Yes as long as we ride the coattails of the uber rich we should do ok.
Iâll admit to being at least as cynical as the next 5 guys. Long long time ago I developed the modus operandi âFind out what the game is and play that. Get something out of whatâs there because you canât get anything out of whatâs not there.â Go along to get along, I guess. I think in NASCAR itâs called âdrafting.â But try not to sell your soul in the process.
1 Like
FWIW, the correlation between SPY and QQQ is 0.86, which is quite high.
DB2
3 Likes
By way of background I stopped counting programming languages in which I had done development when I got more than 50. In 1990 I developed a tool called Specification Driven Development which allowed a programmer to do development of many functions by defining specifications and clips of special code where needed. The system was designed such that we could change the underlying model, e.g., to take advantage of a new language feature or a new requirement and then regenerate all the code to incorporate the change with no additional programming effort. This allowed average productivity of approximately 1000 lines of 4GL per hour or the equivalent of 3 to 10 thousand lines of 3GL. This was used to produce well over a million lines of 4GL. Later in my career I worked on, but did not finish a system to automatically write 4GL code directly driven by UML models.
With that background, I was thinking about the current use of AI to replace programmers or enhance their productivity and it occurred to me that the problem with using a base of existing code as the reference for the model is that most existing code is terrible. How is the AI going to do anything but produce more terrible code?
Compare this to the UML approach, which is used in some industries. While there is significant development work used to build the translation engine, no development work is required for each new function. Instead, the work goes into defining the model, a symbolic representation of what the function needs to do, a model which needs to be created by a programmer type person, but a model which can be understood and validated for correctness relative to requirements by a subject are expert or analyst.
I just donât get how we get good code out of the AI approach and it seems like we are left with the same long term maintenance problem we have with existing hand-written code. E.g., when the Y2K thing came along, most code had to be manually examined, evaluated, and fixed to accommodate the change in specifications. With the model-driven approach, one just changes the definition or perhaps the translator, pushes the button, and out comes new code with the new specs. With AI code, we are going to be left with the manual method.
2 Likes
tamhas
Thank you for the terrific knowledgable thoughtful post.
I mentor two Mexican kids studying math on their way to AI based careers, one of them actually recruited by Facebook for work in Menlo Park HQ last summer. I also do special tutoring for more normal students needing help for their classes in advanced algebra and programming. I feel like we all might be on thin ice.
Background on me and [I have nowhere near your breadth, was always a maverick, but I did hard core programming long long ago, actually starting off using Fortran to control missiles⊠totally insane, and participated with the losing âblueâ team in the competition that led to the creation of ADA, which i still think to be a klugy abomination. In the late 70âs I fell in love with Petri Nets, Petri net - Wikipedia
and at the end of the 70âs was using that âgrammarâ in software design hooked to Pascal and descendants boosted by Forth
Forth (programming language) - Wikipedia
as programming languages for contracts requiring very fast running code in early microcomputers.
I and the Mexican scholarship organization I volunteer with (âJovenes Adelanteâ) have become extremely concerned about what seem to us to be the hopelessly antiquated curricula for computer programming in most Mexican universities and colleges. UML is a very interesting option as providing a mode of thinking useful for driving the use of AI in programming. I had never heard of it.
More please!
david fb
2 Likes
Yeah, you have to specify to the AI chatbot that you want âgoodâ code or code that runs at a speed of X. Good luck with that.
Thatâs sort of what those new kids thought they were doing when they invented compilers
Mike
3 Likes
tj,
If ZeroHedge believes it then there is no way it will ever happen.
Really horrible source.
The company I was working with at the time was Pathfinder Solutions
They had the most open system such that it was possible to add a new language. All such systems I know of use 3GLs, but I was working with the Progress 4GL ⊠an interesting little switch. I still think it is the right way to go.
UML is itself an interesting beast. It has expanded to a huge monstrosity, but actually only 5 pieces are necessary to do the modeling for this kind of work. When one has it going for a particular application, no other approach can compete.
2 Likes
Tamhas
Thank you!
I have sent off to Pathfinder for more info. Any other links into UML that are more mother load than the scattered attractive but lesser placer flakes I have found off the google?
david
1 Like
Ahh, found the official UML org site. I have lots of reading to do.
david
Tamhas good because it is all BULL BLANK!
Back in art school weâd go to gallery opens as an assignment and write a paper on the art and artist in the exhibit.
One Friday in the school gallery we had a NYC artist put up massive murals about five of them. He gave a talk about going into a mystical state to create his abstracts.
On Saturday and Sunday I went and sat in that quiet gallery with a pad of paper to meet my assignment. I threw out a lot of balled up paper. I had nothing.
On Tuesday I was in the cafe waiting for class. The associate professor of art came up to me. This was odd. Note I was the leading student in the curriculum. She had one quick question to ask, âwhat did you think of the show on Friday nightâ? With no thought at all, I said, âI am not turning in a paper. I have nothingâ. She scream for all in the large cafe to hear, âgood because it was all BULL BLANKâ!!
We were all given an A for showing up. The paper was not required this time. The two professors had talked it over.
It is important for us to know when we see BLANK.
1 Like
@tamhas with that in mind it is a sad day when Meta is more honest than Google or Open AI. Yet just being a spoiler in a completely worthless battle.
Dealbook by Andrew Ross Sorkin, NYT email column no link.
snippet
The memo, entitled âWe Have No Moat, and Neither Does OpenAI,â was leaked to the Substack newsletter SemiAnalysis. Zoubin Ghahramani, a vice president for research at Google who helps oversee A.I. work, confirmed the memoâs authenticity to Cade.
It says:
Weâve done a lot of looking over our shoulders at OpenAI. Who will cross the next milestone? What will the next move be? But the uncomfortable truth is, we arenât positioned to win this arms race and neither is OpenAI. While weâve been squabbling, a third faction has been quietly eating our lunch. Iâm talking, of course, about open source. Plainly put, they are lapping us.
For Meta, releasing LLaMA may serve several goals. It may undercut the early lead that OpenAI and Google have established. And as the journalist Ben Thompson of Stratechery noted, A.I.-powered content-generating tools could also be âgrist for the millâ on Metaâs platforms like Facebook and Instagram.
We could end up with Google and OpenAI and their proprietary systems on one side of the battle, and an armada of open-source competitors, fueled by Facebook, on the other.
1 Like
Note that UML itself goes on and on and on ⊠some people like to model all of that stuff, perhaps to help understand the problem better. But, to generate code, you only need about 5 diagrams and AAL (Abstract Action Language used to describe the procedural elements). So, donât get too hung up on the zillions of different pieces in UML.
The companies other than Pathfinder who do this kind of work have proprietary translators, so there is no way to implement a new language.
2 Likes