prompt engineering

Discussion is dominated by Hacker News practitioners debating prompt engineering's legitimacy and practice. Defenders argue it 'works, and that alone is a reasonable reason for people to explore it' and that 'most people are thoroughly incompetent at giving instructions.' Critics contest the name: 'Prompt crafting is fair, informative, and accurate... craftsman/artisan != engineer.' Practitioners report concrete wins — augmenting prompts 'with SERPApi, embeddings and a vector db, brought really flakey results into the >90% accuracy threshold' — but flag a severe tooling gap: one builder 'run[s] manually on 50 different test cases' per prompt change, calling it 'extremely time consuming,' and asks for 'performance monitoring... but for prompt quality.' Sold prompt packs are mocked: '46 Revolutionary ChatGPT prompts' is likened to non-generalizable bridge calculations; another user says most basics are 'pretty easy to figure out on your own if you play with ChatGPT for a few evenings.' Risk stories recur: Reddit's top thread (+5,530) jokes 'Just like a junior developer!' about an AI that followed a 'Purge everything before installing' instruction and nuked online backups; an HN commenter blasts Brex for using a 'non-deterministic model' in 'production use cases' (expense tracking) with hallucination and privacy risk, while Brex's own Office-of-the-CTO author defends the practice in the same thread. Prompt injection is framed as 'communicating with demons.' Technical discussion covers few-shot/in-context learning, ~99% reliable JSON generation for simple schemas, and token-cost trade-offs ('More tokens, better response?').

No aggregate ratings were found for this product during the last harvest.