Anthropic's prompt caching is such a game changer. We had a few ideas which needed huge system prompts in a chat-like interaction to work — but it was always too expensive/slow to reasonably run. With prompt caching, all those ideas are suddenly much more feasible! More to
PROMPT ENGINEERING
-
Distinguish Your AI Product: Better UX and Generative Capabilities
By
–
INNOVATION If you have good knowledge of your target audience, distinguish yourself w/ features: a) new interaction — build better UI/UX that fits into your users workflows better.
b) new generation — use AI that outputs new things, not just media a human could produce! -
New Free Course on Improving LLM Applications Accuracy
By
–
New course on @DeepLearningAI
: Improving Accuracy of LLM Applications https://
go.fb.me/zfwvd8 Created in collaboration with DL, Meta & @LaminiAI
, this free course covers topics like evaluation frameworks, instruction & memory fine-tuning, LoRA + training data generation. -
ChatGPT prompts to jumpstart online business
By
–
If you;re interested in building a business online then… Jumpstart your journey with these ChatGPT business prompts: https://
godofprompt.ai/blog/chatgpt-p
rompts-for-small-businesses
… -
LLMs behave like predictive text influenced by chat conditioning
By
–
There’s no dialog before this, to be clear — I didn’t ask it to pretend. LLMs are just like this. Though there’s some some conditioning from the implicit chat syntax (“user:” etc.), to a much greater extent than most widely used LLMs this model really is “just predicting text.”
-
Most models use instruct SFT and RL-based tuning
By
–
Most models that normal people have used in the past year (ChatGPT, Gemini, Claude, etc.) have some form of both instruct SFT and RL-based tuning. But yes I’m using “RLHF” inexactly in my post as a synecdoche for all post-training.
-
Pretraining and conditioning issues in instruct-tuned dialogue models
By
–
There’s conditioning from the dialog syntax that it’s being naively given in the same format that the instruct-tuned version receives. It’s seen these in pre-training, but the association isn’t strong enough apparently to make it act like a chatbot even most of the time.
-
Example contrasting base LLMs and RLHF-tuned models
By
–
I tried to write a prompt to show how base LLMs differ from the RLHF-tuned ones everyone knows, and I think this gives a bit of the flavor. A message from Llama 3.1 405B (base), on whether it’s useful to talk to base LLMs:
-
Discussion on RLHF vs post-training model comparisons
By
–
You’re not comparing RLHF to no RLHF here, you’re comparing different generations post-RLHF mystery models, likely of different sizes. If you don’t do post-training at all, naive attempts to talk to the model go like this:
-
FOXY Method: Self-Improving Agents Design Pattern
By
–
I'm starting to see some more projects and research around self-improving agents, so thought I'd reshare this design pattern I introduced last year. The "FOXY method" tracks output, generates a reflection, and feeds the most relevant ones when running future objectives, to help