@eaton@phire.place
@eaton@phire.place avatar

eaton

@eaton@phire.place

Social justice advocate, content strategy ne'er-do-well, co-founder of Autogram, co-host of Christian Rightcast. Anti-fascist, partner to https://zirk.us/@sparrowpost

This profile is from a federated server and may be incomplete. Browse more on the original instance.

timnitGebru, to random
@timnitGebru@dair-community.social avatar

"Leike had said that OpenAI's "safety culture & processes have taken a backseat to shiny products" in recent years, and that his team struggled to obtain the resources to get their safety work done."

Lol. But there never was a "safety culture"!

The founding of OpenAI was premised on "existential risk" of AI to "humanity," the new grift white men created to siphon even more resources for themselves under the guise of saving all of us.

https://mashable.com/article/openai-sam-altman-greg-brockman-respond-to-jan-leike-resignation

eaton,
@eaton@phire.place avatar

@timnitGebru Whenever someone’s pitch for safety is “X is so dangerous, WE have to be the ones that make it first!” I know there’s bullshit comin’

eaton, to random
@eaton@phire.place avatar

This week's discursive thread about magic, AI, and the tricks that language plays on us has been worked into a new piece over at The Interconnected. I suspect it won't be the last…

https://theinterconnected.net/eaton/incantation/

glyph, to random
@glyph@mastodon.social avatar

If you are not interested in the goings-on at , my feed is about to get truly annoying. I usually try to be mindful about posting volume but for the next week or so I will definitely not be doing that. I will do my level best to use the hashtag consistently, and to boost stuff that does the same, so if you don’t want to be flooded please go ahead and mute the hashtag if you haven’t already.

eaton,
@eaton@phire.place avatar

@glyph buddy, this is what i’m HERE for

eaton, to random
@eaton@phire.place avatar

With apologies to @Wolven, some thoughts on the magicklike conception of LLM Prompting.

Telling computers what to do has always been a job for specialists — almost by definition it is an act of translation, of adaptation, of understanding two worlds and bridging the gap between them.

Some academic computer science folks may spend inordinate amounts of time on the computer-side of that divide, figuring out better algorithms and more efficient mechanisms for what happens behind the curtain…

eaton,
@eaton@phire.place avatar

…But the vast majority of "computer programmers" and "developers" and even tinkerers spend much more time straddling the divide, trying to understand human needs and processes and tasks and mental models and translate those into the kinds of inputs and decisions and outputs that a computer will need to wend its way through to accomplish specific tasks.

eaton,
@eaton@phire.place avatar

Even when the system a programmer interacts with is so complex, poorly documented, or hermetically sealed that it feels non-deterministic, there is a baseline understand that they are dealing with the artifacts of other humans' past decisions.

Jokes about certain "incantations" being necessary to make the system work are jokes about knowledge-of-mechanisms lost to time. Not a description of real Words Of Power. Not a description of true communication between human and machine.

eaton,
@eaton@phire.place avatar

LLMs twist that normal understanding around, in ways I don't believe our culture has mental models or vocabularies for.

The questions we ask them, the commands we give them, the outputs we receive from them, are not translated to and from a precise, deterministic language. What we get from them is fuzzy and organic-feeling and startling.

(Under the hood, of course, it is very much that precise and deterministic stuff — still code, still math, that is — but our interface to it is not.)

eaton,
@eaton@phire.place avatar

Coaxing and directing and nudging LLMs in one direction or another feels — at least, in the conversational interface most users are immediately presented with — like a matter of voice and tone, an act of convincing and cajoling rather than specification and testing.

Research papers are published encouraging users to add "emotional language" to their queries. To "explain how important the question is." To lie, or to promise a reward for a job well done, to convince the machine.

eaton,
@eaton@phire.place avatar

Behind the scene, as always, this still collapses into complex but ultimately straightforward math.

LLMs have been trained to mimic the shape and structure of the conversations we humans have with each other, the messages we use to coax and cajole and convince other humans.

“Please answer clearly, my job is on the line," then, is not a desperate plea to another thinking, feeling being. It is a search filter once removed, narrowing a range of probable responses by mimicking past emergencies.

eaton,
@eaton@phire.place avatar

"You are an expert salesperson, confident in the quality of your product and skilled at listening to customers to understand their needs before recommending a solution.”

A prompt is not a persona, a character we have "convinced" the machine to slip into. It is a vector — a particular line speared through the sea of syllables and sentences, successful if it pierces the fish but otherwise meaningless.

baldur, to random
@baldur@toot.cafe avatar

One of the things that the Stack Overflow brouhaha demonstrates is that it doesn’t matter if a service was founded by people trusted by the community (Atwood and Spolsky) and was broadly community-led. If it’s a VC-funded startup, they will sell out their users at some point.

eaton,
@eaton@phire.place avatar

@baldur back in the day, that origin story was a real inspiration. Now it feels like a “live to see yourself become the villain” punchline

mcc, to random
@mcc@mastodon.social avatar

Dead internet theory but for IRL

eaton,
@eaton@phire.place avatar

@darius Is a copy of that talk up anywhere?I’d love to see/read/etc.

eaton,
@eaton@phire.place avatar

@darius @c0dec0dec0de @cstross man, I need to dig up more of the notes from the stuff I wrote back in the late 90s/early aughts. It was never fully fleshed out, but the idea several collaborators and I anchored on at the time was an AI future in which “true AI” never happened, but probabalistic machines fed on huge quantities of human-authored material were convincing enough that civilization just started using them to deskill creative, assistive, and analysis work.

danhon, to random
@danhon@dan.mastohon.com avatar

re the rabbit r1's “ability to point the camera at a chunk of text and tell it to “explain this to me like I’m a 3rd grader” is fairly incredible”

-- feels like this is a sort of accessibility feature that Apple could do really well at, given their history and approach to accessibility. They've already been making little steps of progress here

https://houseofkyle.com/2024/05/01/rabbit-rabbit/

eaton,
@eaton@phire.place avatar

@danhon Yeah, to me at least that feels like interestingly equivalent levels of utility magic to the point-your-camera-at-a-foreign-sign-and-see-it-in-your-native-language apps.

eaton, to random
@eaton@phire.place avatar

A few years back, I wrote that the right wing in America believed that human rights were a reward for ideological correctness, rather than a baseline all persons were entitled to. The designation of “stateless” enemies who could be targeted for assassination, and the insistence that noncitizens have no claim to humane treatment or protection under the law, would inevitably slide into arguments that “insufficiently loyal” citizens have no rights.

Keep an eye on this one.

https://www.rightwingwatch.org/post/jack-hibbs-says-campus-protesters-have-no-constitutional-rights/

tanepiper, to random
@tanepiper@tane.codes avatar

Why do Americans insist on completely ignoring the metric system?

eaton,
@eaton@phire.place avatar

@tanepiper I mean, I can’t really complain, I maintain a spreadsheet just to quicklly calculate novelty units (quibis as a unit of money/time, yoctobears as a unit of weight, etc), so

eaton, to random
@eaton@phire.place avatar

I dunno, man, I'm kind of getting the impression you CAN wait for a response from me

zachleat, (edited ) to random
@zachleat@zachleat.com avatar

Reading this blog post from @dries (creator of Drupal) about the post-pivot definition of Jamstack.

https://dri.es/the-new-old-jamstack-and-mach-journey-towards-traditional-cms-concepts

Dries rightly notes that Jamstack 2.0 shares more in common with monolithic/traditional stacks.

For me—I read this as a huge indictment of the broadened definition—it’s not a healthy evolution of the term if we must sacrifice technical clarity and simplicity to do so.

#TheFutureOfJamstack

eaton,
@eaton@phire.place avatar

@tanepiper @zachleat @deadparrot @dries “Headless CMSs are spicy domain optimized databases, change my mind” was one of my favorite LI shitposts

eaton,
@eaton@phire.place avatar

@deadparrot @tanepiper @zachleat @dries “tags” and “the ability to put content in a hierarchy” are some examples. The issue isn’t that they are impossible with most headless systems, but that teams are left recreating CMS features from first principles with data primitives — along with the editorial interfaces to manage them.

Sometimes that’s what a team wants, but often it’s an unexpected shock.

eaton, to random
@eaton@phire.place avatar

Finally finished Benjamin Lorr's "The Secret Life of Groceries: The Dark Miracle of the American Supermarket” and it ... is a good read? That lives up to its ominous title?

It's broken into six parts, which basically go from “Wow, Trader Joe's has a fascinating origin story" to “…I …did not realize so many of the Myanmar people had suffered decades of slavery and inhuman abuse to keep shrimp prices low”

An excellent book about grocery store, but also about the grim mechanisms of capitalism.

eaton,
@eaton@phire.place avatar

@lewiscowles1986 they’re about 50% cheaper than they were in the early 90s, ajusted for inflation. On the long term, the big shift was the global emergence of industrial-scale shrimp farming in the 70s. Before that they were a luxury product that cost more per pound than a good steak.

Production at that scale, at the prices we’re now used to, is tremendously labor intensive but also relies on slave labor.

Calling that “guilt porn” seems pretty weird.

eaton,
@eaton@phire.place avatar

@lewiscowles1986 It’s a bit like saying electronics prices are too high today; they are expensive in comparison to many other things we consider essentials but historically very inexpensive. And even then, often rely on labor practices and environmental impacts we believe are wrong but are unable to alter via individual action.

“So? We know!” Well, yeah.

eaton,
@eaton@phire.place avatar

@lewiscowles1986 A couple of thoughts.

First, have you looked at the actual price of shrimp, or are you just applying general CPI correction to the current advertised price?

https://fred.stlouisfed.org/series/WPU02230501

eaton,
@eaton@phire.place avatar

@lewiscowles1986 Second, the shift in both price and production levels I referenced is missed if you cut off at 1977. Industrial shrimp farming emerged in the mid 70s, and that change is what transformed it to a commodity protein.

Your earlier comment (that shrimp should be cheaper since they're ‘harvested from the ocean’) is off in two respects: that's not how the shrimp industry has worked for decades, and even if it did, ocean-caught is more rather than less expensive, production-wise.

eaton,
@eaton@phire.place avatar

@lewiscowles1986 Third, complaining about the high price of shrimp when it's historically low, both in real terms and when adjusted for inflation, is a perfect demonstration of why the shrimp supply chain, like chocolate and coffee and cotton and a bunch of other commodities, keeps whoopsing its way into horrifying human and ecological costs.

The societal problem of those costs across a bunch of industries, and the difficulty of altering the systems that produce them, are not “guilt porn.”

  • All
  • Subscribed
  • Moderated
  • Favorites
  • •
  • JUstTest
  • kavyap
  • DreamBathrooms
  • thenastyranch
  • magazineikmin
  • tacticalgear
  • cubers
  • Youngstown
  • mdbf
  • slotface
  • rosin
  • osvaldo12
  • ngwrru68w68
  • GTA5RPClips
  • provamag3
  • InstantRegret
  • everett
  • Durango
  • cisconetworking
  • khanakhh
  • ethstaker
  • tester
  • anitta
  • Leos
  • normalnudes
  • modclub
  • megavids
  • lostlight
  • All magazines
    •