[kingkongjaffa]

You don't _NEED_ to know it, but it's probably good for your career to have some exposure in 2025 to LLMs.

Download ollama and mess with a model locally, implement your own RAG system so you get a feel for what that entails and what good and bad use cases look like. I use LLMs every day for random stuff, but not really because I need the output, more because I need to know and understand where it's strengths and weaknesses are.

You're not trying to become an overnight expert here. Nobody expects that. But there's this gap right now between what stakeholders think AI can do (basically everything) and what it actually can do. And guess who needs to bridge that gap?

As a software engineer part of your responsibility is to advise non-software engineers on what is just now possible with technology, because when a stakeholder comes to you with some wacky idea, you need to be able to judge it and decide on the investment of time the idea might require, or if it's even possible today/now.

[TZubiri]

>Download ollama and mess with a model locally,

Ollama locally is very slow (or low quality). I feel a good middle ground is renting GPU or TPU per minute and running a local model there.

[LorenDB]

Not if you have a gaming GPU or a recent Mac.

[mattmanser]

The op's not talking about running a local AI.

That's pretty much a complete waste of time as the op is not a developer.

The op's talking about using AI in their normal work workflows.