In the ever-intensifying race of generative AI, often dubbed the “AI war,” major tech companies are vying for dominance. Microsoft appears to be leading the charge with a formidable lineup of generative AI-powered offerings like Bing Chat and Image Creator. Not far behind, Google is steadily advancing, although it has yet to introduce a standalone generative AI product or fully integrate the technology into its flagship, Google Search.
However, a senior software engineer at Google boldly asserts that the true contender to be reckoned with is not Microsoft or OpenAI but rather the power of open-source solutions.
Luke Sernau, Founder of Better Engineering and Senior Software Engineer at Google, Highlights Open Source as the Dark Horse in the AI Arms Race
Luke Sernau, a prominent figure within Google and the founder of Better Engineering, has expressed his belief that neither Google nor OpenAI currently possess the upper hand in the race for AI supremacy. According to Sernau, Google’s rivalry with OpenAI has diverted the company’s attention away from the significant advancements being made in the realm of open-source technology. In a thought-provoking document, Sernau warns, “While we’ve been embroiled in internal disputes, a silent contender has emerged to take the lead. I’m referring, of course, to open source.”
Sernau’s remarks were included in a document that surfaced on an internal Google internet system in early April. The document quickly gained traction among Google employees and has reportedly been shared extensively, according to an individual familiar with the situation. On May 4, consulting firm SemiAnalysis published the document, causing it to circulate within the Silicon Valley tech community.
When it comes to large language models, Meta’s LLaMA has emerged as the favored choice within the open-source community. Released in February, LLaMA is touted by Meta to surpass GPT-3 in various tasks, including natural language processing and sentiment analysis. What sets it apart is its high adaptability, allowing for customizable weights that enable it to operate efficiently on less powerful hardware. This characteristic makes LLaMA particularly appealing to developers.
However, Sernau acknowledges that LLaMA is not the sole developer-friendly large language model available. He notes, “While our models still maintain a slight advantage in terms of quality, the gap is rapidly narrowing. Open-source models are exhibiting remarkable speed, enhanced customization, increased privacy, and an impressive level of capability. They are accomplishing tasks with $100 and 13B parameters that we struggle with at $10M and 540B. Moreover, they achieve these feats within weeks, not months.”
“We have no secret sauce,” Sernau wrote. “Our best hope is to learn from and collaborate with what others are doing outside Google.”
He also suggested that clients would be unwilling to pay for restrictive models when other high-quality models are available for free.