AI Update: Slowing Down AI—The Internet Is Too Small
As I have reported, AI has been on a rampage of growth so fast it is hard to keep up with it. But now comes word that its growth might be slowing due to a completely unexpected source—at least to me. Apparently the internet is too small.1 Who would have guessed that?
The newer more powerful programs require larger quantities of information to learn from. And here I thought the internet was the encyclopedia of everything! Not really, but I certainly did not think it would be a well run dry for AI. But then, the more I thought about it, there are telltale signs that this might be happening.
We have reported on the fact that only public information is reachable by these AI programs. It cannot go behind pay walls. And, as we also reported, companies are becoming more sensitive about information that is reachable by the AI machines. As we noted in our last AI post, the New York Times has sued Microsoft and OpenAI for theft of its proprietary information.2 That could lead to more restrictions.
But, according to the WSJ article cited below, “[m]ost of the data available online is useless for AI training because it contains flaws such as sentence fragments or doesn’t add to a model’s knowledge.”3 These AI “machines” are gaining in knowledge and abilities. It is exciting to watch this all happen. It is truly uncharted territory. Stay tuned for more in the future.
1 Deepa Seetharaman, "For Data-Guzzling AI Companies, The Internet Is Too Small," April 1, 2024; wsj.com/tech/ai/ai-training-data-synthetic-openai-anthropic-9230f8d8?mod=dhjem10point.
2 See The New York Times Company Civil Action No. ________ Complaint Jury Trial Demanded Plaintiff, v. Microsoft Corporation, OpenAI, INC., OpenAI LP, OpenAI GP, LLC, OpenAI, LLC, OpenAI OpCo LLC, OpenAI Global LLC, OAI Corporation, LLC, and OpenAI Holdings, LLC, Defendants; United States District Court Southern District of New York; Filed Dec. 27, 2023.
3 See Note 1. The author suggests that you read the full WSJ article for more context.