Let's talk about Artificial Intelligence

[Jerry_Atrick]

10 hours ago, nomadpete said:

 

Had the automatic rapid military response been controlled by A.I. You would not be reading this now.

You may be right - or it may have ended up in the same outcome. For us to come up with a hunch, we take inputs, compare them to what we would expect, and arrive at a hunch. It is quite possible and AI could have come up with the same outcome if it were pure AI and not just a discrete program directed to retaliate on identifying a missile launch. 

 

In fact, you could argue if it wasn't Petrov on duty, but some other commander, we would unlikely be reading this today, as well. 

[Jerry_Atrick]

It's a bright future he have. This came from a weekly AI newsletter I get:

 

npm is a package manager for javascrip and used to deploy most web apps. 

 

[onetrack]

You forgot to mention how a readily-available Chinese, AI-enhanced, satellite imagery programme, is helping Iran locate U.S. assets very quickly.

 

https://www.abc.net.au/news/2026-04-06/us-intelligence-chinese-satellite-imagery-middle-east/106508322

 

 

[nomadpete]

13 hours ago, onetrack said:

You forgot to mention how a readily-available Chinese, AI-enhanced, satellite imagery programme, is helping Iran locate U.S. assets very quickly.

 

https://www.abc.net.au/news/2026-04-06/us-intelligence-chinese-satellite-imagery-middle-east/106508322

 

 

Globalisation works in mysterious ways.

[Jerry_Atrick]

Indeed.. It looks like the "other alliance" is a bit stronger than ours: https://www.reuters.com/world/europe/russia-supplies-iran-with-cyber-support-spy-imagery-hone-attacks-ukraine-says-2026-04-07/

 

That't thank thanks Chump gets from Putin.. Of course, Chump won't see it that way.

[Jerry_Atrick]

Claude.io from Anthropic is considered the bee's knees for coind. They have just released Mythos, it's next gen.. except they haven't released it to the general public. Apparently, without specific training, it identified zero day and other vulnerabilities. It is claimed to exceed the capability of all but the best developers: https://www.forbes.com/sites/jonmarkman/2026/04/08/what-is-claude-mythos-and-why-anthropic-wont-let-anyone-use-it/

 

The software development industry is already being turned on its head. I used claude.io to quickly develop a prototype app that is fully industrlalised (i.e. uses all the correct data, application, user iunterface, security, distributed, and scaling techniques, complete with application and server health monitoring and the like). On the free version, it took my about two months because they have much stricter usage limits than ChapGpt. But, it was exceptionally well written code - and I didn't have to tell it what technoogy to use.. I basically said this is what I need, and it did the rest. 

 

Many companies are turning to AI.. a lot of Intergrated Development Environments (IDE's), which are th programs where you write code, debug, ibuild, etc, have AI already integrated. However, Claude is next level stuff already. Mythos basically will make the software development industry almost entirely automated.  We are currently hiring a lot of software developers for a big new program of work - it is huge. I won't go into numbers, but from about September/October last year, I have increased my analysts fby a factor of 10. We have been hirign developers (we already have a reasonable number) - looking quadruple our developer count. Despite it being a buyers market, we are finding ti tough to find quality candidates and we are paying decent rates for London. So, I will be proposing to our AI team adopting an AI dev tool for our project. We have to deliver a shed load in a year, which is why we are looking to hire an army of people. But, for literally a few hundre dollars per user per month, we can get exceptional throughput. And here's the interesting thing: It can write the code and a separate AI bot will review the code and suggest improvements. If they are accepted, the improvements are fed back intoi the code writing bot as well.

 

So, what does this do to the industry? It doesn't kill it entirely... 

 

The first thing I predict will be a virtual removal of programming languages.. Of you think about it, programming was originally writing a bunch of 1's and 0's in a specific way and took an eternity. And then, Assembler/Assembly (depending on which side of the tracks you were born) was developed as a human readable extraction of the CPU instructions. You had to learn Assmbler for each processor family (and sometimes individual chips). That saved a lot of time compared to the binary coding and the instruction set was somewhat standardised for standard operations. This is effectively a second generation language. Then came 3GLs - these were BASIC. fortran, COBOL, C, and these days Java, C#, Python and the like. They are general purpose langiages that are feature rich and provide a programmer all of the control needed, but the difference is that the language applied to any computer (wlell, almost).. as a compiler or interpreter to convert that code to binary code (executable code). So, in theory, if I wrote a COBOL program on an IBM Mainframe, it would compile and run on a PC with a COBOL compiler. The truth is there are extensions and platform specific functions. Although, Java, C# and Python have fixed that by being interpretative, and the use standard library extensions. 

 

4GLs made a relatively brief appearance, in that they were higher level 3GLs - an attempt to make programming more English like and democratise development to the users. Ironically, langiages like Python, which are modern 3GLs have done more for that than the 4GLs. There are 5GLs, but they are largely obscure. 

 

But AI - especially such as Claude will, IMHO, replace computer languages as we know it. They will become English (or other human language) builders. At the moment, they are code generators.. I asked Claude to develop my app using Java and specific frameworks (though it suggested some better ones than I requested, so I went with its suggestion). When it did create a bug, a quick prompt fixed it. The article above talks about usign humans for reviewing the code, but I would suggest once there is critical mass of confidence in the technology, code will no longer be required. 

 

For example, you have your series of prompts to build an application. Claude (or other AIs) can generate the executable (or interpretive bytcode, a sort of executable) direct - no production of human readable code required. It will be tested and if it works, it can be deployed; if not, adjust the prompt, rebuild and retest. 

 

Back to my team; about 1/2 of my analysts are ex developers/software engineers - like myself. We have very good business domain knowledge and reasonably good, if outdated software development knowledge.  My idea is that we are given a system architecutre (data lake, data frameworks, service platform, user interface platforms - e.g. desktop, tablet, phone), asd we start crafting the prompts to build the system. The first will be the data model, of which there are industry standards we can tweak. I would suggest we will have the data model done, say, in two days. We can use AI to populate the database and then write tests to ensure it works property. Say 2 - 3 days. Within 5 days I can have something that would take as maybe a month to get what we think is right and would probably take 2 analysts and some database admin support. 

 

The I can start building the individual functional services asnd UI, as well as integration to other systems. Let's say 2 months to build, system test, performance test, and have the users acceptance test it. The it is a deploy to the production environment. 

 

In fact, we could easily automate the testing, as well as for updates, any regression testing. Say add 2 months, maybe with a lead tester/QA person, an architect part time, a senior dev part time to review stuff, and maybe a database administrator part time. Otyherwise we would use analysts - probably between 3 and 5 because of the number of functions we have to deliver. 

 

And the best thing is, even if it produces spaghetti code (which it doesn't), I don't care, because what is important to me is the promtps that build the system. And you don't need to be a linguist or an expert in the coding or the specific language, database, user interface technology to use it. In fact, in my little home experiment, Clause inferred much of what it had to do without me needing to even hont at it. 

 

As I said, I can't give you exact numbers, but our non-AI plans are close to 100 people all up, with a minimum viable product at about a year, and then the full enchilada in two years.. and there will still be kinks to iron out, performance issues, regressions in releases, etc. 

 

But  with maybe 5 analysts and a smattering of other support, I cam have the whole enchilada within 6 months. 

 

Now work out the man days. In the UK, there are about 220 working days for professionals.. 100-ish people * 220 days * 2 years = 44,000 working days. And my business has a cost of using manual processes and elevated operational controls from day 1, but reducing as more is delivered over a two year period. 

 

Now I have, say all my staff in an AI world on it full time (which I doubt, but let's go with it) = 110 days * 9 (5 analysts, 1 developer, 1 QA, 1 DBA, 1 Architect): 990 days. 

 

Assume $1000/Day per person. Without AI, $44m; With AI: $990,000. Assume $500/mo per user (that is what a company a friend of mine is paying Anthropic for Claude): $27,000 gives a total with AI of $1,017,000. Even if I am really, really optimistic, and we quadruple the estimates, that is $4,068,000, noit quite 10% of the costs of doing it the old way. 

 

It is going to be a bumpy ride, but in theory, all services should get cheaper once widespread AI adoption kicks in. 

[Marty_d]

So.. what you're saying is that the shareholders of your company will save $43m, but 95 people won't get work.

And a good slice of that $43m will go to the top execs as bonuses for claiming credit for your ideas.

Would that be accurate?

[Jerry_Atrick]

At this early stage that is what will happen

 

I would expect concentration in the AI market and costs to go up a bit, too

[old man emu]

Sounds like teh AI you describe will put half the population of Asia out of work.

[nomadpete]

4 hours ago, Jerry_Atrick said:

Claude.io

Apparently, without specific training, it identified zero day and other vulnerabilities.

Did it start with the number 42, then work back from that?

[nomadpete]

So, it has come to pass.

 

Back in 1968, I was an apprentice TIT. One of a hole bunch of TITs. We watched in awe as our instructor held aloft a beautiful object about the size of an A4 page. It was a maze of fine golden wires in a matrix of tiny ferrite toroid rings.

 

"This" he proclaimed, "is a matrix memory. It holds a thousand bits of data for a computer. It was made last year and today is obsolete. At the present rate of development, in a few years we will see computers so smart that they can develop their own programmes!"

 

So I have seen the original Matrix.

 

It took a bit longer for it all to happen, but it has come to pass.

 

OME, never fear, all those redundant coders will soon be free from enslavement.

[nomadpete]

BTW,

Back in 1968 my apprenticeship was called, "Technician-In-Training". Which made us a whole class of new TITs.

Edited 33 minutes ago by nomadpete