Raw LLM Responses

Well, I’m a forty year IBM mainframe vet and I’m currently trying to teach Aitrium to archive the company’s 20 legacy application… and it’s not going well. First, let’s de-mystify the processing… the corporate version is a character-string analyzer that prioritizes the number of hits in the document dump it is given and formulated a grammatically correct response NOT a “correct” response (because it is not sentient Duh!) So it’s really only useful for “syntactically” checking your code base for “hints” on How to code a very SPECIFIC parameterized problem. What Codex, Co-pilot and the IDE don’t admit is this really only works in a C language framework with an Object/Process oriented design… THAT is NOT how legacy COBOL applications - that still run CORE processes for Government, Banks, Financial firms, Insurance, Transportation and Manufacturing were built… they weren’t “designed” but EVOLVED to meets the IMMEDIATE needs pushing massive amounts of data thru all against processors AS FAST AS POSSIBLE! And the complexity is the sheer number of code upon code that has been layered on - most of it dead or just playing possum - so you don’t know what you might execute on a given run… so you let it run and only tweak it when it doesn’t work… AI has no answer for case instances that haven’t occurred (yet) and - even if it came up with the “correct” fix - the programmer would have to be able to determine IF one of many responses IS the correct resolution… sooo… FIRE AI when it stops working in the middle of the night and the bank is hit with million dollar fines for not processing credit card transactions by 5AM 😏 For the last decade, the Uber managers of these old IT shops were sold on “modernization” (I worked at the largest credit card processor that handed Amazon and after a decade and MILLION$$$ pissed away on “modernization” and the “cloud” and other such buzz verbiage.. the CORE credit card processing is STILL on TSYS and ACAPS ie IBM mainframe COBOL, CICS and Assembler code bases because NOTHING they tried to replace the functionality wit’ could run faster (for reason that are obvious to a programmer but incoherent to someone who has never got up at 1AM to figure what “abended” in an assembler program 🤓) Bottomline with “modernization” it makes no - economic - sense to piss away MILLION$$$ on “replacement” system that will cost MORE to run and will run TEN to HUNDREDS of times… longer. And AI is only as good as the people who are training it - as the output require TRAINED people to tweak it! (And the “tweaks” aren’t persistent!) Will AI replace programmers, writers, photographers… managers YES! But only those who can’t provide more value that ‘vanilla’ AI results! 😉