected the open source moment for LLMs would happen like this?
π€― Hiroshima dropped and there was pin drop silence. Weights for a powerful large language model by Meta are now being torrented into ubiquity βοΈ
π§ What this means is everyone now has access to a GPT caliber model, without the (often) acutely limiting trust & safety layer of an OpenAI API. This is sex without a condom, and I donβt mean waifu diffusion π
π¬ This will do for text as a modality what stable diffusion did for imagery this past year vs. DALL-E 2. Of course new capabilities will arise working in concert with it, RLHFing it, empowering use cases aplenty π
ποΈ Calls for government intervention are fascinating. I get that position but have also seen the benefits of open source first hand. Make no mistake β itβs a double edged sword βοΈ
πΈ Guaranteed awesomeness will ensue with a level playing field in LLMs eroding OpenAIβs advantage, but also unleashing large scale spam and misinformation on the internet π§
βοΈ Itβs a fine balance to strike. You can let things acclimatize in open source or you can treat AI like nuclear β system level surveillance to stop rogue AI while also controlling access to GPU clusters β’οΈ
π° Will be interesting to see how USG reacts. Other nation states already had these systems. And now the community does too. What do you think β How might this play out? π€¨
Join the conversation on Twitter, plenty of speculation on how/why this went down the way it did: