One of the fears of open sourcing this tech is that now instead of one company hopelessly out of control, you may have many players who are clueless as to what’s happening under the hood and it’ll be too late when shit hits the fan.
I can see logic in this.
There are very small numbers of entities who can train LLM models of this scales. Not everyone can access to 100000 GPUs. Open sourced codes will not let people train it because of this reason.
Instead, smart but not necessarily wealthy scientists can check the code, I get your side of story but no regular people can actually train it even thought they could manipulate code in their ways. If they are able to get that much computational power, they probably are probably wealthy enough so that their people able to write their own codes. I think.
