A Geek's Life

(Gen) AI and its impact on life, the future of work and pretty much everything in between has been on my mind for quite some time, more so of late. It hasn’t helped that Co-pilot has been rammed down our collective throats at work for the better part of a year, in the name of digital transformation. This is perhaps why @iamgregb’s post How I Use AI resonated deeply. Two other posts I have read recently also touch on the same topic from different aspects, Cafelog’s AI Compass and AJ Jacob’s [Update on] How I Use AI.

The main thrusts of @iamgregb’s post, as I read it, are that:

• AI is a (partially deterministic) machine
• Given the above, it is best used for structured work – formats, metadata and “doing the boring connective tissue work between apps”
• Personally, @iamgregb doesn’t use AI for creative work or for finding meaning.

This aligns with a framing which I read somewhere but alas can’t find a link to. As I recall, it considers AI (LLM) use cases in the knowledge economy world as falling into one of summarisation, interpretation or generation.In this framing, LLMs work very well in summarisation, barely passably in interpretation but terribly at generation (of new knowledge).

Working as I do in the so-called knowledge economy, the impact of AI on jobs is a very present concern, and something I have reflected on as far back as 2021. I remain convinced it is a matter of when, not if, that a combination of an LLM, a system that stores knowledge (eg a knowledge graph or graph database) and a suitable orchestration framework could passably replace the data ingestion and analysis bits of my day job. There will always be the need for explainability which will require judgement of the domain expert. For now at least, the domain expert’s bell has not yet been rung, at least so says Ethan Mollick.

My current AI setup, if I may call it that is a Claude Pro subscription for a year, after which I intend to evaluate the value and then decide if I want to re-up, pretty much what AJ Jacob’s is doing.

So what am I using AI for:

• Cartoonifying pictures: Primarily ones which include my family as a means of blurring/obscuring/ changing them just enough to preserve some privacy
• Artifacts: For creating dashboards and/or visualisations of data I capture as part of my continued (light-touch) quantified-life experiment.
• Simple coding tasks: Such as migrating two old blog archives (an old the3six5ng project and the 2007 to 2024 version of this blog) to cloudflare pages.
• As a quizzing agent: As part of my job searching in the background/ being a tactical slacker, I have created a Claude Project where I periodically get quizzed about key bits of knowledge associated with my day job as a means of revision and keeping mental atrophy at bay. This is probably the biggest use of my plan so far.

Nothing particularly earth shattering then I think but keeping my AI use to the routine and structured, and outside that, areas in which my domain expertise allows me query the output and push back. My small experiment with asking Claude to review my investment portfolio suggests it has some way to go!