Majority of AI Researchers Say Tech Industry Is Pouring Billions Into a Dead End

You can just throw a lot of money in a problem.

This, more or less, is the line that artificial intelligence researchers take in A. Conversation. When asked whether the “expansion” in the current artificial intelligence methods could lead to the achievement of artificial general intelligence (AGI), or a general purpose of artificial intelligence that matches or exceeds human perception, 76 percent of the respondents said it is “unlikely” or “very unlikely”.

Posted in A. New reportThe results of the survey, which inquired from 475 researchers of artificial intelligence and conducted by scientists at the Association for the I intelligence service, provided a resounding refusal on the way the long technology manufacturer of achieving the gains of Amnesty International-by providing obstetric models, and data centers used to train them, with more devices. Given that AGI is what the developers of artificial intelligence claim to be their final game, it is safe to say that scaling is widely seen as a dead end.

“The vast investment in scaling, unaccompanied with any efforts similar to understanding what was going on, always looked to be in its place,” Stewart Russell, the computer world at the University of California in Berkeley helped organize the report, He said Newscientist. “I think, about a year ago, it started clear to everyone that the benefits of scaling in the traditional sense have fallen.”

Huge sums of money are thrown into the intelligence arms race. Artificial Intelligence Investment reached more than $ 56 billion in financing investment capital alone in 2024, Techcrunch I mentioned. Many of this are spent to build or operate the huge databases required by obstetric models. Microsoft, for example, has He is committed to spending $ 80 billion On the infrastructure of Amnesty International in 2025.

It follows that the demand for energy is amazing. Microsoft signed a deal for Launch an entire nuclear power plant Only to operate its data centers, with its competitors Google and Amazon Also get rid of wonderful nuclear power deals.

The hypothesis that artificial intelligence could be improved indefinitely by scaling was always on a shaky ground. For example, the recent existential crisis in the technology sector that the Chinese company Deepseek, which can exceed the artificial intelligence model from the soles of the soles of the leading chattings in the West, and billions of dollars in a small part of the cost of training and strength.

Of course, it was writing On the wall Before that. In November of last year, reports indicated that Openai researchers discovered that the next version of its large GPT language model A much less improvement offerIn some cases, there are absolutely no improvements from previous versions on their predecessors.

In December, the CEO of Google Sundar Pichai Go in the record By saying that Amnesty International’s easy gains have ended, “but stressed with confidence that there is no reason that the industry could not” continue to expand. “

Cheaper and more efficient methods are explored. Openai uses a method known as COMPUTE Test time with its latest models, where artificial intelligence spends more time to “think” before choosing the most promising solution. This has achieved a performance boost that would take the scaling mountains to repeat it.

But this approach “is unlikely to be a silver bullet,” Kerfind Narianan, the computer world at Princeton University, told the University of Princeton, Newscientist.

Meanwhile, Dibsic was a pioneer in an approach called “A mixture of experts“Which enhances multiple nervous networks, each specialist in various fields – the ideal“ experts ” – to help reach solutions, rather than relying on a“ general ”model.

However, if Microsoft’s commitment to spending tens of billions of dollars to databases is any indicator, then the brute force will remain the favorite of the mighty industry – while it will be left to the most famous startups to avoid doing more with less.

More about artificial intelligence: China announces all artificial artificial intelligence materials online