vaughn tan

@vt

299 Following

1343 Followers

vaughn tan pfp

lucky bb 👏👏👏

0 reply

0 recast

1 reaction

vaughn tan pfp

0 reply

0 recast

1 reaction

vaughn tan pfp

what will you do with it

1 reply

0 recast

1 reaction

vaughn tan pfp

sign me up #homecooking

1 reply

0 recast

1 reaction

vaughn tan pfp

humans can do it, but are bad at it unless trained in discernment of subjective value and kept in continuous practice — and the more we let machines do the decisionmaking about subjective valuation, the less practice we'll get at it. slop slop slop all the way down 😑

1 reply

0 recast

1 reaction

vaughn tan pfp

also think that "smartness" is not the right criterion for thinking about AI systems the right criterion is whether they can make decisions (like humans can) about inherently subjective value

0 reply

0 recast

0 reaction

vaughn tan pfp

i'm not optimistic — i'm actually pretty _gloom_ but the only silvery lining will be that we can stave off everything turning into slop if humans are hypervigilant not just about AI outputs but in a more meta sense, about not giving machines things to do that they can't do reliably. on which topic: 1. https://uncertaintymindset.substack.com/p/ai-meaningmaking 2. https://uncertaintymindset.substack.com/p/ai-mirage

1 reply

0 recast

0 reaction

vaughn tan pfp

def! but it'll be easier to avoid silly (or dangerous) hallucinations by being clear about what to ask AI systems to do and what to carefully ensure that only humans get to do. my answer is meaningmaking should be for humans only (for now)! https://uncertaintymindset.substack.com/p/where-ai-wins feel like i'm a lone voice in the wilderness on this topic though

1 reply

0 recast

1 reaction

vaughn tan pfp

<3 this framing: being willing to invest the effort in making the meaning of a thing right is an instantiation of both care and awareness the problem may also be that effort itself is inadequate when the person doesn't know enough to detect misinterpretation or hallucination (especially when very subtle — one might even said insidious)

1 reply

0 recast

2 reactions

vaughn tan pfp

Innovation stalls when we treat uncertainty like risk. In my latest article, I share a 4-day program helping public sector teams embrace uncertainty as a strategic resource—not a threat to be planned away. Ready to transform policy work in your govt? https://vaughntan.org/transformpolicy

0 reply

0 recast

1 reaction

vaughn tan pfp

i instructed it to not make up articles that don't exist or urls that don't exist :-|

0 reply

0 recast

1 reaction

vaughn tan pfp

just spent 5h undoing a conceptual mess: an AI-generated essay with subtly incorrect references (some don't exist at all) that had been subtly mis-"read" by the AI system. this is further evidence that when using AI systems to do work, humans need to be clear about what the meaning-making part of the work is, and be vigilant about keeping ownership of that part: https://uncertaintymindset.substack.com/p/ai-meaningmaking

1 reply

1 recast

4 reactions

vaughn tan pfp

don't people here like giant fiberglass roast ducks?

0 reply

0 recast

2 reactions

vaughn tan pfp

not so /europoooooor now huh

1 reply

0 recast

3 reactions

vaughn tan pfp

itap of a big fibreglass roast duck

0 reply

0 recast

0 reaction

vaughn tan pfp

not at the scale of a whole bldg. but i made a bunch of built-in shelving last year for alcoves which had no level/plumb flat surfaces, and where the walls were not robust enough for anchoring. so built foundations and made freestanding support structures. one of them was less than half a degree off level — which caused no end of grief. the other one was perfectly level and a joy to build on. also, i learned that a tiny amount off level translates to a lot when the structure is 3.2m high 😑

0 reply

0 recast

0 reaction

vaughn tan pfp

1 reply

0 recast

1 reaction

vaughn tan pfp

2 replies

0 recast

5 reactions

vaughn tan pfp

ask me how i know ⚰️

0 reply

0 recast

1 reaction

vaughn tan pfp

0 reply

0 recast

2 reactions