I bet we're going to see public APIs for most platforms following this trend and getting nerfed, as they are no doubt being abused for harvesting massive amounts of training data for pseudo-"AIs".
That's a job killer for other business consumers of those APIs. #stochasticparrots
When Sam Altman declared, “I am a stochastic parrot,” we should take that to mean he doesn’t know what he’s saying and can be relied upon to make up facts.
@therealjimlove @briankrebs Lemoine’s "interview" is no more convincing today than it was a year ago. LaMDA is no more "Artificial Intelligence" than are #ChatGPT, #Bard, or any other large language model. Calling them "AI" is fraudulent. The charlatans wringing their hands over the bullshit fears of 'takeoff' and 'hyper-intelligent AGI' all have vested interests in that fraud.
"The #technical phenomenon is the preoccupation of the vast majority of people of our time to seek in all things, the absolutely most efficient method".
Wrote Jacques #Ellul in his book "Le Système technicien".
#efficiency #technocriticism #ethics #sociology #engineering #quotes #AI #ChatGPT #digitalisation #philosophy #ethicalAI #AIEthics #technology #tech #stochasticParrots #generativeAI
We (mostly) seem to have made it through the Blockchain and cryptoscam hype with only medium collateral damage to most of the population (highly profitable ransomware and burning much more dirty coal than would otherwise have happened), but unfortunately at the cost of stumbling right into the next one. I would hope that we could get that one over with with less collateral damage, but I am not optimistic at this point.
Which damage do I fear?
* Again, massive energy consumption through training too-large models on too much data. Most probably not on the level of Blockchain proof-of-work idiocy, but still too much for our already-unstable environment.
* Further, potentially irreversible erosion of public trust in online information because of verified facts being drowned in the to-be-expected, heavy industrialization of fake information production in all types of media.
What can we do to mitigate that collateral damage?
* Stop calling it "Artificial Intelligence". Stop calling its errors "hallucinations". Stop speculating about "sentience" or "consciousness". We are talking about Large Language Models (LLMs) as a very specific sub-field of Machine Learning (ML) right now. Nothing else.
* Instead of blindly training and playing with huge models treated as black boxes, invest much more in human understanding of what actually goes on inside - whitebox, interpretable, deterministic, repeatable, and reproducible behavior of tools we intend to build applications on. Instead of being baffled by seemingly emergent properties, maybe ask more deeply why those properties emerge? Instead of slapping it onto every product/service you can reach, maybe ask if that particular use case actually benefits from adding a stochastic parrot?
* Stop neglecting basic security and privacy principles just because "this is a new thing". It's not. Security means separation of code (initial prompts with priming statements) and data (untrusted user input) instead of mangling them all together. Privacy means well-defined notions like unlinkability, anonymity without the possibility to re-identify, and informed consent before ever touching any personal data in the first place.
* It probably means again paying for good journalism. You know, actual reporters cross-checking and verifying their supposed facts before shouting them out.
* And yes, maybe it means regulation. China regulating Bitcoin mining had a positive global effect on its energy consumption (it's still a dumpster fire, but slightly less bad than it was before). Regulating “AI” is probably what it will take to mitigate some of the worst effects of this hype, but I fear that such kind of regulation will come much later than the damage.
#Rant #AI #LLM #LLMs #StochasticParrots #CollateralDamage #FakeNews #IndustrializedFakes
Did you ever stop and consider that the opposite of Artificial Intelligence would be Real Stupidity?
Just going to pop this here for future reference.
(#DuckDuckGo )email protectionサポートページ 、(DuckDuckGo)プライバシー規約 、(NYmag)あなたはオウムではないし、チャットボットは人間ではない 、「#AI の一時停止」書簡に対する『#StochasticParrots 』執筆者たちの声明文 など。
» And what I was struck by there, was how it combined generic things that were true, with very specific things that were wrong
Lot of people say that large language models hallucinate,
I rather think they can fabulate
Seamlessly make stuff up, fill in gaps in a way they're not even aware of doing «
#chatgpt #gpt4 #LLMS
Really good post, allegedly from a Googler, that is about the revenge of open-source coming to roost now with AI and LLMs (with echoes of #StochasticParrots & @emilymbender on the folly of really large models):
I just started using Bard to help me learn Google Cloud console.
I’m considering the various models, but need an AI assistant and Learning & Development (#EdTech space) to search and discover the appropriate context.
Otherwise, democratized machine learning apps are like the Library of Alexandria without a card catalog or a Hypatia to explain the Gods of Lies™ and what you need
#GoogleCloud #learning #democratization #AIapps #librarian #cybrarian #GoogleBard #stochasticparrots
Proof that even people can be #StochasticParrots when they repeat things without understanding the context
a colleague of mine, an expert in NLP and driving a community at my employer, called me the other day "You are our Emily.".
My attempts to educate my colleagues doesn't go unheard.
Thanks @emilymbender , @timnitGebru and @DAIR
every single time a hyped A.I. a #SALAMI antidote comment is applied by me pointing to your contributions.
PS: I refused the intended compliment, just being little old me.
Gerade im ICE einen Gesprächsfetzen mitgehört von vier Marketing-Mitarbeiterinnen eines nicht identifizierten Unternehmens, in dem der Satz fiel: "Ich schreibe jede E-Mail mit ChatGPT."
Jetzt denke ich darüber nach. Einerseits gehen dadurch berufliche Grundfähigkeiten verloren, andererseits entsteht hier auch ein neues Berufsbild von Maschinen-Bediener*innen, für die generative KI ein massiver Ideen- und Output-Beschleuniger sein wird.
If it weren't for Artificial "Intelligence", the GOP wouldn't have any intelligence at all.
What is the training data of #LLMs?
#ChatBots / #GenerativeAI / #stochasticParrots get education from everything we ever wrote for the web.
An article by Scott Rosenberg @scottros: https://www.axios.com/2023/04/24/ai-chatgpt-blogs-web-writing-training-data #dataGovernance @dataGovernance #copyright #microBlogging #webPublishing #blogging #blogs
An interview about #generativeAI / #stochasticParrots with the European Commission's executive vice president and commissioner for competition:
"I do hope that we can find political agreements later this year, so that [new rules] can come into effect."
#Vestager said the proposed rules will not affect research.
"You can test, you can innovate, you can pursue your ideas. What we are regulating is when it's put into use."
https://asia.nikkei.com/Editor-s-Picks/Interview/EU-eyes-new-rules-for-generative-AI-this-year-Vestager #ChatGPT #OpenAI #EU #dataPrivacy
“If ChatGPT doesn’t understand anything and can’t reason, then how do you explain [this ability] or [that ability]? If it just guesses the most likely next word, would it be able to do all of that?”
If the training data set is large enough the model is complex enough, and—crucially—if both the inputs and the outputs of the task you’re describing CAN BE REPRESENTED IN WORDS, then yes, an LLM can do (or appear to do) those tasks without understanding or reasoning.
More proof that Large Language Models #LLMs are not "intelligent": you can "hack" them with something that a human can immediately recognise as a prompt injection. Calling the stochastic parrots "artificial intelligence" is maybe marketing genius, but as of yet a gross misrepresentation.
#artificialinteligence #ai #ki #generativeai #stochasticparrots
The problem with #stochasticparrots isn’t the technology itself, it’s people: https://techhub.social/@Techmeme/110223895710586581
Más ejemplos de comportamiento de los modelos de lenguaje que desacreditan las metáforas de “Loros estocásticos” y "generadores de gilipolleces"
#chatGPT #BullshitGenerator #StochasticParrots
as frustrated i am with the whole #stochasticParrots meme. the differences of language models and people are also just interesting and important in their own right.
their lack of intention alone is such a rabbit hole. can there knowledge without intention and autonomy? i don't think so, it's more like pure familiarity without any pragmatics, except maybe about text and language itself?
what a weird frickin thing.
@benjedwards To say 'it tries' implies intention and volition. While I agree it's become routine to anthropomorpise computer programs, we should be especially careful about ones about which claims to intelligence are made.
It does not 'try to'. It has no semantic model of the text, which is precisely why it produces plausible untruths.
It is *programmed* to complete a transcript of a chat.
And its algorithm is plausibility-seeking, not truth-seeking.
Thank you for this 7 post thread on
#AGI #AIHype #AISafety #EthicsInAI #LLM #AISEO #AIPause #StochasticParrots #EmilyBender
Excited to see my school, University of Michigan - Dearborn, once again, doing the right thing by insisting on an opt out of Turnitin's flawed and problematic #EdTech
This tech will harm students. The company is rolling it out in a way that is particularly harmful.
Offering this tech as an option is one thing but it is being forced and rolled out too quickly!!
Since we've been looking for more things to do, @firstname.lastname@example.org @email@example.com @firstname.lastname@example.org and I wrote a statement about the horrible "letter" on the AI apocalypse, the very first citation of which, was our #StochasticParrots paper.
#justdrafts | #linkedlist — A Response from the “Stochastic Parrots” Authors to the Future of Life Institute’s Open Letter
Gebru, Bender, McMillan-Major, and Mitchell being on fire.
A 'pause' in the context of AI development if only a 'pause' does nothing to address problems with the way AI has developed - 'the explosion of synthetic media in the world' has already happened and, as is, it 'reproduces systems of oppression and endangers our information ecosystem': important @DAIR response by @emilymbender @timnitGebru @meg and Angelina McMillan-Major to the AI 'pause letter'
#StochasticParrots #AI #PauseLetter #AIEthics
Another #AIWinter is coming
As soon as the investors realize that none of this tech is actually ready for use they will bail and cut their losses
As soon as "self-driving car" investors realise that true #autos are still decades, maybe even centuries, away they will stop funding this nonsense
As soon as investors realise that these "language models" and "generative AI" are all legal minefields and lawsuits waiting to happen they will distance themselves faster than you can say "#SyntheticMedia"
And once all the current patents have expired and everyone involved has moved on, someone will have a bright idea and find someone with money willing to listen, and they will start another AI cycle all over again...
All this has happened before, and all this will happen again
Italy puts temporary ban on ChatGPT.
I’m assuming it’s the risk of psittacosis 😄
Sadly, this article’s authors choose to imply causation by saying the ban comes “days after” the open letter on AI, as well as detailing all the *actual* reasons specifically advanced by the Italian authorities.
If you want a reasoned critique of the open letter, look here:
US Schools are opting out. My school is one of them.
But no one has been told opting out is an option.
You have to ask (and I've heard sometimes pressure) Turnitin to #turnitoff
And you should!
This is flawed #EdTech snake oil! This is Sea Monkeys! They can't prove that it works. They are not letting you test it. They are forcing it on your school!
I've also heard that over 100 institutions in the UK have asked that the feature be turned off thanks to pressure from JISC and others ( @Lawrie can you confirm?) And I have to ask where the heck are @educause ISTE and other US based #Edtech orgs?
I do know of at least three major news outlets (one focused on #highered and two national outlets) working on this story so maybe that will get some attention.
Turnitin Customers were told that this AI detection would be rolled out as new feature on April 4th and that there was no option to turn it off.
That in January the feature would become a paid option.
This is a drug dealer sales tactic. First hit is free. I've seen this tactic many times but I'm not sure I've ever seen it with such a shady product that has such a huge implication for harm.
I've been spending a lot of time thinking (& not enough time writing; sorry @DonnaLanclos) about #highered #edtech procurement processes so this Turnitin AI Detection thing has me all wound up.
Let's take stake and attempt a little thread.
First let's just call AI Detection what it is - a flawed measure that is needed in the big picture but when reduced to a single score and in the hands of those who are looking to convict - is just plain dangerous.
Can anyone recommend an approachable, high-level, hype-free introduction to how neural net text-to-image generation works, that i can assign to my students who aren’t (universally) CS-literate? Including discussion of ethical issues around data collection, ideally.
This week's net.wars, "Unclear and unprsent dangers", complaining about the distraction of futurist fantasies was rendered near-obsolete by the Italian DPA's announcement that ChatGPT must stop processing Italian users' data: https://netwars.pelicancrossing.net/2023/03/31/unclear-and-unpresent-dangers/ #NetWars #AI #chatGPT #StochasticParrots #SocialJustice
I just got word that Turnitin has allowed our institution to opt out of the AI Detection features they were forcing on customers till Jan when they planned to start charging for the feature.
I've heard that any institution can opt out but you have to request to be put on the "suppression list"
They are not advertising this - you have to ask for it. They are providing no research or proof about the accuracy of their AI detection.
People are claiming that gpt5 which is scheduled for december will reach AGI. What do you think?
https://www.linkedin.com/posts/alighodsi_hello-dolly-democratizing-the-magic-of-chatgpt-activity-7045014694385917952-JUVH?utm_source=share&utm_medium=member_desktop "We are open sourcing Dolly, a ChatGPT-like model that can do instruction following, brainstorming, summarization. The remarkable thing is that we created it for $30 using one server for 3 hours on a small dataset using a 2 year old open source LLM model. The secret in magical human-like interactivity probably lies in a small dataset, not huge models, not training for 100,000 GPU hours." #StochasticParrots #LLM #MachineLearning
What is your opinion on the wolfram chatgpt plugin?
Does this accelerate the road towards AGI? Or is it just corporations trying to cash in on the hype? Or somewhere in between?
And how would you compare wolfram+chatgpt with vipergpt ? Would it be easy to hook some more functionality into vipergpt to be able to do similar things?
#wolfram #ai #aihype #stochasticparrots #agi #chatgpt #chatgpt4 #gpt4 #wolframchatgpt #vipergpt
I think you are right. So let's regroup:
AI will disproportionately harm marginalized people
AI detection will disportionately harm marginalized people
Where does this end?
#Handelsblatt moechte einem gewissen Sebstian Matthes Erkenntnis und eine differenzierteren Blick wuenschen, gerade hinsichtlich #ChatGPT; in einer vernetzten Welt sind fundierte Informationen durchaus verfuegbar.
Moeglicherweise hilft der Hashtag #StochasticParrots und #StochasticParrotsDay .
Hier sind beide nicht vertreten, oh welch Ueberraschung.
#StochasticParrots paper reading ...
me is stunned by the example given of GPT-3 response on #WagnerGroup which clearly shows the dangerous bias incorporated; in my conception I even can't state the crimes those stand for, even the link chosen to Wagner and the historical implications that go with it (Fascism, Nazi-Germany) are deeply concerning me.
members even have tattoos underlining the intended historical context. (=>major criminals)
GPT3's response is very different.
Machine-Augmented Humanity: Or How We Handed Loaded Firearms to Children, Turned the Safety Off, and Hoped for the Best
#humanity #generativeAI #education #children #risk #consumerharm #ESG #misinformation #activemeasures #socialmedia #stochasticparrots #OpenAI #LLMs #AIgovernance #datagovernance #attentioneconomy
#StochasticParrots Day including #catcontent. Love it so far.
This week's net.wars, "Performing intelligence", looks at GPT-4, invokes Clever Hans, and asks "Are we *nuts*?" https://netwars.pelicancrossing.net/2023/03/17/performing-intelligence/ #NetWars #AI #stochasticparrots
@clhellisen Full disclosure, I too believe this. The conclusions of On Computable Numbers is inescapable: either the human brain is Turing complete or it isn't. In either case it can necessarily be emulated by a machine which is Turing complete.
However, nowhere is it said this would be easy, and #StochasticParrots are not a step in the right direction.
Since we published the #StochasticParrots paper two years ago, the issues discussed in it have only become more urgent and salient. Join me, @timnitGebru @meg Angelina McMillan-Major and an esteemed group of panelists for discussion and reflection, March 17 2023.
This was a dig at the #StochasticParrots paper, a comprehensive, measured roundup of criticisms of AI that led Google to fire @timnitGebru, a respected AI researcher, for having the audacity to point out the Emperor's New Clothes:
Gebru's co-author on the Parrots paper was @emilymbender, a computational linguistics specialist at UW, who is one of the best-informed and most damning critics of AI hype.
I really enjoyed this episode of the Tech Won't Save Us podcast with @timnitGebru
Listening to Timnit reflect on how fast the things we warned about in the #StochasticParrots paper came to pass left me wondering: what if instead $Billions we put into clean energy (or making the grid alternative energy ready or carbon capture or...)?
Wow, this really is the definitive text about #ChatGPT & Co.
You Are Not a Parrot
And a chatbot is not a human.
And a linguist named @email@example.com is very worried what will happen when we forget this.
"#LLMs are tools made by specific people — people who stand to accumulate huge amounts of money and power, people enamored with the idea of the #singularity. The project threatens to blow up what is #human in a species sense. But it’s not about humility. It’s not about all of us. It’s not about becoming a humble creation among the world’s others. It’s about some of us — let’s be honest — becoming a superspecies. This is the darkness that awaits when we lose a firm boundary around the idea that #humans, all of us, are equally worthy as is."
#AI #Chatbot #stochasticparrots #newstodon
As always Barbara Fister is thoughtful and devastating
INFORMATION LITERACY, WRITING INSTRUCTION, AND THE PROBLEM OF STOCHASTIC PARROTS
"Powiedzmy, że A i B, oboje biegle mówiący po angielsku, zostają niezależnie od siebie uwięzieni na dwóch niezamieszkanych wyspach. Wkrótce odkrywają, że poprzedni goście na tych wyspach pozostawili telegrafy i mogą się ze sobą porozumiewać za pomocą podwodnego kabla. A i B zaczynają radośnie pisać do siebie wiadomości.
Tymczasem O, hiperinteligentna ośmiornica głębinowa, która nie jest w stanie odwiedzić lub obserwować żadnej z wysp, odkrywa sposób, aby podłączyć się do podwodnego kabla i słuchać rozmów A i B. O początkowo nie wie nic o języku angielskim, ale jest bardzo dobry w wykrywaniu wzorów statystycznych. Z czasem O uczy się przewidywać z dużą dokładnością, jak B zareaguje na każdą wypowiedź A.
Wkrótce ośmiornica wkracza do rozmowy i zaczyna podszywać się pod B i odpowiadać A. Ten podstęp działa przez jakiś czas, a A wierzy, że O komunikuje się tak samo jak ona i B - z sensem i intencją. Potem pewnego dnia A woła: "Jestem atakowany przez wściekłego niedźwiedzia. Pomóż mi wymyślić, jak się bronić. Mam kilka kijów." Ośmiornica, podszywająca się pod B, nie potrafi pomóc. Jak mogłoby się to udać? Ośmiornica nie ma żadnych referencji, nie ma pojęcia, czym są niedźwiedzie czy kije. Nie ma możliwości udzielenia odpowiednich instrukcji, jak na przykład pójść po kokosy i linę i zbudować katapultę. A jest w kłopocie i czuje się oszukany. Ośmiornica zostaje zdemaskowana jako oszust..."
Cała tekst o tym co się stanie jeżeli zapomnimy, że chatbot nie jest człowiekiem:
#ai #octopus #stochasticparrots
this is excellent. it's excellent.
We’ve learned to make "machines that can mindlessly generate text," @emilymbender told me when we met this winter. "But we haven’t learned how to stop imagining the mind behind it."
and from blake lemoine, the engineer who got fired from google for asserting that the #LLM was sentient:
Whether these things actually are people or not...the whole point is you can’t tell the difference. So we are going to be habituating people to treat things that seem like people as if they’re not.
#ai #chatgpt #openai #effectiveAltruism #stochasticParrots #linguistics
Very nice profile of @emilymbender in this thought-provoking and hard-hitting NY Mag piece — on text generators posing as humans and humans calling themselves parrots https://nymag.com/intelligencer/article/ai-artificial-intelligence-chatbots-emily-m-bender.html #chatgpt #stochasticparrots #nlu
It keeps getting better:
"'This is truly an intolerable situation. If you both lie,'
"-- 'And lie convincingly,' added Bingle-Dum --
"'Yes, thank you. If that is so, then how am I to ever trust either of you?'
"Googl-E and Bingle-Dum turned to face each other and shrugged."
Chatbot search is a terrible idea, especially in an era in which the web is likely to fill up with vast mountains of AI #bullshit, the frozen gabble of #StochasticParrots:
@timnitGebru @emilymbender @mmitchell_ai
Hmmm. I clicked on the link - #Google & Co. now know that I'm interested in #LanguageModels and the #StochasticParrots Day.
#DataIncontinency #Privacy #donotfeedthegoogle
Outright lying about the claims of the #StochasticParrots paper from @emilymbender, @timnitGebru, et al. — it's not a good look for Yann LeCun.
The paper indeed explores harms and climate impact. But nowhere does it claim that LLMs are useless, and the paper is saturated with examples and discussion of precisely the uses that LeCun claims it denies.
Zen and the Art of Metal Musicianship https://decafbad.net/2023/01/27/zen-and-the-art-of-metal-musicianship/ #StochasticParrots #ADayInTheLife #ChatGPT3 #misc