Artificial intelligence (AI) may hit a dead end thanks to data shortages and technological limitations, according to a report of Bloomberg.
It’s been almost three years since generative AI burst onto the scene in the form of chatbots, AI image generators and music generators. Such a big leap forward has left many wondering what’s next.
So have excited tech companies and their shareholders Bloomberg reports that Google, OpenAI and Anthropic are all struggling to build more advanced AI.
According to Bloomberg’s sources, the model OpenAI is working on, Orion, is not meeting internal expectations. Likewise, Google’s latest iteration of Gemini isn’t much better than the previous one either. Anthropic also delayed the release of its Claude model.
One of the reasons cited is that “it has become increasingly difficult to find new, untapped sources of high-quality, human-made training data that can be used to build more advanced AI systems.”
It is truly fascinating. It is well known that AI companies have raked virtually all available data across the open web to build various models. It’s safe to bet that almost all photos are taken online for AI training purposes.
This is something I wrote about in June 2023 in which I pointed out that AI image generators cannot survive without fresh photography.
Bloomberg says that with all the open web scraped, tech companies are finding it hard to fill the gap. Some of them are turning to AI imaging, but researchers have found that this method has limitations. One study revealed this AI trained on computer-generated material turns to mush.
All this puts a damper on the dream of artificial general intelligence (AGI). It refers to hypothetical AI systems that are more intelligent than humans. OpenAI and Anthropic have both previously stated that AGI is close.
“The AGI bubble is bursting a bit,” Margaret Mitchell, chief ethics scientist at AI startup Hugging Face, tells me. Bloomberg. Mitchells says that AI companies will need to take “different training approaches” to achieve AGI.
“It’s less about quantity and more about quality and diversity of data,” adds Lila Tretikov, head of AI strategy at New Enterprise Associates and former deputy chief technology officer at Microsoft. “We can generate quant synthetically, but we struggle to get high-quality unique data sets without human guidance.”