My use of AI in Jan 2025
In December OpenAI announced that o3 would soon be available, making it very clear that things are going to change soon—indeed, are already changing. It's not going to viable to avoid using AI in profession settings for much longer, any more than it is to avoid using a search engine: it might be fine for casual and social situations, but if you're fundamentally unfamiliar with the technology you'll be at a big disadvantage.
So one of my themes this year is AI
: making better use of AI in my daily and professional life.
This article describes my current use cases in January.
My first use case is asking easy questions, where I don't expect to need in-depth information, and can afford for the information to be a little bit wrong or out-of-date. Some examples of this are: language questions (what's the difference between ad
and vesz
in Hungarian); word-choice questions (what are some synonyms for discovery
); and high-level brainstorming (what types of hobbies are good for brain health).
I don't expect these sorts of questions to lead to a drawn-out conversation, and they could also be answered by a web search or a traditional resource like a dictionary, textbook, or mindmap. So they don't need a high powered AI system.
For these, I use Windows Copilot. It's available on the desktop with an easy shortcut, and is free to use.
Some questions are more detailed and I want accuracy, normally when I'm trying to make a decision. For example, maybe I'm deciding whether to install solar panels even though my house doesn't have a south-facing roof slope: I want to know energy production figures and costs of panel installation, and they need to be accurate and up to date.
For this type of question, I use Perplexity. It costs £20/month, and is very good at citing its sources and giving more detail. It still can't be fully trusted, but then neither can the sources it's using—it's just harder to detect a bad source when you're another layer of indirection away.
I do have concerns about Perplexity's long-term future. The big AI companies all have deep pockets and can cover losses for a long time, but Perplexity is running a much tighter operation, and they've started having sponsored suggestions for followup, which could degrade their use if they get carried away.
For more in-depth thinking and planning I use the latest Claude and Gemini models via the TypingMind frontend. This frontend is set up to use the pay-per-token APIs and is easy to switch between back end models and custom system prompts.
I find Claude to be better at following system prompts to take on a character, whereas Gemini feels a little bit snappier all around but less flexible.
I've had the most use here from the AI in getting it to lead me through a long processes, and generate templates for me to fill in, something to break through blank page syndrome
. I probably could have used traditional methods here too—like randomisation or combinatorics—but having the AI write a customised planning template seems easier.
In theory coding is a huge area for professional use of AI, but I don't have a good solution here. GitHub Copilot is always available, but it's not a game-changer, just a boost in typing speed. Claude works for small scripts or Q&A, but I've not yet used it for large projects.
This is probably the most important area for me to develop!
I don't use voice chat much, but when I do ChatGPT advanced is the only real option.
That's my AI use in Jan 2025! A bit of Windows Copilot, with Perplexity for research and Claude & Gemini for branstorming, and a bit of GitHub Copilot for coding. Hopefully over the next few months I'll be able to add a bunch more use cases to this list.