Many of yesterday’s talks were peppered with the acronyms you’d expect from this gathering of ambitious panelists: YC, FTC, AI, LLM. But there was an underlying—some might say—enthusiasm for open-source AI.
It was a distinct left turn (or back, if you’re a Linux fan) from the app obsession of the 2010s, when developers seemed content to containerize their technologies and hand them off to larger platforms for distribution.
The event came just two days after Meta CEO Mark Zuckerberg declared that “open source AI is the way forward” and released Llama 3.1, the latest version of Meta’s own open source AI algorithm. As Zuckerberg said in his announcement, some technologists no longer want to be “limited by what Apple lets us build” or face arbitrary rules and fees for apps.
Open AI is simply the OpenAI approach NO using GPT for its largest, despite what the multibillion-dollar startup’s name might suggest. That means at least some of the code is kept secret, and OpenAI doesn’t share the “weights” or parameters of its most powerful AI systems. It also charges fees for access to its technology at the enterprise level.
“As complex AI systems and agent architectures evolve, using small but finely tuned open source models yields significantly better results than [OpenAI] GPT4 or [Google] Gemini. This is especially true for enterprise work,” says Ali Golshan, co-founder and CEO of Gretel.ai, a synthetic data company. (Golshan was not at the YC event.)
“I don’t think it’s OpenAI versus the world or anything like that,” says Dave Yen, who runs a fund called the Orange Collective for successful YC alumni to support early YC founders. “I think it’s about creating fair competition and an environment where startups don’t risk dying the next day if OpenAI changes its pricing models or policies.”
“That doesn’t mean we shouldn’t have safeguards in place,” Yen added, “but we also don’t want to unnecessarily throttle bandwidth.”
Open-source AI models come with some inherent risks that more cautious technologists have warned about – the most obvious being that the technology Is open and free. People with malicious intent are more likely to utilize these tools to cause harm than a costly private AI model. Researchers have pointed out that this cheap and easy so that criminals can learn to ignore any security parameters present in these AI models.
“Open source” is also a myth when it comes to some AI models, as WIRED’s Will Knight reports . The data used to train them may still be kept secret, their licenses may restrict developers from building certain things, and they may still ultimately benefit the original creator of the model more than anyone else.
Some politicians have opposed the unfettered development of AI systems on a huge scale, including California Sen. Scott Wiener. Wiener’s AI Safety and Innovation Act, SB 1047, has stirred controversy in tech circles. It aims to establish standards for AI model developers, who cost more than $100 million to train, requires certain levels of pre-deployment safety testing and red-teaming, protects whistleblowers working in AI labs and gives the state attorney general legal recourse if an AI model causes extreme harm.
A notable speaker at Thursday’s event, added to the lineup at the last minute, was Andrew Ng, co-founder of Coursera, founder of Google Brain, and former chief scientist at Baidu. Ng, like many other attendees, spoke in defense of open-source models.
“This is one of those moments when [it’s determined] “If entrepreneurs are allowed to innovate all the time,” Ng said, “should we spend the money that would go into building software on hiring lawyers?”