AI & Cloud Meetup Karlsruhe - Prompt Engineering for Software Developers [DE] - David Farago
"AI will not take over your job, but people skilled in AI might" [Philip Hodgetts]. But what does that mean for software developers?
The application of large language models (LLMs) and the engineering of their prompts are producing wow effects and innovations on a weekly basis, but mainly for "open-ended question tasks", where the set of answers that are considered correct is relatively large, like the Duolingo conversation partner and AutGPT's web-based research.
For "closed-ended question tasks", which have more restrictive requirements, e.g. the generation of software engineering artifacts like code and tests, much fewer answers are considered correct. Here, LLMs often answer incorrectly, which is a show stopper for many of those tasks. Due to this problem, many emerging prompt patterns try to improve the correctness and other quality aspects of the LLM's answers.
This talk presents various prompt patterns (e.g. Chain-Of-Thought, external information, self-critique) and other means (e.g. a CoPilot UI) to improve the quality of answers for "closed-ended question tasks". The main part performs prompt engineering to iteratively improve the generation of unit tests for a given function implementation. We conclude that the correctness and comprehensiveness of the LLM's answers greatly depend on the prompts, the LLM, and how the LLM is applied.
After this talk, you will know the main weaknesses of LLMs, and how to cope with them. In particular, you will know various prompt patterns and what they can (not) accomplish for "closed-ended question tasks" like the generation of code and tests.
About the speaker:
Dr. David Farago is a developer and entrepreneur at the cutting edge: Doing model-based testing before it was widely accepted, agile software development before it was cool, cloud just because he can and AI because we need people that understand it, he always takes an academic approach in evaluating the newest trends in IT.
This talk was recorded at the AI & Cloud Innovation Meetup in our Karlsruhe office on July 20th, 2023.
