Eli

[Eli @document_body]

@rauchg I hope he sees this king

[Eli @document_body]

@frederic_ooo @steveruizok Pretext uses canvas https://t.co/4Xt1PWm8g4

[Eli @document_body]

@matthewcp If there is a 10% chance of 0% of the population surviving, what’s the % chance of 90% surviving, and are you okay with that too?

[Eli @document_body]

@lik_ai_ @jaffathecake IMO one of the pitfalls of responsive design is placing too much emphasis on the ‘resize’ case. I’m definitely guilty of it, because it’s so much fun, but real users don’t resize the window very often

[Eli @document_body]

@JLarky @hd_nvim Maybe he knows that C++ is more than 40years old

[Steve Ruiz @steveruizok]

I'm kind of iffy on "skills discourse" but I think I came up with one so good that its worth sharing. The basic ask is to "describe the UX of my application completely". In doing so, agents will discover inconsistencies, bugs, and design errors that have never come up before

[Eli @document_body]

@tszzl Because Reddit did it first (I think?)

[Eli @document_body]

@paulg

[Eli @document_body]

@trq212 Everyone is projecting human intuition for how they would detect a watermark. Don’t LLMs have an implicit fingerprint by nature of having fixed weights? A model could say ‘prefer “overcast” when causally discussing historical Autumn weather in Edinburgh’

[Eli @document_body]

@aboodman Somehow the analogy is helpful, but it’s funny that it reads as “imagine if serverless functions weren’t serverless or functions”

[Eli @document_body]

@invict37 @Spectromachina @M1Astra Couldn’t you just use another non-Anthropic model to rephrase the text? Pangram might still say ‘looks like AI’, but the Anthropic fingerprint would be gone

[Eli @document_body]

@steveruizok @brendanrc2 In my org we do ‘something vaguely resembling semver’, and use the version bump to decide whether to prompt for an update

[Eli @document_body]

@GergelyOrosz @tanishqk Isn’t code review more about shared responsibility and ownership of a potentially user-facing change to the product? The code quality is usually not the problem, rather that they’ve mis-understood a feature request or missed some important context

[Eli @document_body]

@trq212 I never understood CLAUDE.md. Aren’t the models trained on tens of millions of README.md docs? Looking forward to the day the advice becomes ‘keep your readme up to date’

[Eli @document_body]

@whashywash @__alpoge__ @BrunsJulian1541

[Eli @document_body]

@aaron_lou Even with no explicit web search, is there really no kind of higher-level / semantic caching going on? Like a global ‘collective concept cache’ that could be interfering?

[Eli @document_body]

@__alpoge__ (guy who had to look up what a conjecture is (me)):

holy fuck

[Eli @document_body]

@icyphox It also doesn’t mean ‘local only’

[Steve Ruiz @steveruizok]

I feel the same way about Linear, where a product wonderfully designed around a human activity feels sort of less good or pointless as it pivots towards AI workflows, where the biggest competition comes from more convenient throwaways like md files or one page sites

[Eli @document_body]

@rough__sea Should repositories have README’s?