Will Large Language Models end up like the Metaverse?

It is widely known in my circles that I am “The guy that is interested in AI” (A term which I reject, as I stated in my post “Geniş Dil Modelleri [sic] Yapay Zeka Anlamına Gelmez”), and some have raised their concerns that my views on the the field of artificial intelligence, just like on the metaverse, is flawed and wrong -that I am wasting time on a temporary trend that will not catch on, just like the metaverse-. This posts analyses why they are both right, and wrong.

To begin understanding the situation, we must first understand the context. Large Language Models, unlike the metaverse, is not an endpoint, no one knowledgable in the field claims its the final evolution of AI systems. To understand what I am discussing, we are to look at a Podcast. In a Podcast (Yann Lecun: We Won’t Reach AGI by Scaling up LLMS) with Alex Kantrowitz, Yann LeCun (Who is the Meta Chief AI Scientist, the company well known for having tried to establish “The Metaverse”) states what Intelligent animals do and Large Language Models cannot do, which are; understanding the physical world, having persistent memory, being able to reason and being able to plan.

I as an end user agree with his statement, whereas I think Large Language Models try imitating a human, it does not faithfully act like a human. This contradicts my earlier view that “A Large Language Model can do everything a human can”, it does try imitating a human being, and as a result gets most things right, but this required correct prompting, context injection, and special handling that a single model file cannot solve, but a program largely can. We are still having difficulty getting smarter large language models, such as omini models right (Qwen is recently working on those as the time of writing this article), as humanity.

Therefore -in my opinion at least- in the podcast, Yann LeCun rightfully warns against a hyper-fixation for researchers and professionals within the field for Large Language Models, as it simply is not an endpoint, but a step towards a smarter artificial intelligence. It will be a stepping point, but being fixed in the field would be disastrous for any individual.

In the previous section, I emphasized on a transitionary stage, but how is this relevant to the fact that Large Language Models are in fact, not a temporary trend? How do we know Large Language Models wont explode like the dot com bubble? We don’t, and I do think we are headed towards the dot com bubble exploding as many companies are found solely on utilizing large language models alone. There are just too many “solutions” on the market, most being recent as technology (Large Language Models) started to be more stable and predictable, as people learnt how to use the technology on proven and stable methods, most people fail to realize that their solutions are not built on something stable, but built on something fragile -a rapidly evolving technology that is transitionary-, and when the next development, next iteration of technology occurs, all these companies will suddenly explode and struggle to adapt, causing a market crash in the process.

So, then the article should end right here, as the people warning me are right, that like my investment on the Metaverse, Large Language Models are doomed? No, not at all. Here comes the Metaverse. Metaverse is a forcefully invented term, it is not recent, platforms like it have always existed such as Second Life, ROBLOX, all before mid 2010s, it is a corporate terminology, invented for the market. Metaverse in essence is a virtual world with e-commerce for real life built in, unlike normal virtual worlds where you earn and spend virtual currency, Metaverse involves real currency, for virtual things (Which I always have criticized despite my hype for the Metaverse)

In mid-late 2010’s, especially early 2020s, we saw most of these virtual worlds evolve into “Metaverse” entities, doomed to fail as a result of their corporate nature, people work, regular people do work hard, they get tired in the process, they earn their livelihoods, they earn they money to make their ends meet. A regular person, by how the Metaverse was presented, would -and did see it- as a dystopia, a virtual world where you enter unwillingly, “own things” digitally (where you actually don’t, you rent the right to use, usually for personal use too), being “happy”.

Metaverse did not claim to be a transitionary state, it tried to show itself as the ultimate endpoint of user experience in social media, virtual world, and e-commerce, and a dystopian one at it. As for me, my personal opinion on the Metaverse itself is simple; I support the existence of personal, de-centralized metaverse entities, where each individual can enter an immersive metaverse of a theme of their choice and be immersed in a positive way, where they can socially interact, express themselves, and order things physically to their door. This seems hard to implement at this stage due to the public perception, therefore my thoughts are purely a part of my own utopia of thoughts.

Now that we analyzed the both in depth, the reader should already see that the two fields simply don’t end up similarly, one is an endpoint, the other is a phase. This simply is not the “only divergence point” however (if we fall into the fallacy of comparing both), as how they started are also different. As a part of developing technology, Large Language Models started off in a paper called “Attention is All You Need”, addressing previous techniques such as RNN (Recursive Neural Networks). Metaverse on the other hand as aforementioned were invented for the market. One simply cannot compare natural process of multiple researchers to a an individual corporate decision, it would be a gross injustice.

As Conclusion, people are both rightfully worried and gravely wrong in this matter. People should be worried, that hyper fixating on Large Language Models and missing out new developments, or jumping on the trend directly, will negatively affect the individual and society, but comparing it into the Metaverse is simply plain out wrong, not because of my personal opinion, but because how they developed and meant to conclude. Large Language Models are currently one of the best AI tools we have, and we must utilize the most effective tools we have while looking out for better methods to come.