Why 'open' AI might be more marketing than reality

Mark Zuckerberg's AI strategy it is all about openness transparency collaboration democratization this technology is so important and the opportunities are so great that we should open source and make it as widely available as we responsibly can so way everyone can benefit he wants to open up AI to everyone but should we trust him and just how open is his AI anyway this week on Tac check we are breaking down the myth of open Source while chat gbt is quickly becoming a household name meta's version of AI brings to mind a tall woolly long- necked animal but among developers those working in the AI trenches meta's open source model called llama is the hottest thing around meta is turning into uh into the the needing uh open source AI big big technology company um and it's really uh paying dividends in in the Public Image Wall Street is also waking up to meta's AI proposition it stock over the last year has well outperformed Google and microsofts but calling meta's AI system open source is only part of the story we're here to break down the promise and the myth in its purest most simplified definition open source is where the source or the underlying code is freely available to access copy modify redistribute closed software on the flip side means that the source code is only accessible to its creators and fully controlled by them what data was used to train Sora we used publicly available data and licensed data so videos on YouTube I'm actually not sure about that okay videos from Facebook Instagram you know if they were publicly available um available yeah publicly available to use um there might be the data but um I'm I'm not sure I'm not confident about it she can't or she won't tell us and because the model is closed there's no way of getting that information so if the chatbot is biased if it hallucinates or if it spits out misinformation developers cannot go back and find out how it happened another key component of building and evaluating an AI model are its weights essentially how much value the model assigns to piece of its training data there's also all the training data itself a huge amount of data that is supposed to be transparent and trackable in open source models letting researchers and users understand how a model works unlike open- Source software the weights and training make it so generative AI models require huge amounts of compute power to build which in turn requires huge amounts of capital it's why so far the largest players in the space have been confined to those with the most money most of them only release enclosed models when it comes to most of the progress made in AI today it has come from those Clos Source models this is companies like open AI with their GPT Gemini from Google anthropic but a select few are marketing themselves as open source the biggest meta llama 2 Google's new Gemma and the Mis large proponents of open source include some of Tech's most prominent figures like meta's Mark Zuckerberg and Tesla's Elon Musk the name opening eye refers to open source um the intent was what's the okay so what was the opposite what's the opposite of um of Google would would be a an open- Source nonprofit because Google is closed sourced for profit um and that profit motivation can be potentially dangerous musk and Advocates of that philosophy they say that open source is crucial to preventing the concentration of power in the hands of the few and the richest and that transparency is critical to making sure the systems are safer and better when push comes to shove let's say they do create some some digital super intelligence almost Godlike intelligence well who's in control those building Clos Source models on the other hand they say they offer better quality control and money-making opportunities to State the obvious there are near-term commercial incentives against open source but there is another longer term argument against open sourcing as well when the capability is on the lower end I think open sourcing is a great great thing but at some point the capability will become so vast that it will be obviously irresponsible to open source mods now the two sides they may sound like they are diametrically opposed except that some of those open- Source models may not actually be as open as they say the biggest open source model out there right now it is meta llama 2 right now they have the leading open- Source AI model in the world that's llama 2 I think it's downloaded something like 30 million times in the last couple of months the company says tens of thousands of startups they were built upon it including Buzzy ones like snowflakes scale AI door Dash meta benefits too when a model is open source any developer can go in and fine-tune it also adding those improvements to the model itself essentially free labor for the benefit of all meta says that has improved llama's performance by nearly 10% but it's also good PR now Mark Zuckerberg can tell investors things like our long-standing strategy has been to build an open- Source General infrastructure while keeping our specific product implementation proprietary the catch is llama was only publicly released because of a leak it was initially only available to researchers By Invitation less than a week later someone posted the full model online for anyone to download and the dirty truth critics say that llama's not actually open source the open source initiative founded in 1998 to support the open source ecosystem writes that unfortunately the tech giant has created the misunderstanding that llama 2 is open source it is not the biggest issue is who gets to use it open source systems should in theory be free for anyone to use but meta restricts its licensing dictating that you can't use llama to improve any other large language model and that you have to request a license if your company has more than 700 million monthly active users some analysts also predict the model won't even stay free for long and I'd put this probably 3 years out where we we'll likely start to hear more about this is the opportunity for them to start to charge for it Monster believes that eventually meta could build a cloud service that the model runs on and charged to use it I.E closing it to non-paying customers Mark Zuckerberg is not doing this uh for the good of humanity uh he is advancing this because he sees the Power of artificial intelligence and the opportunity to build a big business around it meta is not the only company known for being open sourced that doesn't seem fully committed to it mistal is the biggest open- source AI company in the startup World AI startup mistol is Raising nearly $500 million in a funding round that includes Nvidia and Salesforce investors but their most recent and most powerful model called the mistra large that's closed in the case of maestro what they've announced is working with Microsoft and that's actually a closed product that's very different they're maintaining their open uh product they're doing both of it they're doing both touting itself as an open generative AI company while at the same time closing off their most Cutting Edge technology researchers in the paper open for business they say that most open- Source AI it's just marketing and they write that some companies are embracing open AI as a way to entrench their own dominance letting them set standards of development while benefiting from the Free Labor of Open Source contributors some of the most successful open- Source models will eventually want to find ways to make money for it here's another example musk has not been shy about his belief in open- Source AI Elon Musk posting on X the following uh that this week his AI startup xai will open- Source grock musk unveiled his AI platform grock back in November of last year and said it can access X formly Twitter in real time he's sued open AI of course now and Sam mult and alleging they broke the company's funding agreement by seeking profit and criticize them for not being open but a trove of emails that open aai published in defense seems to suggest that musk is only an advocate in public in private he wrote that open aai was burning cash that the funding model couldn't reach scale and that a for-profit pivot might create a more sustainable Revenue stream when one open AI executive told musk as we get closer to building AI it will make sense to to start being less open the open and open AI means that everyone should benefit from the fruits of AI after it's built but it's totally okay to not share the science Elon simply replied yep even if the biggest tech companies aren't truly buying in the community of developers and users that are passionate about the open source cause has been around long before them and they will stay long after there is I think will be a vibrant open-source Community this is just a natural part of what happens when there's a paradigm shift and I think it will be a smaller part relative to the paid for Community but if for those developers who want to have that ability to customize and not pay as much they're going to have plenty of options in the decade to come what is clear is that separating truly open- Source AI from just marketing that will have a huge commercial impact shape policy and determine whether the Revolutionary Tech will stay in the hands of the few or the many [Music] sh

Channel: CNBC Television

Views: 66,233

Keywords: app, business news, cnbc, digital, disrupt, funding, innovation, investors, nasdaq, nyse, online, silicon valley, startup, stock market, tech, techcheck, technology, venture, wall street

Id: tpZ5mKByEfE

Length: 10min 18sec (618 seconds)

Published: Fri Mar 15 2024