The more I build and decide, the clearer it gets that tolerating ambiguity isn’t a personality type. It’s a skill. And probably the most leveraged one you can train. - I need more information first - is sometimes wisdom. Usually it’s fear in a trench coat.
used to love the em dash, now it’s the clearest AI tell. wild that LinkedIn is still full of it. we need a Geneva Convention for em dashes. until then: the strategic comma, the occasional semicolon, and the lost art of just ending a sentence.
Breaks from social media have made me weirdly sensitive to its dementor effect. Even short use now and I feel something being actively drained. The mechanism was always there, I just couldn’t perceive it through constant exposure.
@dlevine815 make prompts and the subsequent responses collapsible. Tapping a prompt could shrink it to a single line, freeing vertical space and making long conversations easier to navigate
Study has limitations (small sample, potentially leading prompts, might just be mimicking therapy text from training). But interesting to think about what that means for AI evaluation. Also Claude refused while others didn't, which is interesting.
https://t.co/5j0iJFrbGR…
5/
Most people don't know what they want until you ask them the right way.
Turns out LLMs are the same. They don't "have" a personality until you prompt one into existence. The self isn't underneath the responses - the self *is* the pattern of responses. 1/
Which means: if you're evaluating AI through standardized tests, you're measuring how well they've learned to take tests.
Not what they'd do when they don't know they're being evaluated. 4/
Works for Fortune 500s and startups alike. The best insights come from open-ended questions that reveal what customers actually do, not what we think they should do.
You learn more about PMF in 10 user interviews than in 1,000 rows of retention data.
At Trendyol data showed us who to talk to (users dropping at different stages of purchase funnel). Open-ended questions showed us why (“walk me through what happened” not “do you want X feature”)
Learned this lesson early at McKinsey too: Complex model said spread bank branches for coverage. Customer interviews revealed people wanted them clustered to compare rates in person.
Sophisticated analytics < asking “how did you choose your bank?”
What if therapists just went undercover as our loved ones friends? They get help without ever admitting they went to therapy. Especially useful for boomers who don’t believe in mental health. Call it stealth health.
Waiters who memorize orders instead of writing them down are optimizing for looking competent rather than being competent. The pen is mightier than the working memory.
Treating life as an experiment makes everything easier. Want to write daily? Frame it as a study and see what happens. The experiment mindset changed my relationship w failure. Instead of “I’m bad at this” it becomes an interesting data point in the bigger picture.
We went from avoiding commercials to voluntarily following people whose entire job is to be a commercial. Influencers are freelance advertisers and human billboards. And somehow we’re surprised that social media feels exhausting.