has the era of active sabotage of the autoplag inputs begun? let's hope so
this post was submitted on 21 Oct 2024
1 points (100.0% liked)
TechTakes
1758 readers
166 users here now
Big brain tech dude got yet another clueless take over at HackerNews etc? Here's the place to vent. Orange site, VC foolishness, all welcome.
This is not debate club. Unless it’s amusing debate.
For actually-good tech, you want our NotAwfulTech community
founded 2 years ago
MODERATORS
It would be funny if someone was literally beating up servers with a wooden shoe.
"percussive maintenance"
momty python style giant sabot descends on Microsoft data centre
Considering Glaze and Nightshade have been around for a while, and I talked about sabotaging scrapers back in July, arguably, it already has.
Hell, I ran across a much smaller scale case of this a couple days ago:
Not sure how effective it is, but if Elon's stealing your data for his autoplag no matter what, you might as well try to force-feed it as much poison as you can.
It's almost completely ineffective, sorry. It's certainly not as effective as exfiltrating weights via neighborly means.
On Glaze and Nightshade, my prior rant hasn't yet been invalidated and there's no upcoming mathematics which tilt the scales in favor of anti-training techniques. In general, scrapers for training sets are now augmented with alignment models, which test inputs to see how well the tags line up; your example might be rejected as insufficiently normal-cat-like.
I think that "force-feeding" is probably not the right metaphor. At scale, more effort goes into cleaning and tagging than into scraping; most of that "forced" input is destined to be discarded or retagged.
yeah this is the thing I’ve been thinking a lot about
fucking reCaptcha is literally mass-weaponising users for data filtration, and there is no good counter besides just not using reCaptcha (which is something one can’t easily pull off without things like regulatory action, massive reputational problems that make people gtfo, etc)
I have similar worries about cloudflare being such a massive chokepoint and using that position to enable “ai bot filter” services. feels extremely monopolistic, but ianal and I’m not entirely sure what the case grounds/structure on that would be (if any)
the only other viable strategy at the moment is fully breaking contact with any potential bad traffic systems, and that’s extremely fucking dire because that’s yet another nail in the coffin of the increasingly less open internet
The whole Cloudflare bot detection is so weird and eerie. I've had issues where I can't get past it presumably just because I'm using some in-application browser just to get a login cookie, but other times it just lets fucking curl through no questions asked.
it just lets fucking curl through no questions asked
Fucking what. I've heard of sites blocking curl and I've been able to get around it by copying user agent and sometimes cookies from the browser. Now I'm cursed with the knowledge that I could probably just scrape stuff from everywhere
I saw people say they would add 10% opaque layers of the musk with Epstein's accomplice (whos name i forgot for a second and too lazy to look her up) photo. Would be nice if there was a tool to do so automatically. (Not that i post on twitter anymore).
tbh that sounds like a pretty easy script to write! Too bad I am not near a computer rn
I got nerd sniped into trying to resize felons_musk_and_maxwell.webp to the same size as some base image before compositing it on top with a 10% dissolve in the same magick invocation but I need to sleep so I'm giving up for now.
They added sleeps to training jobs? Sounds like they deserve a raise for improving energy efficiency instead…
I thought they were gonna do that themselves by feeding on their own outputs littered all over the www. Maybe they can use some help.
that's also happening, but yeah it's going to have to be a team effort
we call it clogging, folks, we put a little clog in the machine
I'm sorry Mr. Musk, grok's a bit constipated today. Someone fed it too much cheese. Then it started hallucinating.
ooh I like that
In other news, there's been a statement on AI training that's racked up over 10k signatures, which is unsurprisingly lambasting the rampant stealing that went into creating the autoplag machines:
Now, I'm way too much of a fan of sidenotes, so I'll whip one out:
Beyond simple content theft being publicly lambasted, I suspect that even licensed use of artists' work for gen-AI will ignite some controversy - if Eagan Tilghman's run-in with controversy last year is any indication, any usage of gen-AI, regardless of context, will be met with hostility.
Update on LLM reviewer situation:
PM is down to let us pitch them our argument. Good news: PM seems like a cool person, is open minded, and is being pretty frank about the forces at work here. Bad news: taking action on this will open a whole can of worms, so any proof has to be ironclad. After conferring with our local grant wizards, the battle plan is to crank out a 15 minute pitch consisting of:
- a 2 min elevator pitch of our tech, highlighting what the reviews mangled
- intro to LLMs for people who know what glycosylation is
- intro to semiotics for the same
- show how transformer architectures transform symbols into symbols to produce text-shaped objects without actual intent, ideas, or context (and why "automated AI detection" is also bullshit).
- show a few examples of plausible-at-first-glance gen-ai slop (the nonexistant turkish fortress, mouse dck, etc)
- Highlight how our weird reviews (both good and bad) fit exactly into this bin (absolutely mis-interpreting a table, inventing a bacterial species we didn't use and talking shit about it, miscounting our team members, etc)
We'll be leaning on the Stochastic Parrot paper pretty hard, because it's a good entry into the field on the skeptical side and is just well constructed in general. I'm also on the hunt simplified diagram for how LLMs convert tokens to arrays to tokens from the original transformer literature. Unfortunately, so much of the literature is obscurantist on purpose, and I want to avoid falling into the "It can't be that stupid" trap. Any pointers in that direction are most welcome!
Wish us luck, heh!
If your Wikipedia page contains as many random arXiv preprints for references as the "prompt engineering" article, consult your physician.
Actual message I got while renewing my insurance plan last night. Thank you for adding a shitty chat bot which will give me false information about my life and death decisions, bravo.
This tool solely exists so that you can ask it questions and get assistance, but also we disavow any responsibility for the answers to the questions we just told you to ask it. Has this kind of clause been held up in court anywhere? Like, I'm sure it has but it seems like the same logic would be ridiculous in any other context. Like, consider the fraught legal history of the anarchist cookbook.
The Bookseller: Penguin Random House underscores copyright protection in AI rebuff
Penguin Random House (PRH) has amended its copyright wording across all imprints globally, confirming it will appear “in imprint pages across our markets”. The new wording states: “No part of this book may be used or reproduced in any manner for the purpose of training artificial intelligence technologies or systems”, and will be included in all new titles and any backlist titles that are reprinted.
Now that the content mafia has realized GenAI isn't gonna let them get rid of all the expensive and troublesome human talent. it's time to give Big AI a wedgie.
It's weird how rarely I see people point this, but in theory this kind of boilerplate should be technically meaningless. If copyright protections include the privilege to use the work for training a machine learning algorithm, you need explicit permission anyway. OTOH if it's fair use or otherwise not something copyright law is concerned with, the copyright holder's objection doesn't matter.
For the record, I think AI models are derivative works and thus they're not only infringing on typical "all rights reserved" works, but also things such as Free software whose license terms require attribution if used in derivative work, and especially share-alike copyleft licensed work.
I thinkt it's pretty well-lknown that Spotify got all its initial music from Oink. They moved fast, got dominant, and were able to present the record labels with a big audience prepared to pay for streaming music. The labels quickly ensured they'd get the lion's share of that revenue.
OpenAI and friends tried the same thing - scrape everything, build AGI, reap the rewards. Except it didn't work, and they're in a much worse position morally. Even if they can get a judgement that what they're doing is legal, it will cost them a lot in litigation fees, coupled with the public perception that these culture vampires are ripping off the poor honest author. Not a good place to be in.
Now that the content mafia has realized GenAI isn’t gonna let them get rid of all the expensive and troublesome human talent. it’s time to give Big AI a wedgie.
Considering the massive(ly inflated) valuations running around Big AI and the massive amounts of stolen work that powers the likes of CrAIyon, ChatGPT, DALL-E and others, I suspect the content mafia is likely gonna try and squeeze every last red cent they can out of the AI industry.
At some point, something is going to reveal that all the money in AI has gone into power costs for datacenters and NVidia chips and that the AI companies themselves aren't doing so hot. I hope it's the discovery process for some of the inevitable lawsuits.
it's pretty publicly known
the VCs are gonna take one heckuva bath
More AI generated shite from Ireland: Transport for Ireland decided, for some reason, to use AI for its Hallowe'en themed ads. This was roundly complained about online. Then someone decided, for genius reasons, to ring Liveline and complain about it.
For those who are unfamiliar, Liveline is a national phone-in show presented by JOEEEE DUFFY, who could start a fight with a brick wall. Every episode is about either a petty grievance or a real horror story. It's like a national whinge-in. I am going to listen to the episode (available here) and see if there are any highlights.
‘They wish this technology didn’t exist’: Perplexity responds to News Corp’s lawsuit
“There are around three dozen lawsuits by media companies against generative AI tools. The common theme betrayed by those complaints collectively is that they wish this technology didn’t exist,” said the Perplexity team in the blog. “They prefer to live in a world where publicly reported facts are owned by corporations, and no one can do anything with those publicly reported facts without paying a toll.”
I wish the AI bros at Perplexity and elsewhere a very cope and fucking seethe.
Okay, quick personal sidenote:
With how much misinformation, manipulation, outright theft and other horrific shit this AI bubble has caused, I suspect we're gonna see some attempts at an outright ban on AI. How successful they're gonna be, I don't know, but at the bare minimum it'll enjoy some popularity on the political fringe.
Burglars telling homeowners to cope and seethe when questioned about their possession of crowbars at time of arrest.
You know you're dealing with serious people when the dogwhistles come out.
They prefer to live in a world where publicly reported facts are owned by corporations, and no one can do anything with those publicly reported facts without paying a toll.
Yea, down with corporate IP trolls, information gatekeepers and idea landlords! Anyway, what was Perplexity's business model again?
idea landlords: making sure that no one is living rent free in someone elses head
they wish this technology didn’t exist
this is supposed to be invalidating, but like... yes? what's wrong with that?
Crypto mining firms based in Sweden are accused of withholding around $100M in unpaid taxes.
Mostly VAT fraud.
News in Swedish: https://www.svt.se/nyheter/lokalt/norrbotten/kryptoforetagen-lurade-staten-pa-en-miljard
But your honor, we're a crypto mining company; we don't add any value!
HEARTWARMING: Baldy McDickface to step back from podcasting now the Russian money has dried up
Pretty good sneer there:
BREAKING: Tim Pool announces he will be stepping back from full time content production to look after his family. He states he's tired of being made fun of for not having a wife and kids so he will also be using the extra time to pursue acquiring that family
view more: next ›