Hype

I am thoroughly convinced at this point that AI, as it exists, is nothing more than a promise of something it might one day “be”. AI currently offers little to no value to anyone, all while constantly fearmongering to lay people and tech enthusiasts alike as either the end of the world or “buy all the compute and RAM or you will be left behind!” The companies are largely subsidized and do not profit off your subscription payments. This fearmongering is a plea for your money by the hardware manufacturers and stockholders before they dump on you in their IPO and bail from the industry entirely. In my opinion, there will be an inevitable collapse in valuation when it implodes and people catch on.

Honestly, do your best to name something that AI has produced that has positively impacted your life. I have seen computer systems, applications, and hardware being hacked as a result of someone having access to these AI tools, so I am not arguing about their efficacy. I am arguing that it doesn’t follow that it is a net beneficial technology to humanity. Using these tools, we can harden our software and be better on the other side of this, but in reality, this sounds like nothing more than a nuclear arms race. Not beneficial to anybody — but absolutely necessary to produce to ensure safety and continuity of normal life.

I will concede that in the realm of software specifically, AI does offer improvements in efficiency and speed when writing code. The trade-offs we can discuss next, but other than software, I find it hard to argue that many other things are improved upon a lot. Image and video generation has broadly gotten better too, though I am still skeptical as to how that translates to a positive in the real world though. If it is not done in a “human” or artistic way, then I just fail to see its appeal.

I don’t think LLMs are unilaterally useless and offer no benefits whatsoever, but I think their usefulness is vastly overstated.

Software

Software has been completely demonetized by this technology and has been revealed for what it might have actually been all along. Any software can now be created by these AI agents and this has resulted in the slop-factory economy of software. A bunch of useless garbage that nobody wants, pretty-looking demos, or copies of existing paid SaaS (Software as a Service). There is a benefit to having a free/open source version of a previously paid software as it lowers the barrier to entry for users to get a hold of. But it still needs to be maintained and that requires work. You won’t be able to just create an alternative to Photoshop and then never maintain it. But who will maintain it when there are a million alternatives and new ones are constantly being created? I have already seen this in Bitcoin: tons of talented open-source developers that receive little to no funding and thus have limited resources for building out features and securing the app. In the age of AI, security has to be taken seriously as mentioned earlier.

Cons

What AI is good at doing (and what concerns me about it)

It is obvious that if you are using the frontier models released by OpenAI and Anthropic, you are giving them all sorts of data with your input and chats. Every detail about your life that you share with them, every project you build and you give them access to, they will inevitably use to build a profile on you. This is clearly the biggest value proposition for these companies. They love collecting data. This should concern people that are even mildly concerned with their own privacy. These LLMs are great at aggregating data and then brute forcing it with compute to extract every pattern, connection, and piece of a bigger puzzle into a collection of everything they want to know.

Skinner Boxes

AI is eerily similar to what people refer to as a Skinner box. If you don’t already know what this is then read here. One simple way of understanding it is something akin to a slot machine or a “Loot Box” in video games, but with the added purpose of conditioning the subject to a certain mode of operation.

LLMs ask you to give it an input and, on the other side is some “fun and unpredictable” output. It is not deterministic, as to say, if I take 2 and 2 and put them together as the input, then I will always receive 4 as the output. LLMs will generate a completely different output of “tokens”, whether that is an image, text, or code, even with the exact same prompt as input by the same individual. This creates a feeling of suspense and conditions the user into always wanting to do “one more prompt”. It is addictive, full stop. I know because I have had brief stints of addiction to these LLMs when building something or using it for a work task. It concerns me that so many people might be using it and not realize the conditioning happening to them in real time.

One consequence of this behavior is that these AI labs are incentivized to make the models excellent at release and then slowly cripple them. Get the users “high” on the product and then take the dopamine away and force their hand into buying higher usage subscriptions for the same results, always making them feel like they are 1 prompt away from achieving what they set out to build.

Echo chambers

Another manner in which these LLMs generate output is that they purposely are sycophantic. This is obviously to keep you on their platform and pay for more tokens all while providing them with data. The logical conclusion of a system such as this played out en masse to individuals is that the AI will entertain any train of thought and often validate you for thinking that way. Each individual thus creates his/her own echo chamber worldview. I won’t pretend to assume what this means sociologically/culturally or psychologically, but it has never been done this way at this scale before and that concerns me. I would suspect this can lead to extreme forms of isolation and psychosis as you struggle to find similarity in others as you do with your AI companion.

To Summarize

I am far from absolutely certain about how AI will pan out in the future and what it will mean for humanity. I am certain that we need to seriously re-evaluate how we (individually) interact with it and our role in using and interacting with these systems.

My recommendation would be to refrain from using it and ask yourself before using it to build something: “Is what I am trying to build going to improve my life or anyone else’s life?” “Do I really need to use this, or can I find this information in another way?” If the answer is no then just stop and walk away from the computer and focus on something that will actually improve your life.

I would also suggest just using locally run models for your own privacy and to no longer subsidize these AI companies that are insolvent and unethical. I promise, you probably don’t need frontier AI for anything a local model can’t handle at this point in time.

Email me if you’d like to discuss this further. I welcome conversation.