An absolute must-read by Erik Salvaggio on how image generators come to be and the abuse hidden within them.
”Stanford Internet Observatory’s David Thiel — building on crucial prior work by researchers including Dr. Abeba Birhane — recently confirmed more than 1,000 URLS containing verified Child Sexual Abuse Material (CSAM) is buried within LAION-5B, the training dataset for Stable Diffusion 1.5, an AI image tool that transformed photography and illustration in 2023. Stable Diffusion is an open source model, and it is a foundational component for thousands of the image generating tools found across apps and websites.”
”An additional point of serious concern is the likelihood that images of children who experienced traumatic abuse are influencing the appearance of children in the resulting model’s synthetic images, even when those generated images are not remotely sexual.”
« LAION’s data is gathered from the Web without supervision: there is no “human in the loop.” Some companies rely on underpaid labor to “clean” this dataset for use in image generation. Previous reporting has highlighted that these workers are frequently exposed to traumatic content, including images of violence and sexual abuse. This has been known for years. »
https://www.techpolicy.press/laion5b-stable-diffusion-and-the-original-sin-of-generative-ai/
Per Axbom
Grieving how lots of people with good ideas are opting for gated platforms for their content. It's hard to navigate all this, I get it, and I'm not blaming anyone but the platforms themselves.
Continuously hitting a wall on Medium is what got me this time. I just wanted to post a supportive comment...
General-purpose search engines can lead you down all sorts of AI-generated delusions. While not a fail-safe, mastering DuckDuckGo's ability to bypass big search engines in favor of narrow ones can be helpful.
https://axbom.com/duckduckgo-bangs/
@iamdavidobrien That I will do!! Lovely tips.
Here's I am holding a book at Waterstones on Princes Street :) It's funny, because it was actuallly our most recent trip to Edinburgh I was thinking of when I was considering where we probably bought the most books.
Happy new year David!
Tänkte skriva något om skolan och digitalisering men jag har redan skrivit det. För snart sex år sedan.
Det är ibland lätt att glömma att vi ältar samma frågor om och om igen. Och att det vi skrivit och kommit fram till för många år sedan fortfarande har värde. Eller aldrig ens omhändertogs.
Jag la mycket möda bakom min text om mobilförbud. Jag har fått mycket positiv återkoppling genom åren (och de senaste dagarna).
Kanske har den ett värde i att signalera hur man på tydligare och tryggare sätt kan angripa frågor om digitalisering ur ett samhällsperspektiv.
Jag inbillar mig inte att du har tid att läsa den idag, på nyårsafton. Den beräknas ta nästan 15 minuter att läsa. Men kanske kan du hitta en stund under nyårsdagen att fundera över hur det kan bli om vuxenvärlden tar lite mer ansvar för sitt (vårt) agerande.
Planerar, utvärderar och resonerar. Inte bara säger vad vi ska göra utan också tydligt talar om hur vi följer upp följderna av de experiment som barn och medborgare överlag utsätts för.
I stället för att ideligen ignorera den djuplodande utvärderingen och hasta vidare till nya snabba lösningar som likt en hydra ständigt alstrar nya dilemman för någon annan att hantera.
Och om du ansvarar för en publikation där du tror att denna text eller något liknande skulle göra nytta... hör av dig.
Gott nytt omtänksamt år! ❤️
How do you say "We have no idea how our product works", without saying those exact words.
OpenAI: "Training chat models is not a clean industrial process. different training runs even using the same datasets can produce models that are noticeably different in personality, writing style, refusal behavior, evaluation performance, and even political bias,"
They can say 'refusal behavior' and get away with it.
https://businessinsider.mx/chatgpt-accused-of-getting-lazier-2023-12/
What EM failed to acknowledge when he told advertisers to go f*ck themselves is that they already said it first.
They just said it in a way that actually makes a difference.
@ajswritesthings I've actually just discovered this. Love a gadget I can actually tinker with!!
@iamdavidobrien You should see me travelling. If I travel with my wife it's rare we've returned home with less than 20 books.
But some books are just a real hassle when it comes to getting hold of a physical copy.
And our stacks of books are somewhat of a challenge after we moved to a smaller place... 😅
@ruprecht I know I know...
Oh how I hate DRM. Remind me to never buy a book via Apple Books again.
I am now downloading some titles I've previously bought. Some of them I've bought twice. And part of me is still feeling like I'm doing "something wrong". It's tragic how we ended up in this reality.
(The backstory for all this is that I gifted myself a Kobo e-reader from my company.)
Om du vet att ett möte spelas in…
- Påverkar det vad du säger i mötet?
– Ställer du krav på hanteringen av inspelningen?
– Tror du att det påverkar hur andra deltar i mötet?
Om du vet att ett digitalt möte transkriberas automatiskt (för minnesanteckningar och ”automatisk summering”)…
– Vågar du säga precis vad du tänker?
– Oroar du dig till exempel för att något du säger om en chef/kollega sedan kan bli sökbart i organisationens system?
– Tror du att det påverkar hur andra deltar i mötet?
Hur bekväm skulle du känna dig med att tacka nej till en inspelning/transkribering om din organisation har det som praxis?
Hur trygg känner du dig med att inspelning/transkribering av alla dina möten inte blir åtkomligt för andra via interna sökmotorer eller ”AI-kompanjoner”?
Har du någon gång sagt något i ett möte som kanske inte bör lämna det mötet? Som kanske inte ska höras av precis alla?
Några frågor att diskutera när jobbet drar igång igen.
2024 blir sannerligen ett år med många frågetecken. Kanske startar en tankesmedja för att reda i dessa.
Till dess, gott nytt år och ta hand om dig.
#DigitalEtik #Dataskuggan
@thatandromeda With you all the way. It was me toying with the idea that many consultants are very focused on saving minutes here and there as a way of selling their products and services.
"You can save millions in time with this new CMS that allows you to publish 30 seconds faster!"
If those simplistic arguments/calculations held up both ways then much of what the same consultants are advocating for when it comes to adopting AI would be "lost time", and hence lost money.
I wonder if anyone has calculated any estimates of "worktime lost" to playing around with language models during 2023.
And compared it to "worktime gained" through improved output.
On broader populations, not individuals, mind you.
Thoughts on organisations setting goals.
If an organisation sets these goals, what other goals might be necessary to complement with?
1) We create opportunities for customers/stakeholders to improve their wellbeing.
2) We support autonomous, informed choices. (internally and externally)
3) We use resources in a sensible, sustainable and compassionate manner.
4) We proactively listen to, and do a good
job of managing, feedback (internally and externally)
5) We invest in the long-term wellbeing of our workforce.
He had me at "egg".
"[AGI] interprets the Turing Test as an engineering prediction, arguing that the machine “learning” algorithms of today will naturally evolve as they increase in power to think subjectively like humans, including emotion, social skills, consciousness and so on. The claims that increasing computer power will eventually result in fundamental change are hard to justify on technical grounds, and some say this is like arguing that if we make aeroplanes fly fast enough, eventually one will lay an egg."
From the book Moral Codes - Designing Alternatives to AI, by Alan Blackwell
https://moralcodes.pubpub.org/
@jesse
Sure. A decent amount of my writing has been on LinkedIn practices lately:
– using an abundance of emojis (can make it hard to read, and some have really long names read out by screen readers)
– using unicode fonts to simulate bold text (often not recognised by screen readers)
– putting links in comments (making it harder to reach and find for some people)