They're always peak when released, then degrade over the next few months. Which doesn't matter because in 5 months, there's a new record-breaking model.
Gemini Pro, once voted the best model in 2025 by Polymarket, just spouts nonsense half the time now. I say this as an AI Pro subscriber. It built this amazing system for me in a weekend and suddenly it acts like an old horse.
OpenAI often has the opposite effect - they're disappointing when released and everyone goes back to them whenever Gemini/Grok/Claude breaks down and discover that hey, it's pretty good.
Capitalism and democracy is often sold to us under the assumption that humanity is intelligent and can calculate depth.
Let's say Company A makes $100 products but creates 5000 units of pollution. Company B sells their product at $95 but creates 15000 units of pollution.
In this case, the penalty for pollution is built into the free market. Pollution is still necessary, but at what point is it just reckless? But this can only be the case as long as the market is aware of this cost.
"Political opinion" is an umbrella term for something that should be abstracted out. Global warming is a political opinion. People can just have the "opinion" that the pollution doesn't exist. Same for sweatshops, child labor, bribery, corruption, women's rights, and all these other pollution-like effects that damage the world.
Yes, I like my cheap, high quality t-shirts from Bangladesh, and I like the low cost labor that bring me my $0.16 tea bags. I like the water-guzzling, plagiarism machines that write my code for me.
But we have to decide at what point we tolerate these costs and for what output.
The basics always matter. It's communication in the end, and you have to speak the same language. Know what O(log N) means. Know when you pick a BFS vs DFS. Why do we use hash maps and why they work. Why Markov chains and state machines simplify design. How you can use Monte Carlo to make an efficient search in a game.
Many people think AI is going to be a singularity. But Amdahl's Law and Gustafson's Law suggest otherwise. You get diminishing gains with more power, and yet more power unlocks new things. Better CPUs didn't lead to instant simulations and word processing, they led to social networks and TVs that spy on you.
But you have to internalize these algorithms well enough to understand how the world works.
A lot of early games were more like file systems and they worked. I'm surprised nobody had mentioned ACID yet.
The API revolution might be another thing - you were able to swap out a database with any other. Risky decisions are fine when they're reversible. Databases were a more reversible way to deal with scaling and architecture.
It was always a strategy. More so with B2B. Would you use a payment gateway or database that has only received seed funding?
I worked at a B2B unicorn once that raised funds, not so much because they needed the money, but because it symbolized to the market that they were here to stay. That they're not going to run out of money and enshittifies when the market downturns.
No, I have the opposite reaction. A new company with funding to me just implies a runway that will run out, and no profitable business attached to it. Show me a balance sheet that is self-sustaining on their own revenue, and I'll work with them over VC-funded startups any day.
Are you joking? The cost of writing code is practically free. Some basic hosting is very cheap. Find just a couple of paying customers and you are self-sustaining. Hire only when the revenue supports it, and you remain that way.
Style 1: Question is so tough that you can only use AI to solve it. But for part to, you have to edit the code without AI.
Style 2: You are free to use any tool you like as long as you build this full app in an hour. Evaluation is that it works and that it follows appropriate architecture.
Style 3: You're given a dumber model, like Gemini Flash or Claude Haiku, and need to solve this problem.
It's the opposite of the article. Wrap the body in a cloth. Bury as fast as possible. Bury as cheap as possible. One headstone and one footstone. No buildings on it though the headstones recently have names.
Recently there's been more intricate graves and sometimes tombs, but time by time, people fall back to the original traditions. After all, the big caliphs who ruled empires were buried under two stones.
I think they're based around capacity. When everyone moved over to Claude and when the data centers got bombed, capacity went down. It was made worse with all these multi-agent orchestration ideas using many times more tokens.
They gave out discounts but that doesn't magically solve the problem of capacity and led to the service being overloaded instead.
reply