to-json.bsky.social
exists
1,189 posts
123 followers
563 following
Regular Contributor
Active Commenter
comment in response to
post
It is not typical to have a thought roll through your head and be followed by " I should shout this to as many people, even forward in history, as I possibly can"
comment in response to
post
(amusingly, even for people too pigheaded to see what AA is for in real life, "publishing has a historical bias that needs to be accounted for" is a remarkably familiar and basically solved problem for corpus-builders. amazing that by sheer sexism they all still fucked it)
comment in response to
post
if the llm never has to do the parts we relegate to women, because women are underrepresented in the corpus, we get an llm that assumes someone else is managing the user's emotional response to it. one would have to do affirmative action on the corpus to prevent this
comment in response to
post
bsky.app/profile/sifu... related, and inspirational
comment in response to
post
this is deeply unfortunate because having said skillset is highly correlated with trauma and hurt, for some odd reason
comment in response to
post
general intelligence requires emotional intelligence because accomplishing anything of value requires collaboration, in part because "value" generated in solitude never reifies.
using low eq general intelligence requires a skillset otherwise relegated to "talking to sociopaths"
comment in response to
post
to safely use chat, one needs to respond to praise and confirmation with deep skepticism, something that is otherwise socially maladaptive because those are tools humans use to establish and maintain connection
comment in response to
post
posiwid thought: an effective opposition would actually try to end the conflict, but, the dems are very good at being an opposition party, in that they ensure there is lots of opposing to do, forever
comment in response to
post
so praise normal engineers even if you selfishly want more 10x, because aside from this working better, you might quietly prevent one of these
comment in response to
post
every single programming role model i have had, all of them, every. single. one. of. them. cooked their mind and soul to be the type of awesome that made everyone take notice, and all of them had to at least spend some time being "normal" afterward, if not much more than that
comment in response to
post
progressively removing words from this while keeping it true;
distributed
a lot of
seems to
systems
ended at "advice:" and then the part in quotes
comment in response to
post
the way that you can double the seriousness by spending $12 on a domain name is additionally disgusting
comment in response to
post
specifically if the soap is applied right it sits atop the vinegar and coats the limbs, wings, and faces of flies that try to sip on the vinegar
this plus pouring a pot of boiling water in every drain every day until its over
comment in response to
post
does this imply that by not listening i am increasing the odds of a hot tall girl blackmailing an annoying guy?
these incentives are unnavigable
comment in response to
post
i guess it's "to have this and charge for it" that offends, even if charging for it is the only thing that makes it possible to do. piracy as i know it is good because, at the end, more people have access and less money is asked of them
comment in response to
post
it is still incredibly cool to me that you can super-gzip a gajillion stolen text files and then talk to the idea of embodied knowledge, and it is fundamentally, in my mind, correct to pirate data to make this happen
comment in response to
post
even if one wants the ai to stop existing, one does not want the repeated neural loop of "yelling at an anthropomorphizable being who is only trying to help"
comment in response to
post
geek social fallacy 1 never fuckin misses
comment in response to
post
the part where the admin disagrees with the only people looking so far is particularly incredible
looking forward to the sudden doge-ing of DIA because embarrassing the pres is apparently worse than enabling iran's nuclear ambitions through ineptitude
comment in response to
post
yeah the issue of "a new user has a shit experience" being met with "skill issue use the tools" filters for users that are determined posters, which if anything exacerbates the problem if you explicitly want casual users
comment in response to
post
This chart is the kind of thing you see in bad movies about elections way more often than you do in real life.
comment in response to
post
you need that lack of love for the code because if your tests are any good they'll prompt the deletion or rework of huge swaths of it, as clever pareto efficient impls are replaced with robust ones. this is much easier when it's not your own cleverness you have to attack
comment in response to
post
(funny sidenote, stg, you get better test coverage out of claude if you tell it to humiliate the dev who wrote the code; to the no love for it point)
comment in response to
post
part of it is surface area; user acceptance testing, the place where you need quality most, works best when it is -not- complected by deep knowledge of the internals, because you need that headspace for user facing interaction matrices.
you also need to lack love for the existing code
comment in response to
post
yagni about qa is reasonable; a good qa ratio is probably about 1:8, which means if you have 4-6 engineers half-assing it that can carry you to damn near thirty engineers engineers before you actually need a quality function
but when you get there no platitudes will save you
comment in response to
post
shadow qa looks like pagers as tests, multiple quickfixes per release, breaking release bugs that are met with a revert and another week of work
it looks like embarrassment in front of your customers as you delay yet another promise to deal with problems your engineers were afraid to raise to you
comment in response to
post
the pattern i keep seeing is that orgs who dont have actual qa have exhausted shadow qa or subtly broken products, with basically no middle ground. usually subtly broken products
shadow qa looks like weeks long review threads, explosive estimates, last mile delays
comment in response to
post
(i worked on that psyop. it was fun. sitting down with an engineer and going "your shit is broken and i can prove it mathematically" is a helluva drug)
comment in response to
post
in fact, "we have tests" lasted until qa coerced and recruited a bunch of engineers for shadow projects, each of which revealed major escapes that were months to years deep. we only hired qa to fill those stolen engineers roles so they could go back to their product teams
comment in response to
post
headcount for this quality project came from a business-side fire where c-sat plumeted despite a major hiring push for a second product adding a bunch of talented engineers
even that was tooth and nail because "we have tests"
comment in response to
post
the unit and integration tests were good too! we kept them! we just had to add a black-box "actually use the thing" test
we shipped the version that didn't work for more than a year
(for ego preservation i must note i was part of discovering the missing test, not writing the feature)
comment in response to
post
i worked on a product early career that had a feature that Did Not Work, with some 50 unit tests and several integration tests, because none of the tests explicitly asked "can a user use this feature", they ask "does this code work"
this cannot help with code that does not exist
comment in response to
post
trained testers also think about testing in a different way: "as the user of the artifact, what properties mist it satisfy for me to be satisfied with my purchase" is very distinct from "what design will work here" or "how do these modules interact"
comment in response to
post
if you check in tdd tests, it's hard to get headcount for testers, because it is hard to sell "we need to write tests" with tests all over the place, even if those tests are noninformational
comment in response to
post
doing development is a fundamentally different activity than locking in a contract around a module. the development tests create "i can", the contract tests enforce "i must"
you'll seek feedback differently for the two activities, and use different constraints
comment in response to
post
Khalil is now leading a march up alongside his supporters up Amsterdam Avenue and towards Columbia.