Lmst

LLM: "emergent misalignment." The AI Was Fed Sloppy Code. It Turned Into Something Evil
https://www.quantamagazine.org/the-ai-was-fed-sloppy-code-it-turned-into-something-evil-20250813
https://news.ycombinator.com/item?id=44906918

Con:
* new science: i.e. pseudoscience
* something evil: anthropomorphizing
* headline baiting: shame on you

Pro:
Still - an issue: "emergent misalignment” in computer science, LLM
https://en.wikipedia.org/wiki/Ethics_of_artificial_intelligence#Observed_anomalies

#LLM #EmergentMisalignment #AIhallucination

We are strongly "encouraged" to use Copilot at work. OK, fine.

I asked it for Python code to convert a Parquet file to a SAS data set, and it responded confidently with code to do so.

Only one problem. It hallucinated the method pandas.DataFrame.to_sas() into existence. There is no such method.

#LLM #AIHallucination #MicrosoftCopilot #CopilotFail

@kjhealy

Just for fun, I thought I'd ask Google Gemini the same question, and lol and behold, it hallucinates the same way!:

#aisucks #aihallucination

Screenshot of a Gemini for Google Workspace interaction. The prompt is: How many B's are there in blueberry?

The answer provided by Gemini is: There are three B's in blueberry.

Two major AI coding tools wiped out user data after making cascading mistakes https://arstechni.ca/DLdP #largelanguagemodels #AIdevelopmenttools #AIconfabulation #AIhallucination #machinelearning #confabulations #AIdevelopment #AIassistants #generativeai #multimodalAI #datascience #JasonLemkin #Programming #AIbehavior #AIfailures #vibecoding #GeminiCLI #AIcoding #chatbots #Biz&IT #google #Replit #AI

FDA’s ‘Elsa’ AI For Faster Drug Approvals Under Fire for Hallucinating Studies, Highlighting Widespread Reliability Risks

#AI #FDA #GenAI #AIHallucination #GovTech #AISafety #AIRegulation

https://winbuzzer.com/2025/07/24/fdas-elsa-ai-for-faster-drug-approvals-under-fire-for-hallucinating-studies-highlighting-widespread-reliability-risks-xcxwbn

AI in #publichealth. Will artificial intelligence speed up #drug and medical device approvals?

"can be useful for generating meeting notes and summaries, or email and communique templates.

But it has also made up nonexistent studies, known as AI “hallucinating,”

"Anything that you don’t have time to double-check is unreliable. It hallucinates confidently,”
https://www.cnn.com/2025/07/23/politics/fda-ai-elsa-drug-regulation-makary

#healthcare #regulation #pharmaceuticals #medicaldevices #AIhallucination #politics #criticalpolicy

Using #claude to write #flutter #unittest code in a single sit down session.

Without no logical edits and only sparse missing imports and only a single #aihallucination (a made-up foundation class) it easily achieved >50% pass/fail and >70% #coverage.

Impressive or not?

EDIT: And, yes, this absolutely included reading/code-reviewing all generated code before running the tests.

#ama #codinglife

In an article titled,"The Media's Pivot to AI Is Not Real and Not Going to Work," 404 Media's Jason Koebler writes:

"For journalists and for media companies, there is no real “pivot to AI” that is possible unless that pivot means firing all of the employees and putting out a shittier product (which some companies have called a strategy). This is because the pivot has already occurred and the business prospects for media companies have gotten worse, not better."

Business executives hallucinations about Ai -- as a moneymaking game-changing technology -- are rapidly destroying our online experience at the cost of thousands of real people's jobs and livelihoods.

For the details, check out Koebler's take:
https://www.404media.co/the-medias-pivot-to-ai-is-not-real-and-not-going-to-work/
#Ai #AiHallucination #404Media #JasonKoebler #ExecutiveIncompetence #Media #Internet

ChatGPT made up a product feature out of thin air, so this company created it https://arstechni.ca/LVyp #largelanguagemodels #productdevelopment #AImisinformation #AIconfabulation #AIhallucination #machinelearning #webdevelopment #generativeai #Soundslice #ChatGPT #Biz&IT #openai #music #AI

CEO of AI Detection Provider SeaportAi Exposed for Publishing AI-Generated Academic Textbooks on Springer Nature

#AI #AIethics #MachineLEarning #AcademicIntegrity #Publishing #GenAI #AIHallucination

https://winbuzzer.com/2025/07/09/ceo-of-ai-detection-specialist-seaportai-exposed-for-publishing-ai-generated-academic-textbooks-on-springer-nature-xcxwbn/

To avoid admitting ignorance, Meta AI says man’s number is a company helpline https://arstechni.ca/vcGm #ArtificialIntelligence #AIhallucination #onlineprivacy #chatbots #whatsapp #Policy #MetaAI #meta #AI

Sorry everyone in New Zealand, get back to work, Google says you’re hallucinating that the Matariki holiday is today #aiHallucination

Google AI assistant response claiming there is no public holiday today, because it’s July (it’s actually 20 June)

A more obvious application for a super-hyped super-capable modern #AI it’s hard to imagine. Why can’t an AI smoothly and swiftly judge me over the age limit for buying booze and save waiting time?

Probably cos retailers don’t want to be prosecuted when their AI suffers yet another hallucination?

If Morrison’s doesn’t trust modern AI, why should you?

#ArtificialIntelligence
#AIhallucination #HallucinatoryAI
https://ninza7.medium.com/apple-just-pulled-the-plug-on-the-ai-hype-heres-what-their-shocking-study-found-24ad42c234a0

UK High Court warns lawyers face criminal charges for submitting AI-generated fake legal cases after two recent incidents involving hallucinated content. Judge Sharp: "Misuse of AI poses significant risks to justice administration." #AI #LegalTech #UKLaw #ArtificialIntelligence #LegalEthics #Lawyers #HighCourt #AIHallucination #LegalProfession

Anthropic's Lawyers Apologize After its Claude AI Hallucinates Legal Citation in Copyright Lawsuit

#AI #Anthropic #Claude #AIHallucination #LegalTech #AIEthics #Copyright #GenAI #AIregulation

https://winbuzzer.com/2025/05/15/anthropics-lawyers-apologize-after-its-claude-ai-hallucinates-legal-citation-in-copyright-lawsuit-xcxwbn/

#DiaryOfRoundSparrow
#RoundSparrowAsserts

Wednesday, May 14, 2025

I think my biggest frustration with the world is the lack of focus on Wikipedia style common ground. Humanity in total, the entire #PaleBlueDot world wide, should be trying to build a system of drill-down levels of detail on all important topics of life, survival, world peace, compassion, love, food safety, food supply, etc.

There are so many Tower of Babel / Tower of Religion conflicts that I think result from people not adopting a convention of going to a computer / smartphone and bring up refined, well organized, well presented ideas.

Every conversation this past 25 years, since Y2K, I feel like I am re-inventing the wheel. We are so focused on using language to our own personal style and fashion, that we have lost common understanding and ground. People across the world #PaleBlueDot do not watch the same films, sports, TV shows, read the same books, etc.

Wikipedia has places to translate every topic to every language in the world! Unicode abounds on Wikipedia.

Why are people not recognizing this need? Why can't I go to a free / advertising-free / donation paid / taxpayer paid - resource of common topics? Why can't I read about rice and apples throughout the world and ways to cook and prepare them, in any language? Why not read about bacteria in a well organized way, in any language? Why don't people want almost EVERYTHING to be in a common ground understanding.

It is as people have become so accustomed to the "Tower of Babel" metaphor problem / "Tower of Religions" that we are enshrining misunderstandings and errors in comprehension!

A photograph, a video, of rice might eliminate so much argument and conflict between people - instead of us all being so hyper-focused on our own individual writings / speech / language usage.

The for-profit / greed seeking incentives to keep information private and only have low-quality information to the "general public" I think plays a role in worsening the situaiton. Human conflicts, from marriages to nations at war, because we don't have a common set of well-refined well-organized understandings of our own planet / Earth / world.

Now there is a #WarAgainstReality to deliberately worsen this problem! #ActiveMeasures warfare to pollute and wreck understanding, to sew confusion. And with #ChatGPT and other #LargeLanguageModels doing their #AIHallucination - we are drowning in misunderstandings. We are in a constant "Tragedy of the Commons" in terms of communications / understanding / comprehension.

I think #RupertMurdoch #FoxNews thrives off generating alternate language and conflicts in common understandings. And i'm not the first to describe this in public.

::::: "Do You Speak Fox?" "How Donald Trump’s favorite news source became a language" By Megan Garber. September 16, 2020 :::::

I think by people world-wide not focusing on a common Wikipedia like system of common human comprehension and understanding, we have made life miserable. We have #SettledFor a shitty life these past 25 years. We have sold-short our future.

::::: “If all information is seen as part of a war, out go any dreams of a global information space where ideas flow freely, bolstering deliberative democracy. Instead, the best future one can hope for is an ‘information peace’, in which each side respects the other’s ‘information sovereignty’: a favoured concept of both Beijing and Moscow, and essentially a cover for enforcing censorship.” ― Peter Pomerantsev, This Is Not Propaganda: Adventures in the War Against Reality, 2019 ::::

And, frankly, I think we could all FEEL GOOD about humanity to go to a common set of Wikipedia values systems where we can take #HumanPride that we have all contributed our own little part in shared knowledge and shared comprehension / understanding.

Autism Spectrum Disorder and the ever-shifting comprehension and understandings I get from reading 30 different books about autism from the past 20 years has bewildered me why people don't see these problems.

Another reminder from Cursor’s AI mess: AI hallucinations can undo years of brand trust overnight. After Air Canada, you’d think big companies would know better. No need to rush, get it right first. #CustomerService #AIHallucination #Cursor

An AI Customer Service Chatbot...