winston the third

Cercle: Worakls Orchestra at Château La Coste, France (Live) by Worakls on Apple Music Album · 2019 · 14 Songs

stumbled on to this cercle session today and holy shit it is so fun and good and i will be repeating it relentlessly for the foreseeable future music.apple.com/us/album/cer...

at the scale of data required to train, i think it's a very questionable assertion that the training captures "truth." it captures... something. but we can't validate that scale of data, and truth has many facets.

going to go insane because every discussion of "this is how the tech works" is met with a response of "well in the future it might be different"

if you don't understand how it works _now_ how the hell you gonna extrapolate a future where it works differently? computers are not magiccccccc

the point i am trying to make is that the LLMs are doing their job accurately, which is to produce a statistically probably response to the given prompt. we cannot frame "mistakes" as "hallucinations" because that suggests a fixable issue. these tools work differently than humans.

no i stand by it, the tool did its job accurately, it does not have the concept of "accuracy" as we care about it and that's what the problem is

Power is power.

the "hallucination" language conveniently implies that things will change in the future, but they WON'T. the same process that produced what you understand as accurate also produced what you understand as inaccurate. only humans can layer on the accurate/inaccurate analysis.

your LLMs do not have any knowledge of "reality" or "truth" or even fucking "accuracy." they have outputs based on training. the so-called "hallucinations" AND the "correct" outputs are the same thing. same goddamn thing. it didn't mess up. it didn't hallucinate.

forever fucking annoyed at calling LLM outputs that aren't what we wanted "hallucinations" because THE TOOL IS WORKING AS DESIGNED. that's not a mistake! it's not a whoopsie! it's what the statistical probabilities inside the model based on the training data ACCURATELY PRODUCED.

the youtube algo has been feeding me hbcu marching band videos and now I’ve watched this approximately 50 times in a row

i will pay $1000 out of my own pocket to attend this conference

On Hold (Jamie xx Remix) [Mixed] by The xx on Apple Music Song · 2017 · Duration 5:07

this song from Jamie xx's 2014 boiler room set, and the version that Romy remixed in her boiler room set a decade later are load-bearing components of my mental health these days. perfectly normal. music.apple.com/us/album/on-...

Fluke, by Brian Klaas is the book that i've been seeking, pulling together a bunch of threads about how the world works and how we make sense of it. I love it for that!

But it's pop science so it has wayyyyy too many anecdotes and dumb metaphors and I hate it for that!

is that in any way similar to software controlling the car while you're driving it? or is your genius argument that because some surveillance exists, there's no argument against other surveillance?

every system is dynamic. nothing stays the same! data that says "this is not that bad right now" NEVER means "thus it is not bad at all"

if your argument is "people are blowing this out of the water, look at the data" maybe step back and consider if the data could fucking change

i have done multiple cross country drives this year and while agree that in urban areas, 65 is fine, it truly makes a difference on long distances

there is a mountain of literature and evidence for how to build roads and infrastructure that makes it inconvenient if not impossible to go fast. if we can theorize putting a networked governor in every car, we can imagine applying knowledge to road design.

there is also an intrinsic element of surveillance to it, what with a car needing to know exactly where you are to determine speed limit. it's a technological solution to a problem for which many better solutions already exist.

the problem with an empire that is making decisions based on feelings, whims, and ego is that there is no "rational" response, you cannot argue back. we are off that map.

we are, however, very much on the map where pettiness and insults punch way harder than they should. anyways happy monday.

anyways @governorwalz.mn.gov i live like 10 blocks from you hmu i bring a certain extremely technically competent vibe to my unwillingness to consider that AI and its proponents are actually building a viable future.

realized that a job i would enjoy would be the tech bro diplomat for a theoretical tim walz presidential campaign. as in, i would be the one extracting donations from them while also cataloguing all their silly demands based on their magical ideas of impossible futures.

love living in a city with so much good football! go loons!

do you ever think about how the hopeful part of Mad Max: Fury Road is all the parts that happen after it "ends" and we don't see?

had this thought today and now i can't shake it

best fucking movie ever but the actual tough stuff is never shown

i stood beside the sign as a joke and said "take a photo" and my partner laughed and kept walking but our friend who is visiting actually did the thing so part of my grin is being dumbfounded that someone is humoring me

a sign says "Cute Stuff This Way" with a big painted arrow pointing to the right of the image, where a tall goofy dude stands with a shit eating grin on their face

too bad i'm taken for good because this would be a great dating profile pic for the meat market

god i know that the problems we are facing are big and complex and structural but addressing them can be so goddamn joyful if we let it! this was so fun to watch

my favorite part is that it only reloads the things you aren't ever focused on

hey siri remind me to make blueberry pancakes tomorrow

alas, a migraine took down a team member so we are sitting out the event. but here's some photos of what we made! sorry @blackhartstp.com

Cheater McGavin