It’s been a long time since I posted here. Seven months!
I’ve been trying to figure out the AI puzzle. Like many/most people, I’m regularly using LLMs. They’ve replaced search, which is a good thing, because search was already dead before LLMs took over the world. I’ve been saying for years that Google’s search business was vulnerable. This might be a good replacement, though I’m not entirely sure. The jury’s still out.
The AI puzzle is finding a way to use AI to (a) make your work life easier, and (b) make yourself more productive at work. There have been some small gains here and there, but by and large I have not managed to find ways to make myself more productive.
The most important breakthrough from a productivity perspective has been the appearance of highly capable open models that you can run on a regular computer. I’m trying to find a workflow that helps me deal with the deluge of email called my normal work day. I could blow through a bunch of money buying services to help with my email. I’m never going to do that. For one thing, I want my email to be private, even if that depends on the other guy as much as it depends on me.
A real, true, genuine AI productivity boost is that I can use it to write scripts that automate all kinds of stuff. I’m in the process of developing a script that will read in a directory of email messages and have Gemma 4 summarize and classify the messages. It’ll output everything in a nice little html file that can be used to do further processing (archiving, creating tasks, etc.) I’m not there yet, but if this does help me to tame the ugliness that is my email inbox, it’ll be worth it.
I’ve found LLMs are awful for writing. I might have them write up an outline of things and then expand on them. Generating blocks of text is a sucky substitute for writing by hand.
There are some legitimate uses for programming. The problem is that they still get a lot of stuff wrong. Much progress has been made since 2024. Nonetheless, even the best Claude models have their share of stuff that either isn’t right or is suboptimal and hard to read. Writing documentation, creating packages, and making things cross-platform are all big wins for productivity.