· AI Engineering · 3 min read
How to Keep Up with AI Without Going Crazy
Pick one or two areas, build a knowledge mesh, and test new tools yourself.
New models, new tools spawn every day. It is easy to get overwhelmed with all the new cool things. And the over-dramatization “Ho li shiet, this change everything!!!” of the online community doesn’t help much either. I share here the way I keep up with AI progress without getting crazy.
There are many areas in AI. Here are some popular ones:
- Models: frontier models, image generation, video generation, multimodal, small models, open source models
- Tasks: coding, writing
- Techniques: RAG, finetuning
- Domain: Academic papers, Enterprise deployments, SaaS products
Following the news daily in all areas is not feasible, unless that is your full-time job. Better just pick 1 or 2 areas that you like or are related to your career the most. For me, it is frontier models and coding.
Then for these areas, you should acquire the skill to be able to independently assess if news is important or just meh. Without that skill, you will be buried in hundreds of “this change everything” or “blah blah is over” each day. To have that skill, you would need 2 things:
If you don’t have a strong foundation knowledge of that area, you could spend time learning important concepts, how they are related to each other. You will build a mesh of knowledge in the area. That also helps you know what the state of the art is in that area, and why.
Try out all of these state of the art models/tools and have your own benchmark to evaluate them based on certain criteria. You use each of them for a while to understand the good, the bad, and the quirks. And see if your experience matches what the community or news says.
Having those two, you scan through a lot of news and updates daily and know what is worth your time digging deeper, what is important to deserve a node in your mesh of knowledge. You also know which sources provide good content, and which sources are pure nonsense.
For other areas that you don’t focus on, e.g. video generation for me, you can skim them and use your own judgement, and that skill is relatively transferable across areas. You don’t need to do that daily as over time, the good ones will stay and keep being mentioned, while the bad ones will die down due to natural selection. And because you don’t use them often, it is okay to miss some models/tools, as new and better ones will spawn regularly.
In short, I just go deep on a few things and let the rest flow past.
