Honestly ever since I started using LLMs I was always thinking of how good would it be to have a practical guide to picking the right AI model.
That's what this post is all about 😁.
tl;dr
- Pick intelligence over speed: choose the smartest model you can afford. Fast models save time upfront but cost you in debugging hell later.
- YOLO only with frontier models.
- Baby-step everything else, including critical services you wanna vibe engineer.
Note
I believe I've mentioned this a couple of times already, but just to be absolutely clear: do NOT YOLO a mission-critical service that will be deployed to production.
If you're deploying to a QA or development environment, and you're confident that even if it fails, the consequences will be limited and won't result in significant harm, loss, or damage to anything of real value, then sure, YOLO away.
Open-source Models
You must:
- Supervise it: keep an eye on the diffs.
- The prompt needs to be:
- Much more precise.
- Much more detailed.
- A clear specification as to what needs to be accomplished.
Also you might not be YOLOing excluding the risk factor. You might have to do more iteration. As of now I have NOT been using open-source models. But I like to know what you think.
But if you wanna use them like I do like to use them we only have to break a big semi abstract prompt into smaller manageable prompts.
Top comments (0)