Summing up AI 2023

So the last 12 months have been amazing, if not rather dramatic with regards to AI. Things have improved a lot and we will see and hear more of this over 2024 I’m sure.

These are some of the things that I’ve found interesting…

We of course have had the big dwrama over at openAI, with Sam Altman being fired / Quitting? and then the board being fired and Sam getting his job back. Rolling stone has an interesting write up about this.

Everyone is wondering and predicting what this Q* (Pronounced Q star) product at open AI is – some think it may be an AGI (Artificial General Intelligence) but very few people have had access to this product so far. Although there is a lot of speculation.

We still don’t know what’s happened with googles LAMBDA and Blake Lemoine is still I think the canary in the coal mine with regards this technology.

The issue of building your own “Bad version of chatGPT” is a very real possibility. We should beware of Bad Robots! And I mean Bad Robots- you could theoretically hook your own bent version of chatGPT up to a mechanical device and let it lose(Who knows what the military are up to with this idea).

In addition to this is the issue of copyright and the fact that everyone is ignoring the importance of related links and knowledge that back up the statements made by AI (not to mention the issue of AI hallucinations) . Although it is possible for these platforms to supply and reference sources, most of the commercial products don’t include this functionality. This I think is going to pan out in interesting ways. Already a number of Authors are attempting to sue OpenAI.

But I think the most interesting thing you could do is build your own chatGPT and train it on your own data. I’ve set up something on an old machine I run my self and gave it a number of my old blog articles and various other bits and bobs to play with. The results were solid and interesting (with references!).

But two things come to mind with regards this. You need CPU and Ram and ideally a few GPU’s to run this sort of software. In short a grunty machine or an expensive virtual machine, that runs at a reasonable speed (although I did manage to get this running on a machine with 4 cores, 8 gig of ram – it was very slow) but the scenario that comes to mind is this.

If you have your own company and fast access to your own data (files, emails databases, financial data etc), and can hook it up to your machine, you could probably gain all sorts of interesting insight. What was the most profitable project? How many emails were sent? What were the time frames for this project. These and a whole lot more questions could be asked about your data. The stinger comes though when you get around to the speed of the computer running this and the connectivity of your expensive AI brain to the content.

If you only have a 100 megabit to all that data sitting in the cloud, it’s going to slow things down. If you have invested in local hardware (and say have 10 gigabit or more connectivity) to your data your going to get results much much more quickly. I see an argument for employing your own sysadmin percolating!

In short 2024 is going to be just as crazy as 2023 it’s sort of amazing to be alive and witnessing all this. Thanks for reading and stay safe over the holiday season!

Steve

 

Related links quoted! _____________________________

Blake Lemoine
https://www.newsweek.com/google-ai-blake-lemoine-bing-chatbot-sentient-1783340

WTF Is Happening at OpenAI?
https://www.rollingstone.com/culture/culture-news/sam-altman-fired-open-ai-timeline-1234889031/

This new AI is powerful and uncensored… Let’s run it
https://www.youtube.com/watch?v=GyllRd2E6fg

Authors sue OpenAI over ChatGPT copyright: could they win?
https://www.businessthink.unsw.edu.au/articles/authors-sue-openai-chatgpt-copyright

Let’s build GPT: from scratch, in code, spelled out.
https://www.youtube.com/watch?v=kCc8FmEb1nY

 

Ai for work and home

As AI is starting to become something many of us are using, it’s interesting to think about the possibility’s of getting an AI to look at our own data.  I’ve set up privateGPT on an older computer and given it data to “ingest”!

It does take some time and it needs a decent amount of ram and cpu grunt, but it does work surprisingly well even on under-powered machines. It also gives you links to source documents you may have got the device to ingest (unlike some products I might mention… cough!).

There is a part of me that thinks that this sort of product could be very useful for insite for a private company (or even an individual). Think about the possibilities of giving it access to all emails sent. What could it learn? All files on the server, who created the most files? Who was involved in which projects and what were the  skill sets of the people involved (past and present).

Lets also think about the possibility of an AI having access to production data and financials!? What could be gleaned from all that information. It’s been said that we will soon have personal versions of AI that can run on phones. I’m sort of looking forward to this … but having recently re watched the most excellent Ridley Scott movie movie Alien Covenant, I’m recommending that we proceed with caution!

 

Battling Digital Onslaught

A couple of times on this site I’ve mentioned some of the garbage that some ip address spit at my wee server. It’s interesting and frustrating especially when you think about how much people do or do not care.

Admittedly the people at AWS on occasion have constructively addressed the odd concern (but they do require lots of information inducing your time zone preferred type of porridge etc) But on the whole if you make a complaint to an ISP about the scanning or bs coming from an IP address you often never get a response.

So one approach is to just set up firewall rules and hope that the Aholes don’t come back via another address, but this does become a vexatious game of Whac-A-Mole.

I made this project public in 2018
https://github.com/nevetsanderson/sshfail

It is useful in that it on a day to day level can bring awareness to how much junk is rattling at your front doors. I remember telling a manager once about the number daily number attacks and he went rather pale in the face.

I hope that someone can find sshfail as a useful tool, even if it does turn managers pale.

Patterns of random

Ever since I started my adventures with computers, I’ve been fascinated by the process and delight of random. Of taking something so precise  as a computer and having it cough up random information. I think my first experience of this was with the venerable “Hypercard” running on a mac. Danny Goodman wrote a wonderful book on how to use and write for this product and it was a good initial introduction to programming.

I was thinking about how creativity contains elements of random – or new concepts (created because of random actions) and how AI is becoming rather good at defining, refining and regurgitating content to create the new.

Machine learning, giving a set of criteria and a desired result – infinite testing and the ability to weight or to project concepts. To test at an incredible rate – add to this the possibility of quantum computing and humanity is at a very interesting nexus.

Recently a group of researchers hooked an AI up to a robotic arm and sensing device (effectively giving the AI access into the physical world via the robotic arm and something similar to vision) when asked to identify an extinct animal it picked up a plastic model of a dinosaur. When asked what object might be the best for hammering a nail, it picked up a rock.

We live in interesting times.