Open source AI foundation models will wipe out closed and proprietary AI models for the same reason Wikipedia wiped out generalist commercial encyclopedia: crowd-sourced human contributions to open platforms can cater to a high diversity of interests, cultures, and languages.
Yann LeCun I’m going to ask again as you’re not answering or covering this vital issue. Ignoring it is signalling that we’re right to worry about bias and even suggests intentionality. What are you doing at Meta to tackle the overwhelming Zionist bias and misinformation, the bots and the biased moderation, censorship of Palestinian voices, and spread of Zionist propaganda that’s leading directly to genocide? The misinformation your platforms is peddling and promoting is deathly. Why won’t you cover this and respond Yann LeCun?
The same statement didn’t hold for operating systems.
It's a hotly debated topic since decades. Open source Operating Systems (OS) weren't able to displace commercial OSs like Windows and Mac. I believe the same will be the fate of open source AI models.
Just like Linux based OS has wiped out Microsoft & MAC OS for home users ? The data says otherwise . Chat GPT is a proprietary model targeted for consumer applications It will work on the same model as Microsft OS and Mac OS which are closed source but still the number 1 & 2 OS used in households. Linux despite beeing free and opensource has yet to take over the closed source Microsoft OS and Mac OS in the home users market . Same will apply to Chat GPT
Yeah as long we don’t continue with the same open theft of copyright works.
In this case, you’re wrong. To achieve what OpenAI achieved requires tens of millions of dollars in extremely careful data labeling. Open source LLMs have been around for awhile now, and they’re a toy, at best, ultra conspiracy theorists at worst.
Truly obvious atp 😊
Open source or closed source won't matter much except for the fact that in closed source we won't know whether the weights are genuinely used without manual modifications. But even knowing that thing isn't quite useful. Who can read into billion numbers anyways! And even manual modifications to weights won't be any different than adding/deleting some documents on internet :) If those are trained on all of the internet as base as they currently are, they might be pretty useless/noisy/biased/full with security issues etc. Both might have different issues from operating stand point of view apart from these basic issues. (Off course this is knowing their power - of what they CAN do, both good and bad) That being said, both open source and closed source will still be used (more open source for the obvious reason of cost and flexibility!), in the same way we still use social media :) (Having both good/bad, paid/unpaid versions in them and living with that!)
Like Linux wiped up everything else. What you have in Wikipedia is both public domain and often not very reliable.
AI/GenAI/ML/MLOps/DevOps | PhD
1yI don’t agree. The classical open source software outperforms proprietary sources sometimes becuse it only needs human resources and time, but open source AI requires computation resource additionally, and this is not free. Humans will voluntarily give their mind and time but not their monetisable resources (computation). A torrent-like computation sharing network for open source AI may solve the problem.