20250522 - You can’t feed gen-AI on ‘bad’ data then filter it to give only ‘good’ data

Pivot to AI

Everything was always ‘prompt injection.’

Text version: https://pivot-to-ai.com/2025/05/22/you-cant-feed-generative-ai-on-bad-data-then-filter-it-for-only-good-data/

Please send money! It’s very helpful!

Patreon: https://www.patreon.com/davidgerard

Ko-Fi: https://ko-fi.com/A1529D5

Buy me nice things: https://www.amazon.co.uk/hz/wishlist/ls/3Q8VZW46J6DM6

Get an extremely cool Pivot to AI shirt: https://pivot-to-ai.redbubble.com

Sources:

Dark LLMs: The Growing Threat of Unaligned AI Models https://www.arxiv.org/abs/2505.10066 Novel Universal Bypass for All Major LLMs https://hiddenlayer.com/innovation-hub/novel-universal-bypass-for-all-major-llms/

Simon Willison’s Weblog: Series: Prompt injection https://simonwillison.net/series/prompt-injection/

Most AI chatbots easily tricked into giving dangerous responses, study finds https://www.theguardian.com/technology/2025/may/21/most-ai-chatbots-easily-tricked-into-giving-dangerous-responses-study-finds

Previously on Pivot to AI:

Elon Musk releases Grok-2 AI image generator for edgelords https://pivot-to-ai.com/2024/08/15/elon-musk-releases-grok-2-ai-image-generator-for-edgelords/

Whoops! Facebook trained Llama AI model on pirate site LibGen, with Zuckerberg’s OK https://pivot-to-ai.com/2025/01/10/whoops-facebook-trained-llama-ai-model-on-pirate-site-libgen-with-zuckerbergs-ok/

How to make a splash in AI economics: fake your data https://pivot-to-ai.com/2025/05/17/how-to-make-a-splash-in-ai-economics-fake-your-data/

video: https://www.youtube.com/watch?v=Y47jBvAN4DY&list=UU9rJrMVgcXTfa8xuMnbhAEA

Wikipedia: Common Crawl https://en.wikipedia.org/wiki/Common_Crawl

Full Pivot to AI playlist: https://www.youtube.com/playlist?list=UU9rJrMVgcXTfa8xuMnbhAEA