No Sloptober

I just stumbled on the No Sloptober page and it sounds like a fun idea to try. I'm not sure I will, though. While some projects I work on require me to use AI for coding or process (WoW) related activities, most of my day to day things are technically also feasible without any help from the bots.

At this point, I have been a software developer for more than 20 years and I understand "the game". I know how to drive an idea through the process of making something useful, something that sells and I'm well versed in how to lead or leverage a team environment so it can be made into something more.

But coding doesn't always have an operational goal linked to it. Sometimes there are no KPIs. Sometimes, what the goal is just that, the process of building and the end result is something completely useless from a functional point of view. If I had fun making it or I learned something new, then completely useless can be a goal or means to an end by itself.

With that said, of course the reality of being involved in software engineering has changed. If you're writing code, reviewing code, building an architecture or setting up the technical vision for an organisation, LLMs are hard to ignore. And maybe they shouldn't be ignored. I think of "AI" as a fancy autocomplete loaded with tools that can choose to sometimes spontaneously execute themselves. LLMs are a bit like holding a fire cracker - you usually get some use out of them, but there are risks.

Over the last two years or so, I've had the chance to use all sorts of models (LLMs and harneses from different providers). Sometimes mandated externally by other parties as part of a project, sometimes for research into the capabilities of such tools. I've been able to combine my experience as a software developer with the pros and cons of using something like Claude Code or locally hosted model running on my computer. I think experiencing the technology from a critical point of view has allowed me to map the boundaries of what's practical (and ethical) use.

LLMs can be an accelerator when applied responsibly for tasks they have been designed to do. Understanding their inner workings (e.g. next token prediction over a large corpus of human-made content) helps one dodge the pitfall of believing AI/LLM outputs can be factual or that they can reason in terms of stakes or make creative decisions of any sort.

LLMs also come in various shapes and sizes - they are not only useful for parroting human creativity. Remember, there was/is an entire realm of ML tools which use fancy math to try (with some degree of certainty) to solve specific problems for which we don't have alternative solutions.

I think that for an AI/LLM setup to be successful, it requires the active participation of its human operator ... or rather a curator.

And so are you taking part of No Sloptober?

PS: Everything on this website is and will be human made.