Posted on 03/16/2025 6:24:41 PM PDT by SeekAndFind
Yes, I’m sure when management comes and tells you to teach this how to do your job you have nothing to worry about.
Bingo! You can't be a one-trick pony. You must constantly learn the latest technology that customers are willing to support. You can't master everything that is flooding out. Javascript frameworks come and go. Some are good enough to be adopted and required by a customer. That's where you spend your training time.
bfl
TLDNR
Where are the Cliff’s Notes on this?
“I guess you never heard of Document AI.”
so, is that what H&R Block and Intuit used to convert the tens of thousands of pages of Fed and State tax laws and regulations into TurboTax and H&R Block Tax PC programs?
When I finished the task, my company took the new OS to a computer show in Europe to show it off, then a week later an earthquake hit Haiti. My code was in the field providing support for military activity to support the rescue/recovery efforts. No AI in those days. Just hours of grunt work making that kernel code bullet proof. The "magic" under the covers that lets people to the high level stuff with ease.
I have ai open on a separate screen all day. It’s great! I can paste a block of code and say: “see this code? I want a sub that is similar, but does x, y, and z”. Or in sql, I can paste two table definitions and say “I want to keep this table in sync with that table, create an INSERT, UPDATE, and DELETE command in a stored procedure that will do it.”
“Insert debugging lines within this code”
“Can you help me figure out what is wrong with this function”
“I want to write a DAX formula that does X. In Sql, I’d do it like this: YYYYY. How can I do similar in my DAX formula.”
Sometimes you go down a rabbit hole of bad suggestions that loop back on themselves. But overall it has helped me expand my knowledge, learn new software quickly, and do things that I may have thought would be more complicated than they were.
If I had the money, I’d invest in nuclear. AI isn’t going to get smaller.
I remember that very article. I took fortran, Basic, Visual Basic and C++. We studied the history as well.
Bingo! Much of those quirks are only known to people who have worked on those systems for years. It's corporate memory that an AI can't mine. When that corporate memory gets retired or laid off, there isn't an easy recovery path. When I left PacBell in 1991, there were 360 major projects underway. I was one of thousands of people who left the company at that time. Out of the 360 projects, 180 were a total loss. No way to continue. It was capability being built for the future. The remaining programs were farmed out to contractors. In one case the contractor worked on a project and proclaimed victory. A program review revealed that the contractor had achieved victory interfacing to ONE other system. The contractor was ignorant of the other 5 systems that were required for the process flows to work. Essentially, that was a "scrap" too as the corporate memory had been retired or laid off. Oops.
Great article.
There is a low-code / no-code movement that enables those with minimal or no programming skills to build business automations with tools like make.com and n8n.
This is popular for marketing agencies and business consultants.
These tools can be used for things like social media content creation (for businesses), sending automated marketing emails and text messages, and fielding phone calls with interactive AI.
Ping!
Woooow - because no programmer has ever come up with a program to produce prime numbers before?
Still voting RINO?
It’s gotten better in the last year I’ve used it but for my coding tasks but it still generates worthless code about 80% of the time.
What it is, is a better search algorithm… but I attribute that to Google et al degrading their search algorithms for ad space. They’ll eventually do the same for AI and we’ll be back to the current state of things.
The Sieve is still a common interview question, so it pays to have some idea how to implement..
I can give you guys some war stories from the AI front. I’ve been a professional dev since ‘86. Assembly/C/C++ and most other popular microcomputer languages. Moving on. At the office I was asked to develop a script to parse a series of PDFs. I knew this was going nowhere, and by the time I was done with the script they would have changed their minds, so I decided to use AI to do it. It got the first 3-4 working, but then when 5 was different it took over a day for it to produce code that code handle it. (This is OCR’d data from the title page of a patent-like document). By the time it had figured out how to handle #5, it had broken 1-3. So I had to point it back to those, and it fixed them, but then 5 didn’t work anymore. These were in essence TABLES that were getting scanned. Sometimes the word “Name” was on the LEFT side of the name, sometimes it was ABOVE. It couldn’t seem to deduce what was doing on despite me telling it. “This data comes from a table, sometimes the word ‘NAME’ is on the left of the name. Sometimes it’s ABOVE it”. I even caught it one time adding code to special case a situation it was running into. (The name had spaces between elements, and it had been asked to remove all by 1 space.) It added a “If name = ‘first last’, then name = “first last” chunk of code. I had to remind it that there are 1400+ PDFs in this collection and that it couldn’t special case ALL of them. Anyway, it was about this point that I was told that there exists an online tool that can spit out this information, and thanks for all the work I had done.
Being a glutton for punishment, I decided to try one of my home projects with it. I write software bots that play classic video games. I’m working on a new one now, and figured I would see if it could reason some code to handle the targeting. Short answer? NO. Long answer: It’s like it has very limited memory. You can only add so much background and corner cases before it starts losing some of it. At one point it produced a 500+ line file that did MOST of the parsing. But had a bunch of bugs. In the process of trying to get the bugs fixed it provided a 200 line file that had code to address the current bug, but had stripped out a bunch of the previously generated code which was REQUIRED. Once again, as though it ran out of memory, and had to “tune the working set”. I finally got it to a better place and called it quits. I’m now working through the code and finding the holes on my own.
Same here. My first love was the Motorola "D2 Kit" for programming the 6800 processor. I taught myself 6800 assembly language during the commercial breaks of one episode of "Saturday Night Live". A few days later, during our computer science lab, our instructor was called out of the lab to address a "computer emergency". When he returned several minutes later I was in front of the class teaching.
I'm working on a little programming project using Claude now. It's something I've wanted to do for years, but not had time - a web app to create a graphical timeline by combining any number of JSON-formatted "timeline files". What I'm discovering is that AI-driven programming is not a all "ready for prime time", and that the obstacles remaining are formidable.
Almost 70 years ago I was writing programs in FORTRAN 4 and using paper punch cards for input and some 25 years ago I learned Java Scripting and HTML. These programing languages are now about as dead as Sumarian and cuneiform. However, I have kept my old slide rule and trusty book of logarithmic tables so if the sun belches out another Carrington Event frying all our computers, I am ready.
That’s my point - it’s already out there a hundred times in Google land and countless other search engines. The code is already there and the “AI” so-called is synthesizing answers from existing results
Disclaimer: Opinions posted on Free Republic are those of the individual posters and do not necessarily represent the opinion of Free Republic or its management. All materials posted herein are protected by copyright law and the exemption for fair use of copyrighted works.