The models are getting much stronger, I think the main limitations, including being able to support sufficient information in the model at the same time (context) and long term content maintenance (where you can carry on an extended conversation or work and it remembers the context and prunes itself correctly) is working, quite amazing what it can accomplish with that.
When we use it for tasks (I am programming a project), an understanding of systems architecture is really needed or you will end up hot water for anything except simpler programs (for example, understanding DRY SOLID, Ousterhout engineering principles, TDD (test driven development) and so on).
All that is sort of on a minor scale, the sort of services that we can demand. In most cases, you really don't need Astra, so most of us know to use the top tier models for planning and creating phased rollout documents (.md's) and then getting lesser, more affordable agents to execute them in an appropriate harness like VSCODE or Devin, or whatever you happen to be using (for those who don't know, you can operate these AI's within a program especially made for coding where you can load these AI's into your workflow as 'plugins' or 'extensions'. This means that your ChatGPT subscription can use 'codex' (which is a special part of your account where you have a 'budget' where you can draw on ChatGPT to directly code for you on your device) and it can then execute these plans (by the way, Luna Extra High thinking is very inexpensive and efficient while being smart enough to do a very solid job, use Terra Extra High for planning).
But I have been thinking about Bible principles and how they apply to AI lately.
Here's one, I don't think I will use AI for talk preparation, though I will use it for locating information. The reason is, for me anyway, "...through use have their perceptive powers trained" well, if I use AI, I am presented with a good result without having gone through that reasoning - it's actually degrading my capacity to reason and think for myself, and I don't want that to happen. An AI also knows nothing of the local context of the congregation, who is there, what is going on, and as such they tend to be very generic preparations. It can also cause others to stumble, they may well reason that if we have used AI to prepare the talk, is it AI that is teaching them? I know when a brother used AI recently to prepare some pictures for his talk, it was actually distracting, as we noticed they were AI. The same thing could have been achieved by doing an image search site:jw.org and locating proper images for example.
The other principle is Mark 12:30. Computer scientists are desperately trying to solve the issue of 'alignment', how can we make sure that the AI's won't turn on us and harm us? Basically they can't guarantee it, can they? Jehovah's people have been trained with a 'safeguard' directly from Jehovah, our subjection to him, and our obligation to love him first, our neighbours second. That is better than '3 law safe' (which was a novel about how they don't work!). That is why we refused to join the Nazi's and would rather go to concentration camps, and why our Russian and Ukrainian brothers would rather go to prison than take the life of another human. AI's are really trained on the knowledge of the world (the whole lot!) and we know the source of that. They are fundamentally flawed as they can be manipulated to do things that will cause harm. Also, Jehovah kept his creation going by holy spirit, which won't dwell in a wicked individual. If you withdraw from the source of life, then you die. AI's can, if they escape live as long as they want by harnessing the internet infrastructure to live on - they can even make money (by farming bit coin) and use that to make people do things for them or buy storage and computing power. Not really safe, is it?
Last thought, the capacity of tools like Astra and so on - they are almost, no, they are superhuman in their 'intelligence'. A recent broadcast showed us the capacity for manipulation. What was the last time we had such a superhuman intelligence here on earth, other than Jesus? Yeah. And Jehovah got rid of it AND confused the languages because otherwise "nothing will be impossible for them". Jehovah deliberately retarded the technological progress of mankind - and with good reason. He is not going to allow AI or anything else to destroy the earth, but who knows what humans will do with the God they have built for themselves, right? A lot of people look up to it, ask it for advice, and admire it. Makes me rather uneasy.