Post by @troed@swecyb.com

2026-08-29 18:35 UTC

@seanm@infosec.exchange Let me quote myself: "consider LLMs to be an absolutely fantastic tool for all use cases that have a clear OK/NOK gate. That involves cybersec, sysadm, software dev, maths proofs etc. It does not include transcribing doctor's notes into patient journals, as a counter example." LLMs are horribly bad at being databases. No one who knows anything about computer science thinks otherwise, no, not the people creating them either. Try it yourself: A good capable local LLM is about 20 GBs in size. With what you know about compression (Shannon's theorem) - can that file include all the knowledge on the planet? @briankrebs@infosec.exchange

Replies (1)

@seanm@infosec.exchange 2026-08-29 18:59

@troed@swecyb.com @briankrebs@infosec.exchange you're right! I was not using GenAI for sysadmin or software dev as you clearly stated were the only acceptable areas for GenAI. I was using it to develop software and admin a database. These are clearly different and do not fall under your OK/NOK usage cases. I'll keep this in mind and hold my phone differently next time. You're also right that GenAI is not good as a database. I'll let Google, Microsoft, Facebook, OpenAI, Anthropic, Twitter, the US government, and every other major GenAI company know that they're also holding their phones wrong and using GenAI wrong. They can also stop putting GenAI into everything and they won't need as much computing capacity. Phew. They don't actually need more datacenters! #GenAI #LLM #AISlop

Open ##5141533