written by owen on 2025-Oct-16.

I avoided writing anything on AI 2025 while the hype train was still hot. I figure that no one lives forever and I might as well drop something before the bubble burst. So I might as well write some notes. I am gonna keep this simple with a list of hurdles;

Poisoning and Dogfooding

You will dump a bunch of data into your Ai model, do lots of testing then compile what you think is a stable model. There is this point before and after the model is released. Before release the input data is pure. After the release your model starts polluting the environment from which the data originates. This is a feed back loop which will make it harder and harder to deduce what is original data as opposed to data created by the Ai. Poisoning your own well. Each rebuild there-after of the Ai model has the potential to run the ship into the dock. The summary becomes the detail. Everything looks like a hammer because you are living in a hammer factory.

Changing anything, breaks everything

Ai could end up in a situation where slight changes can break whole sections of logic. Like flying through space where moving a finger throws off your trajectory by 10 years. The more data you ingest, the higher the chance that some of that data might generate a bad model. The more users you have the faster these issues will bubble to the top of the pile. In the best scenario; one could simply block all the edge cases but eventually you will end up boxing yourself into a corner where every move is fatal. Check mate.

The power consumption

Ai is using alot of electricity to build a huge data cache/index. That "thing" - in my mind - is basically pattern matching search. This power usage is only going up. At some point we will need human batteries and even then there will still not be enough power to produce the model. Also note that this power is not insignificant - it is a lot of power - we know this power curve like we know light speed. It limits who can create and use "Ai". Most companies are using someone else's cloud services and/or data. Ai cant exist without the power draw. Everyone else is pure talk and subscriptions. SAAS.

Glorified Search

From what I have noticed use cases around Ai nowadays is "search". Fancy search but still search. Search is essentially the ability to find something, but with computers you can find it super quickly. Ai lets you find stuff really, really quickly with some extra fluff around it. This extra fluff is nice but the fluff is just that; fluff. I swear most answers I see in Ai web search are reddit/stackoverflow posts. Funneling people into a easy "catch all" solution is a kind of "web search your way to success" - yes you get your answer quickly but you lose all the other benefits that surround the act of "looking for something". This is coming from someone who used an encyclopedia Britannica for research in high school. Fun times.

Ai has no context of time and space

A trend on modern websites is to not put dates on their content because they want the content to stay "ever green". Added to that social media does not want you to stop scrolling because it wants your attention. These things combined are worse with Ai because responses have no concept of the time that has passed between when the data was collected and when it is output. So Ai produces this lovely data but it might be wrong in the past, present or the future. Bits of the data might be from 10 years ago or 5 minutes ago. Lets not even mention censorship and blocking information. ignorance is bliss.

Ai is Blackbox Tech

This is one of those things which layman like to talk about because they have no idea what it is. The more ambiguous the subject - the easier it is to speculate and market it. A plane crashes in the desert and the report says it was caused by pilot error. I have never seen the inside of a airplane blackbox, is it full of mp3 files? video? graphs? little text logs with telemetry?

We kind of imagine what is inside but no one really knows except the few people who can view the curve of the planet from space. All this enables layman to talk about Ai in "suppositions", acting as if something is inevitable based on something else then applying it to everything. Hammer theory - the over reliance on a familiar tool, but in this case Ai hype is Chinese telephone: someone telling someone else's lie. "Ai is were everything is going" and other folk tales. All this while secretly hoping that no one tests their assumptions. What matters is the fire or the ash that remains?

Ai creates nothing new

Besides creepy Ai slop videos and slop art, most Ai content creates is mostly remixing. Everything is a remix. At some point we will hope that the remixes will get better but the most popular Ai slop can almost instantly be followed back to the source(s). Ai isn't helping anybody outside of the internet. And people on the internet tend to only talk about things in terms of internet physics - they stay far away from real work problems and effects. At some point the internet became 100% marketing and having something to market became more important than life itself.

Conclusion

There is alot of investor capital inflating this tech bubble. Ai like every other buzzword pie kite is heading into space. One thing for sure Ai will not solve world hunger, make flying cars or free Palestine. If you are doing neither of those things you are just marketing. What matters.

permanent link. Find similar posts in Articles.