Skip to main content


#Meta says that its #Llama #AI tools are "the best #OpenSource models of their class, period."

From @kylelwiggers: "There’s only one problem: the Llama…models aren’t really “open source”… Open source implies that devs can use the models how they choose…But…Meta has imposed certain licensing restrictions…Llama models can’t be used to train other models. And app developers with over 700M monthly users must request a special license from Meta."
techcrunch.com/2024/04/20/this…

#LLMs #Licensing

This entry was edited (5 months ago)
in reply to petersuber

Update. Another #Zuckerberg defense of the #Meta approach to #OpenSource #AI, this time co-authored by Daniel #Ek, the CEO of #Spotify. Zuck and Ek use the case for OS as an argument against #EU regulations of AI.
archive.is/IybwU
in reply to petersuber

Update. The Open Source Initiative (@osi, #OSI) is trying to define what counts as #OpenSource #AI.
simonwillison.net/2024/Aug/27/…

"There is one very notable absence from the definition: while it requires the code and weights be released under an OSI-approved #license, the #TrainingData itself is exempt from that requirement."

in reply to petersuber

Update. More on #OpenWashing and the complexity of defining #OpenSource #AI.
hackernoon.com/is-that-llm-act…

"The Open Source AI Definition (#OSAID) is still open for public review and feedback. If you’d like to participate in shaping the future of Open Source AI, you can submit comments."

in reply to petersuber

Update. New study: "Our extensive experiments reveal that #OpenAccess high-performance #LLMs can be adeptly reverse-aligned to output harmful content, even in the absence of manually curated malicious datasets. Our research acts as a whistleblower for the community, emphasizing the need to pay more attention to safety of open-accessing LLMs."
aclanthology.org/2024.findings…
in reply to petersuber

Update. If #OpenSource #LLMs are vulnerable to hacks that generate harmful content (this thread, prev post), they also "bring several advantages to cybersecurity systems" that can reduce those risks.
venturebeat.com/security/how-o…
in reply to petersuber

Update. HELIOS Open (@heliosopen) comments on v. 0.0.9 of the Open Source Initiative (#osi, @osi) definition of #OpenSource #AI.

"If the definition doesn’t start by emphasizing the openness of training data out of the gate, [we] worry it will not get added in later."

#TrainingData

This entry was edited (1 week ago)