That artificial intelligence is being a technological evolution comparable to the Internet is something that has already been demonstrated on numerous occasions (and what remains). Tools as disruptive as ChatGPT caught several large technology companies by surprise, including Google. Although Microsoft raised capital to invest in OpenAI and base its products and services on its technology, Google had to make us wait a few months until you have your product ready, something that has given you an advantage over your competitors.
Sometimes it is surprising that such a radical change has not been led by a company like Google. Those from Mountain View have accustomed us to making our lives easier with their search engine and other products and services based on the large amount of information they have. When Google CEO Sundar Pichai warned his employees of a “code red” to focus even more on their AI technologies, we already knew that the company was late.
Google engineers already predicted the evolution of AI
Another demonstration of the fact that AI has caught Google in its infancy is The document that could be extracted from the trial against the company regarding antitrust lawsuits. This document dates back to 2018 and a Google engineer warned that deep learning technologies could put the company in a bind.
“In the near future, a deep ML (deep autonomous learning) system will clearly outperform the relevance accumulation algorithms for web search that Google has been working on for 20 years. Here, I’m just talking about relevance; that is, determining if a document and a query refer to the same thing. There is much more to web classification that an ML seems much less appropriate for. But I think that basic relevance is the main task of web classification and probably “objective” enough as for an ML”.
BERT was the language model created and developed by Google in 2018
As we can see in the leaked document, in 2018 there were already engineers at Google who predicted that an autonomous learning system would far surpass web search algorithms in which Google has been working for more than 25 years. In the case of Eric Lehman, this was what he thought of the company at the time, and indeed it is what seems to be happening almost six years after his words.
“None of us can see the future, but my bet is that this will almost certainly be true within 5 years and could be true even within 6 months. All the web ranking-like problems no longer matter, and there is little reason to think “That web ranking is something exceptional. In fact, this thinking comes from recent advances in web responses, where deep ML (in the form of BERT) abruptly subsumed all previous work.”
It is no surprise that Google had been working on various deep learning systems and language models similar to GPT for years. However, the fact that they did not have a product to show to the public in a solid way caused the firm to lose months of advantage in the race for the development of generative artificial intelligence. Here Lehman talked about the capabilities of BERT, the ML developed by Google and its potential ability to replace all the work Google had done.
“For the web response team, the wave of deep ML that arrived in recent weeks was a complete shock. With this warning, we should not be caught off guard again, but should start thinking about the implications. And now “It’s really time, because in the new year I hope that many web classification engineers will reflect on BERT and start thinking along these same lines.”
“The risk that Google could be beaten by another company in relevance is highlighted by a surprising finding from BERT: the enormous amount of user feedback can be largely replaced by unsupervised learning from raw text. This could have implications for Google.”
Of course, Lehman’s high capacity in this context to predict what was coming to Google is surprising. And his engineering team realized long before the capabilities of autonomous learning from Google’s own internal projects. The company seemed to have the capacity to launch an ML-based service similar to Bard or Gemini, but decided to err on the side of conservativeness. This could be very expensive for the company, although although in this field the one who starts first often wins, It wouldn’t surprise anyone if Google took the lead in the future.
With Gemini, Google wants to get off to a good start when it comes to offering products based on the operation of generative artificial intelligence technologies. Through its scalable model for all types of devices, the company has made notable progress in its efforts to offer a solid product. However, OpenAI still has the upper hand with its powerful language model.
Image | Google
In Genbeta | Apple launches its own artificial intelligence to edit images with text: it is called MGIE, it is open source and you can now try it