Great responses to this thread, thought I'd collect them in one place with my thoughts 👇
x.com/hrishioa/status/1820446545998917817
What's needed for good engineer to pick up enough basics on LLMs (if they've been meditating under a rock like that guy that missed Covid) - enough that they can read the more advanced guides
1. promptingguide.ai/ from @omarsar0 is one of the best I've seen (h/t @EdSealing)
2. @simonw's blog (simonwillison.net/) is a great resource, but you'll need to do multiple passes if you're starting completely fresh on the topic. Once you're spun up it's invaluable. (h/t @arjunram who also has a list on useful articles)
3. @AnthropicAI's prompt guide has not staled as much as OpenAI's. It's hard to keep these things up-to-date the speed at which models change. It's still a good resource: docs.anthropic.com/en/docs/build-with-claude/prompt-engineering
Also partly stale is my own prompting guide (ty @ChrystuS8V) olickel.com/everything-i-know-about-prompting-llms
Staleness is the main issue with a lot of resources. Honestly I can't even tell which parts of my guide need elaboration anymore, since you eventually fold in working with LLMs to a deeper part of your brain that isn't actively doing it, much like internalizing a new language.
4. Once you're partly through learning the basics, the @huggingface cookbook is a great place to start. HF as a resource is hard to navigate, but a lot of what you want to learn is in there, especially outside just LLMs.
(ty @josephpollack)
huggingface.co/learn/cookbook/index
5. @iamwil's work looks insanely creative and well-done. Haven't tried it myself yet, but it's worth a look!
x.com/iamwil/status/1820484964473024809
6. @karpathy's videos are wonderfully done - and some of my favorites - but I worry they might start at level 2 or 3 for AI Engineers, instead of zero. Which I get - it's hard to start at L0, but there's value: almost all of us learned for loops from someone who taught it to us.
Then you have conceptual articles which don't stale as easy, but are rarely specific enough to be directly useful. Best example might be my RAG series (huggingface.co/blog/hrishioa/retrieval-augmented-generation-1-basics), even though the point of RAG alone has now changed because of context windows.
Still looking for good resources that are:
* Engineering oriented, working with models instead of on models
* Starts as close to the beginning as possible
* Up-to-date with current AI
Not sure I'll find it, but open to any and all suggestions! I'll add more below as I find them