Fact Checked by 
Google’s CEO, Sundar Pichai, has predicted the future of artificial intelligence. Speaking at The New York Times Dealbook event, Pichai said that AI development will begin to slow down.
“I think the progress is going to get harder when I look at [2025]. The low-hanging fruit is gone,” Pichai said during his talk.
Right now, most AI development houses will release a general model and then begin to work on a more specific version. For example, OpenAI, makers of ChatGPT, have ChatGPT 4. This has been spun to 4o, which in turn is split into “Mini” (a cheaper-to-run version) and is now being succeeded by o1.
Each new model comes with billions of instructions to work off, which Pichai believes won’t be the way forward. Going bigger in terms of ability is no longer the fight.
The next thing on Pichai’s radar is hitting “deeper breakthroughs”, which he doesn’t see as a wall, “So you can perceive it as there’s a wall, or there’s some small barriers.”
One of these barriers is the potential for AI-generated data to become the training data, which would poison the well. The internet is estimated to be 57% AI-generated in 2024, which isn’t good data to harvest.
AI is trained on billions of pieces of data, sourced from media, books, and the internet. As companies are beginning to run out of material, it is resulting in a plateauing effect as large language models like ChatGPT receive updates.
These large leaps between models seen from GPT 3.5 to 4 will stop becoming so common as fresh data available dries up.
To get around this, companies have made deals to get a direct source of new content. such OpenAI has partnered with publishers like Future. Google itself has made a $60 million deal to harvest Reddit data for training purposes.
AI industry facing plateau as training begins to slow down
AI and its industry are currently in a rat race to be the best. However, its major promises like artificial general intelligence (AGI), where a computer could theoretically mimic a human brain, are being downplayed by leaders in the industry.
As the next advertised major hurdle of AI, OpenAI CEO Sam Altman has been lowering expectations recently. While he predicts it’ll launch in 2025, at the same summit, Altman said, “My guess is we will hit AGI sooner than most people in the world think and it will matter much less.”
OpenAI has seen multiple leaders from the company begin to leave, as it pushes ahead with plans to become a for-profit company.
Pichai did hit back at Microsoft, one of the most important companies in the AI business at the moment. The tech giant is currently working on nuclear plants to power its data centers for AI but doesn’t actually have their own model for the public.
Instead, Microsoft has a 49% stake in OpenAI and uses various wrappers to present the ChatGPT-powered CoPilot in its software. Poking at Microsoft, Pichai said, “I would love to do a side-by-side comparison of Microsoft’s own models and our models any day, any time… They’re using someone else’s models.”
Featured image: New York Times, Google
