Not the base model, presumably that's fine. There'll be finetunes like Wizard or Orca that don't have these issues.
The base model was probably under $2 million on their in-house GPUs. The data that turned it into this cost them another $20 million.
The smaller models will refuse to tell you how to make mayonnaise.
Try 70b here: https://huggingface.co/spaces/ysharma/Explore_llamav2_with_TGI
And 13b and 7b here: https://www.llama2.ai/
--
The only data filtering they did was remove Kiwi Farms
--
Even with no system prompt, it refuses to tell you how to survive a bomb
Jump in the discussion.
No email address required.
There's kids in middle school learning from YouTube how to put together a more competent LLM than anything Meta will come up with
Jump in the discussion.
No email address required.
Meta and Microsoft announced they will work together on AI today.
Jump in the discussion.
No email address required.
More options
Context
There using the leaked llama code as well
Jump in the discussion.
No email address required.
More options
Context
More options
Context