The deeper we get into AI, the bigger gets the debate about whether it should be open or closed source. While the open source community wants to take the power away from the tech giants and not let them control it, closed source advocates like OpenAI and Google talk extensively about the dangers around open source. Well, who is right?
Regulators are hell-bent on putting stops around open source AI models. But, the models keep beating the closed source ones on various fronts.
Open Source, the Harbinger of AI Revolution
Early this week, during the US Senate of Intelligence hearing, Meta AI chief Yann LeCun wearing the red bowtie of the National Academy of Engineering, presented his case for open source. “AI is going to become a common platform, and because it’s a common platform, it needs to be open source if you want it to be a platform on top of which a whole ecosystem can be built,” said LeCun.
He gives the example of the infrastructure of the internet. “It didn’t start out as open source, but as commercial. But then the open source platforms won because they are more secure, easy to customise, and safer.” He also points out how open source is necessary for the larger world. Unlike Silicon Valley, where information flows extremely fast because of a close-knit ecosystem, other countries in Europe still have to work in silos.
This is definitely true for a lot of companies that are focusing on building native AI products. Regardless of the ease of use and simplicity, the closed source AI products cannot guarantee privacy and security. Take the case of Stable Diffusion and Llama 2, even though Midjourney, DALL-E or GPT offer great usage in the easiest ways, they are still not as controllable as the former duo, which are indeed open source.
Is there a hassle to build on top of open source products? Yes, for sure. But is it going to be beneficial for the companies? Without a doubt!
Open Source: For or Against
Every business needs their core AI products. Arguably, if a company is building one of these products that is relying on AI, the first step should be to opt for open source and build it from scratch. On the other hand, using GPT wrappers means outsourcing of core business, often based on confidential data, which isn’t a desirable long-term strategy for companies.
One common argument against open source AI is that it can’t compete with the vast resources of industry labs. Building foundational AI models is undeniably expensive, and it’s believed that only well-capitalised teams can produce groundbreaking results.
Another argument against open source AI is its supposed inability to reason. It’s often claimed that open source models perform poorly on benchmarks and lack emergent capabilities required for complex tasks. However, this argument overlooks the fact that reasoning doesn’t matter for a majority of AI use cases.
Most users and developers don’t have a pressing need for advanced reasoning capabilities. Open source models are highly proficient at handling the most valuable and common tasks, such as summarisation and ELI5, and with fine-tuning and adequate labelled data, they can cover a vast majority of use cases effectively.
Moreover, even closed source models are bound to get leaked too. All the alignment that the creators argue for is going to go down the drain.
Blaze Your Open Source Glory
Last month, the frontrunner of the generative AI revolution, OpenAI, raised the alarm over open source AI dangers. “An important test for humanity will be whether we can collectively decide not to open source LLMs that can reliably survive and spread on their own. Once spreading, LLMs will be up to all kinds of crime, it’ll be hard to catch all copies, and we’ll fight over who’s responsible,” said Jan Leike, from the alignment team of OpenAI.
It’s true that there are ethical implications about generative AI, but that applies to everyone—both closed source and open source. The best way to go forward, just like any other technology, is to let it go its way without putting shackles around it.
