== Leo Qin ==

Why does AI make me feel bad?


By any objective measure, I should be super excited about advances in AI and its recent entrance into ubiquity. However, every time I hear about the latest GPT iteration, the latest startup that uses generative LLMs to revolutionize XYZ industry, I just feel bad.

Here’s a bulleted list (in no particular order) of all the reasons why I think AI makes me feel bad:

  • In a vague science-fictional way I’ve always expected “Artificial Intelligence” to be more neuro-mimetic. However its current form is more like a stochastic parrot
  • It feels artificially limiting to rely on models that generate written language or code. Why not skip a step and generate executables or binaries? If we’re truly ambivalent to the internal logic of these models, why not bypass the toolchain entirely?
  • Capitalism is capitalism-ing - I don’t have an enormous amount of trust that OpenAI’s idiosyncractic governance model is capable of hedging the aggressive monopolist tendencies of shareholder-driven businesses (prove me wrong, Microsoft)
  • The gold rush mentality will kill good, potentially profitable, but not hyper-profitable ideas. This happened to crypto-currencies; every good idea got financialized to the point where it just became arbitrage, and there is nothing more banal than arbitrage.
  • Speaking of crypto-currencies, excessive financialization lead every crypto-currency to eventually approximate a credit card payment network in function, but the “function” also happened to be wildly volatile. I think that all LLM applications are going to approximate either chatbots or no-code frameworks. This is not an inherent limit of LLMs, but simply of our own imagination and the gold rush mentality.
  • The ability of LLMs to confidently hallucinate pushes us ever-deeper into humanity’s post-truth era. Every business that seeks to capitalize on LLMs has a vested interest in NOT correcting these hallucinations.
  • In the medium-to-long term, using LLMs to do SEO positioning (the most basic and stomach-churning dimension of “growth-hacking”) will destroy the ability for search engines to return meaningfully reliable and truthful results, and LLMs will use this data to eventually poison their own training datasets with true sounding but false content that enforces a particular LLM-like style of speech. That will not stop anybody from using them.
  • Just as how there is an externalized cost to low quality but inexpensive consumer goods, there are absolutely unrealized externalized costs to low quality inexpensive digital goods.
  • To the degree that hustle culture is a driver of rather than a consequence of capitalism, the current crop of “prompt-engineering” grifters that want to sell you a course to teach you about “all the newest AI tools that you should be using right now” is such a disappointing replay of the same mentality in crypto-currencies - “With these 10 cloud services hosted on AWS, you too can make your own NFT that represents an immutable, permanent, and self-sovereign chain of ownership”.

To sum up - anyone who believes in the importance of truth and human freedom should view universal and uncritical use of LLMs as an existential threat to those same values. This is because social networks and search engines shape our perception of reality, and there is nothing more capitalist than achieving completely control of reality. That’s the game, and that’s why every social network and search engine wants to run its own LLM platform.

With the benefit of hindsight, we can see that crypto-currencies were never going to achieve anything close to this level of control, and I hope that we eventually get to that same space with LLMs.

While I cannot yet fully articulate a meaningful alternative to this downbeat and pessimistic path, I can say with certainty that a world where I want to live is one where:

  • Not knowing is not a sign of weakness
  • Quality matters, and things have an inherent value beyond that which the free market assigns to it
  • Externalized costs are visible and seriously considered
  • Diversity in all forms, but particularly in this case written tone and voice, is valued and encouraged