Being everyone's friend at arms length and never being any interest groups best friend or getting to close, not bankster best friends, not benefits claimants best friends, not big businesses best friend not the public sectors best friend. You don't get much appreciation being a good chancellor but by balancing the books well and keeping all as your friend but none as your best friend the economy and all interest groups stand a good chance. In troubling economic times yes you might have to increase spending and more tightly balance the books but there's no magic bullet being more friendly or best friends to a particular interest is not going to fix the issue.

Creation of main data trait library and development of the main bot with takes the the input in the space and compiles it into the library.
Making The Pipe based on small efficient stages with dalle 3 and allowing for animation, far more artistic control, more efficiency with high detail language and more symbolic ability like emoticons.
Developing a bot and library generator from the main data trait library a space where large scale functionality converges with small scale functionality and training of neural minds small and medium. Generating a bot and developing the ability for a good functionality match relative to a lot of parameters.
in 3 years maybe about £6000 to generate a bot this way.
a 32 colour unistyle detail generation and function space convergence over quite a few stages to produce a small interactive data inference and calc simulation that you can then open and run on a basic computer using a simulation script and some English making it easier to convey data well.
in 3 years it could do a reasonable job costing me say 60p to generate 4 variations.
By 3 more years a mini PC could do some thing quite well and to make a half decent bot only costs £1000
By 3 more years after that.
The mini PC is getting a lot more capable with pretrained operations and on the cloud you can create a more decent bot for £400
By 3 more years
The mini PC has quite a few more bots operating at any one time and making your own Linux OS a lot easier will be possible,Personal Dynamic Game making is getting easier on a low watt PC.
At this point with £8000 you could produce a smaller mother bot you could then run on the cloud designed for assisting with production of an os coded your way and run that was totally tailored for you.
In the AI economy if the real issue is how and lot less about more mundane technical effort and more using technical insight with AI assistance to develop anything then I could be single man equivalent to main multi nation company. My problem is not how or being creative it's lots of mundane technical work required to complete so much creativity.

A force map function is where you map and upgrade in a second data set a neural force map such that you can use it during some neural training stages for better training efficiency.
Analogue photonic computing could allow for and better form map function fit as the dataset compiles to a small subset when effecting a zone in the neural network you are training.
This would improve overall speed and dynamic efficiency in training the neural network.
With reasonable quantum computing say 256 - 10,000 good Qbits may be 8,000,000 error correction Qbits something like that with not even enough to do much factorisation but photonic and running in parallel so lots of these types of operations per second making good use of quantum communication.
You could have more dynamic optimisation control relative to the force map dataset you may also have a quantum dataset enhancing the force map dataset and quantum functionality in the neural net dataset you are training.
This would mean your mind could be very robust and could Do quantum AI functioning which would help in better modelling of molecular down sub atomic physics it would also help with things like more robust and dynamic modelling of both Physics and scene dynamic in games and help with modelling better engineering design and product testing analysis.
This is what people don't realize, that with even where photonic digital and analogue AI acceleration is capable now and the maturing of EUV fabrication given ten years the AI could be quite capable for human driven more with a lot less mundane work. If quantum mechanics moves from 10-20 good Qbits upto to a photonic equivalent system with at least 256 good Qbits over the next 20 years we would still be able to do a lot of more optimal computation with less effort. So even if you take the lower possible outcome by 2042 you would have at least 80 good Qbit's with some level of photonic ability giving you some degree of quantum optimisation capabilities.
You would still have over 0.6 zeta flops of cloud compute resources Animations, Game making and much modelling and Simulation would be a lot easier to do a lot with.
TB bot making would be more resource intensive and of less capability that if we had got further by 2040 but the bots would enable people and companies to do a lot with less human effort but compute resources would be more tighter for big projects.
A top mini PC would still be able to do at least 0.8 Peta Flop ideal for many applications but less dynamic than if such a system had reached 50+ Peta flops by this point in time.
You would still have analogue being used to some degree to enhance training.
Computing after this point for decades to come would grow and prosper nicely allowing ever better models and capabilities and ever better production range ability and efficiency.

160p unistyle black and white Picture generator.
vocabulary 6000 words 80 program specific technical word functions.
First a wider space is more wildly spayed with text and detail trait
comparative sprays.
Then the spay is to shrink variation and detail closer to rugularized.
Then heavy regulation is done to the picture and then extra form detail
is sprayed as a wider space is compiled to a 160p top mixed res
1bit black and white picture with
detail manipulation and variation control which usually a
recompile takes less time more like 1 min for a medium
modification request.
Data trait library 2.2GB MAX needed per stage 6GB Storage
Brain 400mb max 1.2GB on storage
Process ram required 600mb 128mb GPU
Time to do 3 variations 6 minutes.
But hey it would be real AI prompt generation on todays PI400 if a top
data science team spent 8 months making the brain and optimizing
the data trait library.
Then you take the low definition 1bit picture and processes
it into a coloring frame and you use 100's of fast running mini AI bot tools
more like standard picture editing tools with an AI touch
To assist with coloring it in how you want it in 1080p full
colour. For example you can touch colors in and the AI blends the
coloring idea more than an exact tool use and you can search for
colouring in styles. then there is the add form detail to a region tool.
you can also add small regions of black detail then reform the data
better into the region of the picture in different ways.
vola AI assisted picture generation is possible on on a PI 400.
This means that if the industry had thought
about the issue more efficiently in the first place A reasonable PC
could have done this task range by 2008 and AI image generation
would be a lot more dominated by people who really touched up their work well.
So bigger models with more ability to do it all quickly from the prompt with
less artistic control would be less competition and Animation and video clip generation
with the artists touching many things up in a similar way would have been available by
2012 on the cloud.
With more efficient work in the area people would have been crafting good videos on the cloud by 2018 with a lot less effort, anything they wanted.
I always thought these researchers were more obsessed with more singular approaches using more resources data trying to run more before they could walk.

Further stage information for mother bundle data trait compiling and Training stages...

when a person of a interest group generates new data that they want to share with the AI, A Space is additively and de-additively and re-additively developed that is a Standard space for coding in a range of potentially useful data traits. Then a standard bot that 500,000 people work on intergrate's applicable data into the mother data trait library with the right intellectual rights.

When Training first through variation and a spread space the main functional structure is drawn in for life. from a variation of large functional range and the lower end of the space variation is small functional ranges it all draws into an optimal middle form as a bot you view on the bot browser balancing old data new data functionality and use. As an extension to an app through stages and the smaller initial seed eggs required are worked out whist some of the fruit is morphed into a more efficient data trait library. you do this in quite a few sub stages.

Next you develop the narrow brains required here you fold in your desired mind from a bigger mind space and it is some what coordinated with a more global local neural force map function with map update and the latest training techniques. You do this in quite a few smaller stages.
Next You go through quite a few stages of variation driven optimisation and the latest post workout optimisation stages.