PROMPT ENGINEERING

→ View original post on X — @mattshumer_

Prompt Engineering: Better Results Working Directly With Models

By

@mattshumer_

–

14 December 2022 19h04

No longer using ChatGPT super often. As a diehard prompt engineer, I can nearly always get better results by working with the models directly. BUT For the average person, chat is the best and easiest way to get language models to do what you want. Conversation is prompting.

→ View original post on X — @langchain

Map-Reduce Chain Speed Optimization Through LLM Call Batching

By

@langchain

–

14 December 2022 17h54

This now means that certain chains see a MASSIVE speed up. For example, the map-reduce chain can batch the map calls to the LLM, resulting in a ~75% drop in time

→ View original post on X — @langchain

Calculating Token Count for LLM Context Windows

By

@langchain

–

14 December 2022 17h54

Estimating the number of tokens Each model has a context window with a certain length, where you can only pass in strings of up to a certain length There now exists a method on each LLM class to calculate the number of tokens for a string in that model

ChatGPT sacrifices in-context learning for dialog context modeling

By

–

14 December 2022 11h15

“ChatGPT seems to have sacrificed nearly all of its in-context learning ability to trade for the ability to model dialog context.”

Instruction prompting: powerful, but with performance and style costs.

By

–

14 December 2022 11h15

Instruction prompting is wonderful. Easier than k-shot for nearly every task, and tackles some tasks (e.g. applying large lookup/policy tables) impossible to specify fully through demonstration. But it isn’t free. You pay a tax in performance — and, I’d add, in style.

text-davinci-002 instruction tuning reduces in-context learning, boosts zero-shot ability

By

–

14 December 2022 11h15

“text-davinci-002, released in May-Jun 2022, is a supervised instruction-tuned model based on code-davinci-002. It is very likely that the instruction tuning on text-davinci-002 decreased the model’s in-context learning ability but increased the model’s zero-shot ability […]”

Wishing for editing responses in task-oriented chat UI

By

–

14 December 2022 10h36

Top of wish-list: Editing the responses. For task-oriented chat, prior task requests serve (elegantly!) as k-shot examples. But the UI prevents corrections except via token-wasting repetition or prose feedback. Playground is still a better feedback/correction loop for me.