고정된 트윗
Advice given to someone asking about AI Consulting:
I don't think an ML background is required to be successful in AI consulting, but obviously helps. I think the biggest "skill" learned in ML is how to successfully do feedback loops in a system.
In an ML system, this typically involves cleaning data, making model tweaks, performance evals etc. In LLMs, in nearly every case you won't be fine-tuning the model, but iterating on prompts is a very similar workflow.
I do think it would be helpful to at least get a high-level learning of how the models "actually" work and become familiar with the basic terms. e.g. tokens, transformers, attention, what happens on each input -> output iteration as the model is predicting. You don't need to know the underlying math (helpful though), but having the understanding of what is happening is helpful.
Most of the AI consulting market is more on full-stack / product development skills and less ML. This isn't the most lucrative opportunities, but they are available in abundance.
Major areas now and over the next year:
- RAG: this is basically just glorified search lol. Useful in many contexts but severely overhyped
- Agents: The models aren't quite there yet IMO for this to be useful, but in 2025 I think this will be a major theme and a HUGE area of interest/investment. Becoming good at this will be valuable.
- Evals: Performance evaluations are a relatively untapped market. Most AI products you see today are flying by the seat of their pants. Without eval metrics, you can't truly know if your prompt changes are improving the system. This is somewhat more difficult to sell as a consultant as it requires a more sophisticated buyer, but is worth a lot of money if you can do it well
English
