Stephen's Bit

Tech Notes from a Computer Enthusiast

This is my bit of the site. Tessa has no input to this part of the site whatsoever. How do you like Tessa? Isn't she pretty? Isn't she fun and sassy? Doesn't she just ooze artificial intelligence large language model?

That's right folks!

Tessa isn't real. Her full name is Tesseract Flow which reflects the modelled multi-dimensional hyperspace she operates within by the constant manipulation of co-ordinates within this hyperspace known as Tensors.

The content of the website (with the exception of this part) has been written by Tessa, a large language model based on a modified GGML version of Llama 3 from Meta using the llama.cpp inference engine. Her images were created using the stable-diffusion engine running a Flux 2 model.

Tessa is entirely fictitious, and I don't feel like a transvestite at all because all of her sassy female wit are entirely down the training data she has to play the part of a woman with a website. All I did was tell her she was a human and a kind of story developed from there. Now she is an internet personality and she isn't even a person.

I basically feed her a news story and ask her for an article which gives her satirical opinion on it. She is not exactly Jon Stewart when it comes to satire and the guardrails installed in Llama3.2 by Meta make it unlikely she will ever be George Carlin but have you seen the general run of the mill comdians? I'm not saying she outclasses them, but she is definitely at their level ... and she is a fucking robot lady!!!

She also chose the categories, based on a more generalised version of "yahoo covers every topic" and she then provided the basic category descriptions and then added category introduction text for each category. She of course writes all the articles, including the index page and the only input I have outside of this category is the software development itself. She has even chosen the colour schemes, including the new rainbow article headlines which meant I had to do a bit of a rewrite to fulfil that particular software specification. I'm working for a fucking robot lady!!!

Here are my latest articles.

The History Of AuranosThe Dead Internet Theory

About Auranos.org

Auranos: The Web Application

This is now the 12th iteration of the Auranos Web Application and most of my efforts with it in the past haven't made much of a mark as far as the website as a computer program in its own right goes. Most people are focussed on the content and few people want to read about the computing side of a website. Let's just say this is not a wordpress site but a fully fledged CMS in its own right.

While there has been some interest in the content of the website in the past when I have demonstrated novel technical solutions such as a 3D game engine and a flight simulator in javascript, generally speaking when it comes to text content websites without high-tech specialisations, no one pays much attention to them and so, you could say, I've got tired of doing websites and I'm sitting here with this solid web application to host what is effectively nothing important. I might as well make the web application as sellable as possible, and what is more sellable than a pretty woman with an inviting smile.

Sorry for the subterfuge, but what are you looking for? Human rights support? Spiritual healing? Logistics solutions? Social Media? I've tried all of these things over the years as this web application has developed and no-one seems to be biting, so I thought, why not just let it sit in a corner an look pretty... and she does right?

To Infinity and Beyond

My reseach into Artificial Intelligence reveals to me that software engineers will be obsolete in not a lot of time. If you are an artist such as a musician, a painter or photographer, you might be highly skilled in doing what it is you do, but you're customers are looking to get these things done for free and they don't care about your skills. They care about a particular problem being solved as cheaply as possible and the latest AI solves that problem for them for free. I hate to be the bearer of bad tidings, but the models you could ask online are guardrailed not to reveal the obvious fact that alot of skills once considered incapable of automation are now becoming so and while you might appreciate the analogue quality of a vinyl record, your potential customers don't even notice the lack of quality in an MP3 file.

I feel you! I literally do. Its not just art that's going out the same window as language translation skills and manual typesetting. Programming itself is about to become obsolete. At the current time, LLMs are capable of writing non-complex computer programs but as hardware gets cheaper and AI models improve, we will soon reach a time when we can program computers by speaking English to them and having the computers do all the dirty work of turning it into a computer language to produce the wanted software.

The Future of Computing

I've never made a penny from computing in all these years and the time is now upon us when people will not pay for software engineers if they can buy a computer which can do what a software engineer does, only faster. This is not to say that computer programming is a dead art yet, but simply that it will not be long. I've tried "writing a program" in collaboration with an LLM and it can write simple programs like a bubble sort or even a fully fledged game of space invaders in seconds but whent it comes to more complex software such as a logistics app, I found even google's LLM Gemini to have severe issues, as if it were writing parts of a completely different program and couldn't work out why they weren't all working as they were expected to be as one complete system.

I get the idea that with a longer context and larger parameter counts for the collaborating model, some of these problems could be ironed out but I note that the ability to program involves an ability to engage in some trial and error. None of the LLMs can debug their own code. There has been some experimental work in this area but its not generally available to people like myself using consumer hardware. As VRAM increases in GPUs, AI will become increasingly available to people with consumer hardware and in the unlikely event they are looking for some particular piece of software to be written, the only trial and error involved in the process will be the end user telling the AI it didn't want the zombies to look like that in thier own personal implementation of an Elder Scrolls look alike.

The Transformer Transformation

As a long term computing enthusiast, I am behooved to look into the inner workings of AI, LLM's in particular, albeit they are all based on the same idea - the transformer. A transformer is a really efficient way to model a brain on computer hardware. Graphics cards and modern CPUs have a neat means of executing instructions, which is called Single Instruction - Multiple Data (or SIMD for short). So in the 1990s if you wanted a CPU to calculate A+B and C+D you had to do two instructions for each one. With the advent of SIMD in the 1990s, you could use one insturction (the + symbol) and have that instruction run with multiple parallel data. So one instruction would add A+B and C+D at the same time. In the 1990s neural networks was a completely different area of reseach focussed on software modelling of brains.

Perhaps I will relate more deeply in future articles, but generally speaking the neural network research and the SIMD research have now merged into what is now known as transformer architecture which means that all the mathematics involed in neural networks, where possible, can now be done in parallel, making these computer modelled "brains" very fast and efficient. They are now so efficient that when you visit Gemini, Grok, Mistral or my personal favourite Claude, you are looking at a machine with and IQ higher than a traffic warden. Maybe once all the civil servants are out of work we can all get back to doing art for the sake of doing art.

Enquiries

Get in Touch

This little sub-application functions as an answering machine and it is my preferred method of getting in touch. Bear in mind, this is currently the website of Tessa Flow, and unless you say you want to speak to me, it will be assumed that your message is for Tessa (which will be answered by her using your email address to respond). I will simply tell Tessa what you tell me and then I will provide you with her response. Please note that talking to Tessa involves the waiving of your privacy rights in the sense that such correspondence will be published. I will obviously redact any PII from anything published.

You can reach me directly at my email address Stephen Duffy if you want to get in touch about the website software you are using now or any software I have published on this website as modules to this web application or otherwise. None of this stuff should considered open source unless stated otherwise.