A trio of AI researchers and experts in Lithuania and England have tossed some cold water on the hype surrounding generative AI and predict that in 2025, companies and investors will have to lower their expectations regarding what the technology and large language models (LLMs) can do.
One issue is scalability, according to Ali Chaudhry, research fellow at University College London, in some cases, the technology is not as scalable as expected. “I think we will see diminishing returns in the capabilities of LLMs. Some AI labs are already hinting that scaling laws are not as effective anymore,” he said in a statement.
There will be other obstacles such as regulation and sustainability, according to research by Chaudhry, Adi Andrei, co-founder of AI software vendor Technosophics, and Web data gathering firm Oxylabs’ CEO Julius Černiauskas who are members of Oxylabs’ AI/ML Advisory Board.
Increasing regulation and general concerns regarding the dangers of AI are factors that will play into curbing the Gen AI enthusiasm, Chaudhry says. “I am predicting more regulations for controlling the negative impact of Generative AI. Generally, 2025 will be very important for AI Safety, and we will see a lot of work (technical and non-technical) in this space,” he said,
Černiauskas also pointed at the energy consumption concerns around AI as a dampening factor. “Responsible AI and Green AI will become even bigger topics next year as we are more used to the technical capabilities of AI. Servers that support AI development put a strain on the environment, and there are many risks flowing from a lack of transparency in how AI is developed and functions.”
Andrei notes that there is a rising movement of content creators against the Gen AI movement. “This has raised awareness within the general population of the irreconcilable issues posed by technology and the fact that it is being forced onto people by billionaires and their organizations,” Andrei said.
Still, there is still room for optimism. The experts foresee emerging technologies and positive developments manifesting this year. Andrei believes the advent of decentralization technologies, rather than AI, will be worth following.
“2025 will be a year of paradigm changes in the social system, and we will see the emergence of technologies – including information technologies – to facilitate a decentralized way of life and of building community, most likely decentralized social networks, local currencies, and so on,” he said.
Chaudhry foresees more AI contributions for scientific discovery.” “I think 2025 will be a big year for multi-modal models, especially text-to-video models, which will improve significantly in terms of the length of videos, their quality, and their obedience to the laws of physics. We might get some interesting insights into the inner workings of deep neural networks in foundational models,” he said.
And Černiauskas is hopeful for advancements in automated machine learning (AutoML), which opens this field for those without specialized ML expertise by delegating ML development to algorithms.
“AutoML is a crucial step toward democratizing machine learning and AI. With more experts in diverse fields being able to create ML tools tailored to their needs, AI and ML adoption in business can accelerate and open new possibilities. Thus, companies have all the reasons to invest in AutoML, and we might see impressive results already in 2025,” says Černiauskas.
Source:: Network World