I struggle to write

I have about twenty articles that I started writing and keep hitting the wall about 50-60%. I’m pleased with the beginning but I keep losing the point.

Perhaps I’m trying too hard, so I’ll try a more fragmented form. This has been inspired by Martin Fowler’s Fragments^[1]. Perhaps it will work better while I’m trying to figure the bigger things out.

Computing club - suggestions welcome

I have offered to join the computing club at my children’s school. I visited once to have a look around and talked to some students, and then returned last week.

I would like to bring AWS DeepRacer and prepare racing for the children. Now that the DeepRacer on AWS is available as a solution^[2] I can bring the training and try to prepare the cars as well. But first, I need to teach the children about the neural networks.

This is where I need some help: what would be your recommendation for training resources, and subjects?

First lesson:

• We started with some basics about what AI is, what AWS is
• We drew a conceptual model of a neuron, a natural one, followed by a programmatic one
• We talked about what a neural network is, how the neurons link up and how they do simple maths to do the fancy stuff
• We talked about encoding the inputs and decoding the outputs, using AWS DeepRacer as an example:
  • The camera takes a picture and turns it to grayscale
  • Each pixel becomes a number in the 0-255 range
  • Maths, maths, maths
  • We get a bunch of numbers with value between 0 and 1
  • The highest value in those numbers represents an action
  • The car takes the action
• We talked that training happens (without detail)
• We mentioned transformers, and how (vaguely) AI chatbots respond to you
• We talked a little bit about reinforcement learning
• We talked a little bit about GANs and deep fakes
• We talked a bit about how the responses we get look like an answer, and how they can be wrong and should not be trusted just like that
• We talked a little bit about the natural and formal languages

I have used AWS Kiro and prepared a learning resource: https://neuralearn.mediocr.is/. Luckily we couldn’t use it last week because IT blocks pages from unknown sources. Luckily because I now know my children cannot do whatever they like on school computers, and also because I had time to iterate on it and make it more engaging. I’m bringing it in this week, and IT have unlocked access. Hopefully we’ll get it running now and the students will enjoy it.

I will need to get a DeepRacer up and running, and learn to run the training environment. Lars Lorentz Ludvigsen, a Community Builder and my friend from the AWS DeepRacer Community, has tested the solution. Now I need to dust the cars, pick up the track, organise a training and racing.

AWS Kiro: my projects so far

I have mostly been focussed on preparing static pages using Hugo or just plain html and js. There are aspects of working with Kiro that I enjoy very much: I can try an idea and withdraw if it doesn’t suit me. The second aspect that I appreciate is deployment. With Kiro’s help I started using CDK for deploying projects. I know this is a simple use case, but I find it very valuable.

I know the projects are not exactly production grade, not the way I would appreciate it. I do not have the knowledge to reason about many of the code outputs that I have experienced. That said, they are good enough for a single-person personal project. I learn in the process. I would not have been able to, for instance, improve accessiblity of this page on my own. I would’ve failed miserably trying to visualise a neural network for a page.

Above all, I see AWS Kiro (and the coding agents in general) as an accessibility tool. They help me focus, or understand that I don’t understand what I want. They help me to get past the wall of awful that would’ve normally leave me stuck and not sure how to make a project a reality.

I do not have large personal projects, thus I can only make Kiro sweat on the work stuff. So far I am mostly pleased with it, but I feel I haven’t pushed it enough just yet. More learnings to come

Most problems are people problems

Adrian Hornsby shares his biggest struggles from a year of running Resilium Labs^[4]. The most insightful one: organisations treat organisational problems as tech problems, and this doesn’t work. Only once it is acknowledged that the technical solution is only a part of a much bigger system, progress can be made.

His book “Why we still suck at resilience” is on my reading list.