Can we put all our eggs in one basket? It is too risky. So far, Silicon Valley focused on just one way to create AI. There is over-reliance on data centers to train and run LLMS — large language models. This technology uses the technique of predicting tokens that appear in a sequence.
The investment is productive. AI chatbots keep growing and startups in this space are breaking even. Businesses adopting AI are in the initial stages and are using generative AI. LLMS are the pioneers in mainstream adoption — 700 plus millions use ChatGPT every week.
Of course, full dependence on a single technique is flawed since as soon as the landscape shifts suddenly, the business falters. Blackberry was obsessed with a physical keyboard, and Apple overtook it by a virtual keyboard. Yahoo relied on portals, whereas the rival Google focused on search engines.
A novel approach to AI could put to risk the capital being deployed to chatbot technology. A smaller tweak tried by DeepSeek of China could shake up the conventional approach of the market.
The area Covariant is exploring software that makes machines perceive the surrounding environment, rather than simply detecting the patterns in data. Companies study robotics and drones, drug discovery, climate modeling. They are away from LLMS.
Google’s DeepMind experimented with different approaches to build super intelligence. AlphaGo, RL were used before shifting focus on LLMS. There could be diminishing returns in LLMS. Of course, LLMS are evolving, and before models do appear. Still, hallucinations persist. Social reasoning abilities of LLMS depend on small set of model features. It raises the issue of reliability.
Yan Lecun, Chief AI scientist at Facebook, has long argued that LLMS are a ‘dead end’ for smarter machines. They lack the understanding of the physical surroundings and are not able to plan the future. They are just ‘token generators.’
LLMS would not disappear, but we cannot afford to be obsessed with a single solution. Investors and businesses must remain alive to different breakthroughs and should be ready for the shifts.
Leave a Reply