The future of AI is not in the hands of Google or OpenAI.

That's an assessment taken from a recently leaked internal memo from the search giant. In the same document, Google executives expressed deep concern about the possibility of failing in the ongoing AI race.

A Google representative later confirmed the document's authenticity, stating that it was merely an opinion from someone within the company, but acknowledged they "are not in a position to win this race, and neither is OpenAI."

The search giant believes that the open-source community is beating leading AI companies, even though their products have a slight advantage in quality, but "the gap is narrowing incredibly quickly."

Capital size is not the core factor.

AI has been creating a buzz among the public since late 2022 with the launch of ChatGPT, a chatbot based on LLM (large language models) developed by OpenAI – a startup closely associated with Microsoft. The success of this product has prompted Google and many other tech companies to jump into the LLM race.

LLM models are trained using trillions of words retrieved from the internet to generate text or conversations that resemble human speech. This process takes months and costs tens of millions of dollars, leading to concerns that AI will be dominated by wealthy technology companies.

AI developed by the open-source community is becoming a real challenger to industry giants like Google and OpenAI.

However, internal Google documents suggest that this assumption is flawed. Researchers in the open-source community are leveraging free online resources and achieving results comparable to even the largest proprietary models today.

Accordingly, LLMs can be "tuned" through a technique called low-level adaptation, or LoRa, which allows for optimizing an existing LLM for a specific task quickly and at a much lower cost than training an LLM from scratch.

For example, in March, LLaMa, a model created by Meta—Facebook's parent company—was leaked online. Although it was small and quite "primitive" with only 7 billion parameters, compared to 540 billion parameters in Google's largest LLM, LLaMa was quickly refined to produce results comparable to the original version of ChatGPT on some tasks.

This could have a seismic impact on the future of AI development as "the barriers to training and testing, which previously required a large research center, have been reduced to just one person, one evening, and a high-performance laptop."

The battle between the giant Goliath and the shepherd David

Now, an LLM can be fine-tuned for just $100 in a few hours. With its flexible, interconnected, and low-cost model, "open-source AI has a significant advantage" that large companies like Google cannot replicate. Therefore, the document concludes that Google or OpenAI are vulnerable to these open-source competitors.

The open-source community has advantages that tech giants cannot replicate.

“I don’t think I need something as powerful as ChatGPT-4 for many tasks,” Simon Willison, a programmer and technology analyst, told NBC. “The question now is how much can models be broken down and still be useful? And that’s something the open-source community is exploring very quickly.”

This is entirely consistent with the opinion of many experts that the greatest strength can also be the greatest weakness if exploited and observed correctly, just like in the story of the giant Goliath being defeated by the small shepherd boy David with just a slingshot.

However, when it comes to the development of AI, not everyone may agree with this point. The reality is that proprietary, paid software still exists, such as Adobe Photoshop or Microsoft Windows, and the field of AI may well follow a similar path of development.

(According to Economist, YahooFinance)

Source