@timbray Nice, thanks for sharing.

Definitely something to think about even for those of us trying to be cautious/responsible about getting near this stuff.

Something we seem to have is yet another extension of the reach of mis/dis information - and what else is poorly structured, not-fit-for-purpose code? - through the boundaries and guardrails that were once around these workflows.

@timbray

Lol only for people stupid enough to fall for it. The ratio of value to ruinous horror is quite poor.

@timbray i’ve hit some of these, and the entire article resonates. it’s so easy to get on the dopamine drip of apparent execution that it becomes really hard to make sure you’re double checking the outputs and what it’s actually doing. ideas are too easy to run amok, when they should be experiments and then refined.

the best solution i’ve found so far is structure around features and validation at multiple levels, including adversarial feature review. it’s a hell of a new challenge!

@timbray Are these people _going_ insane _after_ "spen[ding] two months excessively prompting the thing and wasting tokens [and] creating a ton of tools I did not end up using much" or there something wrong with them when they start on that dark, downwards path?

@timbray it only just occurred to me that the kids growing up today with all of this are GenAI