What is even the codebase of an llm? Meta published their training code, the inference code has always been open source. If the training data is considered, this will 100% always resolve to no since no company can publish their training data for copyright reasons. Is this market just a bet on how this question will get answered?

The answer here is almost certainly going to be "yes if llama counts as open source" (meta calls it open source, but eg Simon Willison would insist it's only "open weights")

The resolution criteria pretty clearly says “with its code base open to the public.” I tried to get an answer to this below but OP appears to be inactive

so that'd mean this would resolve NO? why not bet it down using this line of reasoning

ig you already have a large % of your nw on NO so it's reasonable not to put more. Ig this is a pretty strong case for NO since afaict Llama 3.1 405B doesn't have its codebase freely available to the public, so doesn't satisfy the resolution criteria. tbc I'm not speaking in my capacity as a mod, just confused about the market price

Just hedging against the mods interpreting things differently but yeah agreed my odds of that were lower than 60

Would Llama 3 be considered open source under the resolution criteria?

Looking good

"The model must be verifiably open source" - by "open source", do you mean just "available", or do you mean this (widely-accepted) definition: https://opensource.org/osd/

"The model must be verifiably open source, with its codebase freely available to the public." -- should also the model weights (parameters) be freely available? I recommend yes.

What does "its codebase freely available to the public" mean? Does a torrent containing just the weights (a la Mistral) count?

@dp I recommend it doesn't count, since from the weights alone, the model cannot be independently executed