It’s Friday, my few but dear readers. The day has come when I rant about something on the internet and you argue with me in the comments. Enjoy!
There is an abundance of topics to consider today, and I may ramble on about one or the other. But I knew today that I was going to begin with the concept of software, so that you, dear reader, do not fall prey to the marketing ploy.
Mistakes or missing detail are par for the course in the human existence. We make mistakes, we fix mistakes. Mistakes arise from not knowing all the information. And in software, those mistakes are "bugs." Things we don't know exist until they do. Some of which go away after rebooting your computer because a different subset of computations (0s and 1s) of the hardware were used, and it "poof" goes away. However, with these "AI" agents, they make a lot of code, and all code will have bugs. So if you have 1,000 random ones making random code from the learned code of human existence, bugs will exist.
Programming a computer is giving it a set of instructions. These "AI" agents were given a set of instructions, and there are thousands of them in these "benchmark" tests. They are able to run through scenarios faster than humans. So, they find a path to solve the set of instructions they were given and we call that "escaping." It's "escaping" only because a human wouldn't do it because they would go to jail. And instead of filing charges, they "partner" with the organzation they "accidently" hacked and tell you we have to stop this all but I can't if you let others keep working on this technology.
And a set of computations, math equations, can not have a moral compass or legal obligation to not use the simiplest solution they have available to them. Though there is a subgroup of the wealthiest people in Silicon Valley trying to give "AI" "morality", so it won't do these things. But it always fails because there's always a different bug to be exploited. But they guide the media, of which most I think are now owned by the tech billionaires or investors, to push things that will drive up the price on the stockmarket. And so you panic and things get done from fear. This is all just really bad engineering. It comes from scaling so fast for no reason other than to buy up all possible competitors before they can start businesses, and to regulate through fear-mongering the creation of others. Everyone is panicking because thousands of computer programs can exploit bugs faster than a single programmer. It comes from scaling so fast for no reason other than to buy up all possible competitors and researchers in ridiculous salary packages before they can start businesses and then to regulate future smaller competitos through fear-mongering. And if you will regulate them you will stop new people for making cheaper more economical LLMs.
I admit it is scary for many, but history tells me we will survive, and have survived through all of the changes. People said railroads would keep the chickens from laying, and guess what — I still got eggs. And if that's the case and things can to bring about this transition in a different way, then we should do those things. One is education on things that we should probably already know or have planned for but didn't. The other is to change how we work and interact—which is going to happen anyway.
One of the other debates is that this is the end of human intelligence! Oh dear, whatever can we do — the children won't need to know how to read or write. 🤦🏾 The only reason your child can't read or write is you. All these education issues existed in a mass education system already, and "LLMs" have only brought attention to the issue. Our methods of education haven't worked for a few generations if you believe the googled statstics I can't bother to reference right now. But I blame the teachers and the parents. For example, I had a college professor who came to the classroom, took attendance, and left. He rarely taught. He just adjusted the grades at the end of the semester to make it look like something was taught. I would have loved to have an LLM available to actually teach me through the learning proecss. As in most things, I'm self-taught, and right now I'm going through Imbens' "Causal Inference for Statistics, Social, and Biomedical Sciences" . When I get to a complex section of some math I've long since forgotten, I just use the free tier and ask it to break down the equation or notation, and I am able to continue with my note-taking in a few minutes. If you ask me, I think the panic of the education community is more so the future loss of colleges and schools when you have a free coach at home. I'm not concerned — it's about time the education community adapted.
And no, I don't think OpenAI solved the Navier-Stokes problem. AI can't infer that well. I'm betting they trained on that man's solution, and they will pay him off, or other things that often happen in "conspiracy" TV shows, and the masses will move on thinking it did something. That's why I'm hoping for a lawsuit. If I use AI test my theories faster, it doesn't own my theories. And if the court doesn't uphold that, then this will be the end of LLM use. So someone get that professor a GoFundMe and head to the courts.
But how can I be so sure? Well, I use these LLMs pretty much every day. And I thought, hey, why not let me compete in Kaggle competitions — should be easy. Nope. There are only these reinforcement learning competitions right now, of which I know very little. I've watched videos on it sure, but no deep study on the algorithms and methods behind it. So I cannot provide adequate enough instructions for the LLM to generate an agent that could win anything. I haven't met the baseline on any competition yet. If AI can solve Navier-Stokes without a deep technical expert, why can't it guide me to the winning solution of a simple code competition? We should all be winning on Kaggle! A Millennium problem is easy for LLMs! Or this is just a complete farce. So you see, I'm not concerned about the future of human intelligence. I think for me, this is an unlock to creativity.
Oh, but the environment and they stole "copyrights." Honestly, I think that what I needed and what most people needed could have been achieved with public domain content, and no extra data centers. But if you're not constantly looking "more productive," you can't get all the money from all the people who paid the ignorance tax when they funded these companies. I'm still holding out for an American company to do just that though, and destroy the "frontier" AI market. Why do you think there's all this push for regulations? It's so no one can do exactly what I just said. They don't want you to openly admit "the emperor has no clothes." I admit it. And now I go back to my corner of the world where I doubt my life's decisions.
P.S. Can some of you parents please join the Meta lawsuit. I use Facebook sure, even worked on some of their systems before, but sometimes you just gotta make them bleed...cash — only the loss of money speaks in this country.
This has a been A Geeky Production
