Open-Knowledge source with an Open-Inteligence that can guide you through the entire massive digital library of its own brain. Semantic data light-years beyond a Search Engine.
Kagi’s LLM can already do that. I believe so can Perplexity’s. Citing sources isn’t something only open models can do.
Sure. The point is the thing you said only an open-source model can do, it can do. Plenty of proprietary LLMs can cite sources.
The plain truth is most of the benefits of open models are not on the consumer side. (Or at least, I haven't seen any articulated.) They're on the producers'. Open models are better for those of us training models. That's partly why the open data debate is academic--very few people are training large foundation models because the compute and electricity costs are prohibitive.
Genuinely blown away the EU isn't doing this.
In the U.S., the solution may be in carving a legal safe harbor for companies that release their models per the OSI's draft definition of open source.
This sounds like the usual AI marketing with the word "open" thrown in. It's not articulating something youc an only do with an open source LLM (and doesn't define what that means).
I'm personally not thrilled with how locked down LLMs are. But we'll need to do a better job at articulating (a) a definition and (b) the benefits of adhering to it versus the "you can run it on your own metal" definition Facebook is promulgating. Because a model meeting Facebook's definition has obvious benefits over proprietary models run on someone else's servers.
I believe our world fights to destroy ideas like this because our economy drives our entire life.
>> No
>>> You can't imagine it
You haven’t articulated the idea you claim the “world fights to destroy”. (Just throwing around the word open without elaboration isn’t an idea.)
We need dramatic reforms on copyright, as we’ve really let corporate interests crowd out our rights to human culture and ideas. While I alone cannot decide what we as a country should find reasonable, I can say I find 20 years + 5 years extension is perfectly reasonable and that corporations should have never been able to pay off politicians to get what they wanted. Let alone Sonny Bono, that bastard, signing in bills that specifically benefited him.
So, to reiterate, the idea I feel that corporations want to destroy is the idea that we, as a people, have rights to the works that form our popular culture and that no one man, let alone a faceless corporation, should be able to profit from a singular work for hundreds of years.
No. But you haven’t articulated why making everyone’s Facebook chats public is a net good. What does opening that data up confer in practical benefits?
Given what we know about LLMs, one trained only on public-domain data will underperform one trained on that plus proprietary data. If you want source data available, you have to either concede the "open" models will be structurally handicapped or that all data must be public.
Facebook says no, at least for Llama [1].
[1] https://itlogs.com/facebook-uses-user-data-to-train-ai-but-l...
Also just to be clear, if you want to set up a RAG with an open weight model and a large dataset there's nothing stopping you. Download Red Pajama and Llama and give it a try.
OTOH why you want the data is not clear. You don't need it to run Meta'a models for free, or to fine-tune them for your own needs. The only thing the data would allow you is to pre-train from scratch, in other words to obtain the exact same set of weights that Meta is giving you for free.
My company is building an application for an university client, regarding the examination of research data written in "human language" (mostly notes and docs).
Due the high confidentiality of the subjects, as often they deal with non-patented information, we couldn't risk using frontier models, as it could break the novelty of the invention, therefore losing patentability.
Now with Llama3.1, we can simply run these models locally, on systems that is not even connected to the internet. LLMs are mostly good in examining massive amount of research papers and information, at least for the application we are aiming at, saving thousands of hours of tiresome (and very boring) human labour.
I am trying to endorse Meta or Zuckerberg or anything like that, but at least in this aspect, I think Llama being "open-source" is a very good aspect.