Daniel Reeders

Wow people do NOT understand how LLMs work.

People are asking LLMs to generate secure passwords for them.

https://www.theregister.com/2026/02/18/generating_passwords_with_llms/

'Hi ChatGPT! Using your massive matrix of all the tokens you've seen strung together before, please generate a combination nobody has ever seen before?'

Your AI-generated password isn't random, it just looks that way

: Seemingly complex strings are actually highly predictable, crackable within hours

The Register
Feb 20 at 1:37amWeb
Show threadDaniel CarosoneFeb 20

@onekind exhibit #289173 (subcategory, finding out but not learning)

https://timeloop.cafe/@Taweret/116100018229982389

4 8 15 16 23 42 (@[email protected])

Attached: 1 image so you did the work of making a whole ass tool to help a robot write your novel for you instead of just writing 'your' novel yourself

the Timeloop Café
Show threadDaniel ReedersFeb 20

@uep Okay so that person isn't paying enough for a GPT with a large enough context window to remember what happened three chapters ago.

https://genai-arcade.net/20241011/short-term-memory

I wonder if you could enchain a RAG to a GPT so the GPT can ask the RAG to retrieve specific plot details it needs to remember and can't find in its context window?

sickos-yes.png

GenAI Arcade

Welcome to the GenAI Arcade - your refreshingly honest way to learn about AI! Skip the jargon and doomsday scenarios. Instead, poke, test, and play with AI through interactive games that reveal what it can (and can't) actually do.

Show threadDaniel CarosoneFeb 20
@onekind I mean, sure, yes, but
Show threadDaniel ReedersFeb 20
@uep I just want to distinguish 'AI can never do this' from 'better AI could do this' (but it will boil the oceans)
Show threadJonathan Poh 📷 🏳️‍🌈Feb 20
@onekind ugh. That’s such a bad idea.
Show threadDaniel ReedersFeb 20
@jpoh 'Let me ask an agent that can't tell time to act as a random number generator.'