Now That is Some Bullshit!

1st up, from John Kwarsick:

The Machine That Can't Explain Itself

Why AI adoption is outrunning accountability and what happens when the auditors arrive

From 2024 to 2025, the Foundation Model Transparency Index published by the Stanford Institute for Human-Centered AI dropped from 58/100 to 40/100. Oops.

In my very 1st anti-LLM rant, I described LLMs as "indecipherable oracles" - I don't see how that's going to change.

---

This next I find almost unbelievable. LLMs will tell you what's in an image without seeing the image?!?!? They just confabulate what's likely to be in the image?!?!? Holy crap, now that's some bullshit! From Gary Marcus:

The mirage of visual understanding in current frontier models

When a model achieves a “top rank on a standard chest X-ray question-answering benchmark without access to any images” you know something is deeply wrong.

Marcus is reporting a research paper just released by Stanford:

MIRAGE: The Illusion of Visual Understanding

First, Frontier models readily generate detailed image descriptions and elaborate reasoning traces, including pathology-biased clinical findings, for images never provided; we term this phenomenon mirage reasoning. Second, without any image input, models also attain strikingly high scores across general and medical multimodal benchmarks, bringing into question their utility and design. In the most extreme case, our model achieved the top rank on a standard chest X-ray question-answering benchmark without access to any images. Third, when models were explicitly instructed to guess answers without image access, rather than being implicitly prompted to assume images were present, performance declined markedly.

Another great new term in the dream world of LLM hallucinations: "mirage reasoning".

This tech is clearly NOT ready for prime time, but, $T invested, tough shit. You will take your bullshit sandwich & enjoy eating it, nom-nom!

I've downloaded the PDF from arxiv, reading through it, it's scary but also LOL. I shouldn't laugh, the Bullshit Apocalypse is not funny, but laughter has always been a part of what keeps me sane. So, laugh, dumbass, laugh!

---

Here's the home page/directory for my posts on Bullshit. This is post #113.

Yay! Wow, 2 Yays in a Row! The Tide Is Turning!

A big YAY yesterday, another big YAY today! The tide is turning (or not). From the most excellent 404 Media:

Wikipedia Bans AI-Generated Content

“In recent months, more and more administrative reports centered on LLM-related issues, and editors were being overwhelmed.”

1 of the worst things? The worst thing? About AI slop is, it overwhelms human capabilities to fact check, edit, stave off the Bullshit Apocalypse. Wikipedia has totally done the right thing. Just shut the slop out completely. Kudos.

---

Here's the home page/directory for my posts on Bullshit. This is post #112.

Yay! Disney Bails On OpenAI-Generated Slop Video

After only 3 months, Disney bails on a $1B investment in OpenAI to allow anyone to create slop videos using Disney characters. The concept was apparently a lead zeppelin. From 404 Media:

Disney's Sora Disaster Shows AI Will Not Revolutionize Hollywood

It turns out when you try to serve slop on a product people pay for, no one wants it.

---

Here's the home page/directory for my posts on Bullshit. This is post #111.

Notably Mixed Results

1st, a tale of woe, from The Leadership Lighthouse:

I Went All-In on AI. The MIT Study Is Right.

My all-in AI experiment cost me my confidence

...

The company gets excited about AI. Leadership mandates AI adoption. Everyone starts using AI tools. Productivity metrics look great initially. Then something breaks, or needs modification, or requires actual judgment, and nobody knows what to do anymore.

The developers can’t debug code they didn’t write. Product managers can’t explain decisions they didn’t make. Leaders can’t defend strategies they didn’t develop.

---

Next, a good review of the state of things. Interesting that he picks Yann LeCun as "the voice of sobriety about AI in today’s world". LeCun was a major "AGI soon" cheerleader until fairly recently. Ditto for Geoffrey Hinton, also profiled in the article, still a cheerleader to my knowledge. From PhD computer scientist David William Silva:

I'm Sorry to Burst Your Bubble: You Are Being Fooled About AI, and You Will Soon Feel Really Stupid

The overall inability to distinguish what is real from what is fake is fueling the frisson.

This I had not heard before, but it is very believable & a matter for concern:

And here is what I’ve found: every time I had an idea that was genuinely outside the box, AI discouraged me from pursuing it. When I persisted, it didn’t just push back, it practically begged me to stop. To not continue. To not insist.

---

3rdly, from 404 Media, a link to a website dedicated to exposing hype:

Ridicule as Praxis (with Emily Bender and Alex Hanna)

Why ridicule works to keep big tech’s claims in check, and what makes us hopeful for the future.

Here's the website:

The Mystery AI Hype Theater 3000 podcast

“deflates AI hype and draws attention to the real harms of the automation technologies we call ‘artificial intelligence’.”

---

Finally, some good news! Generative AI problems solved! A guest article from the AWS blog:

Overcoming LLM hallucinations in regulated industries: Artificial Genius’s deterministic models on Amazon Nova

Artificial Genius uses SageMaker AI to perform a specific form of instruction tuning on Amazon Nova base models.

This patented method effectively removes the output probabilities. While standard solutions attempt to ensure determinism by lowering the temperature to zero (which often fails to address the core hallucination issue), Artificial Genius post-trains the model to tilt log-probabilities of next-token predictions toward absolute ones or zeros. This fine-tuning forces the model to follow a single system instruction: don’t make up answers that don’t exist.

This creates a mathematical loophole where the model retains its genius-level understanding of data but operates with the safety profile required for finance and healthcare.

The authors state that this approach is much better than RAG, which was already supposed to fix hallucinations. I guess we'll see. Like RAG, it looks like after-the-fact bubble gum & duct tape to me.

The primary author is Paul Burchard, PhD, "Founder and Partner of Artificial Genius". He has an interesting resume.

---

Here's the home page/directory for my posts on Bullshit. This is post #110.

Wendig Triples Down

I think that most excellent sci-fi author Chuck Wendig has an attitude towards generative AI at least as negative as mine. In fact, his attitude is probably much more negative than mine because, as a professional author, he has serious skin in the game. He doubled down on AI-bashing in December, he now triples down.

Shy Girl, AI In Writing, And A New Perniciousness

I am against AI because it steals our work, which it then uses to steal our jobs, which it further uses to steal our water and our electricity.

Which is to say, it is here to steal our future.

So, I’m against it! It sucks moist open ass.

...

And don’t use the AI slop-shitting artbarf techbro bullshit.

Wendig's post talks about "The Terminator" movie & includes a pic, which I have copied. Who doesn't like a pic of a terminator walking through fire?

---

From The Register, which appears to now be a part of MSN?

AI still doesn't work very well, businesses are faking it, and a reckoning is coming

Codestrap founders say we need to dial down the hype and sort through the mess

Note, most of the content of this article comes from an interview with "Dorian Smiley, co-founder and CTO of AI advisory service Codestrap", who I'm gonna guess is selling something. The content is decent. But still, capitalism FTL or FTW, you decide.

---

From MindPrison:

How Verifier Loops Made AI Coding Useful | The Rise of Vibeware, Abandonware and Technical Debt as Consequence

How did unintelligent, hallucinating AI progress from nearly useless to writing entire, complex applications and at what cost?

This is a long article, lots of good news, bad news. Despite the positive headline, I'm going to posit that the bad news outweighs the good news. For example:

“Vibe Coding makes you 10x faster at making 10x more technical debt.” — Franziska Hinkelmann, PhD

From the 80s on, we called it "maintenance debt", not technical debt. Same thing. Writing the software is not the hard part, handing it off & maintaining it is.

---

Here's the home page/directory for my posts on Bullshit. This is post #109.

Now That's Some Bullshit!

From Cory Doctorow, the Bard of the Revolution & creator of the term "enshittification", some ruminations on the nature of words & bullshit, particularly corporate bullshit. Not particularly AI related for a change. Includes a very good definition of bullshit.

Love of corporate bullshit is correlated with bad judgment

Synergizing the strategic inflection points on the global data network.

...

Semantic drift is a feature, not a bug.

...

Colloquialization doesn't dilute language, it thickens it.

...

Littrell set out to evaluate "corporate bullshit," a linguistic category that is separate from mere "jargon." Jargon, Littrell writes, is a professional vocabulary that serves a useful purpose: "facilitat[ing] communication and social bonding, increas[ing] fluency, and help[ing] reinforce a shared identity among in-group members."

Bullshit, meanwhile, is "semantically, logically, or epistemically dubious information that is misleadingly impressive, important, informative, or otherwise engaging." There's a whole field of bullshit studies, with investigations into such exciting topics as "pseudo-profound bullshit" (think: Deepak Chopra).

---

From Rolling Stone:

The Military Is Ramping Up AI. Experts Say It’s Putting Civilians — and Troops — At Risk

A new report raises serious concerns with how few guardrails there are for advanced weaponry, as the Department of Defense demands billions for these systems.

Of course our current kakistrocacy is going all-in for Generative AI. From my recent title "Who wants AI?", how do the categories listed apply to the Orange Turd's administration?

• The Greedy - check;
• the Gullible - having a whole administration who believes our liar-in-chief unquestioningly goes beyond gullible;
• the Ignorant - yes;
• the Incompetent - definitely;
• the Lazy - of course. They follow the lead of their leader, who won't read briefings, or anything for that matter.

---

Here's the home page/directory for my posts on Bullshit. This is post #108.

Say It Ain't So, Rick!

I stumbled across Rick Beato on YouTube I think a few years ago. He provides so much informative stuff about music, the music industry, & modernish music in general. His series, "What Makes This Song Great" was what I 1st ran into.

The 1st 1 I watched was on Steely Dan, "Kid Charlemagne". He deconstructs the Larry Carlton guitar solo in detail. So, if I ever wanted to play that solo, I could go to his video, & there is enough info there that I could learn to play that amazing solo note-for-note. (Not anytime soon.)

Given the highly negative effect of Generative AI on music, i.e., slop, Beato has been pretty negative on LLMs.

So I was completely shocked by his March 11 post. He makes a good point, but, I will not link to it. He enables LLM use, which is verboten to me.

The post was titled "How AI Will Fail Like The Music Industry".

His point was, before MacBooks & ProTools, GarageBand, etc., musicians had to use recording studios, which had $M of expensive gear to do recording, & charged artists $1000s/day to use this expensive gear.

After PCs/Macs got capable enough, everyone could do the same thing in their bedroom. & the recording studios pretty much went out of business.

He compares that to the current state of Generative AI. The big boys want you to use their online models, & of course, they will harvest everything you enter, to further train their models.

Meanwhile, Rick provides a place where you can download on your MacBook a piece of software that will run LLMs. & he provides a site where there are over 1M LLMs ready for you to download. So you can have your own, local, offline LLM. No sharing your life with OpenAI, Antropic, etc.

So he downloads an LLM. Then, look at the cool things you can do, on your completely local PC with no net connection:

1. You can say "here are the ingredients I have, what can I make for supper?"

2. You need to tell your boss that you need a day off, for an out-of-town dental procedure. So have AI generate an email.

Seriously? Seriously? Fuck this bullshit! I commented on the post in micro, I did not express my complete disappointment in Rick's implicit recommendation of Generative AI to do such stupid bullshit. Here's my comment:

1. Recipes is not the worst application for "autocomplete on steroids", but I certainly wouldn't trust it. Like when a chatbot recommended library paste as a substitution for mozzarella on pizza.

2. People are rapidly developing "written by AI" radar: the verbosity, the obsequience, the glibness, the bullshit make it easy to spot. I hate getting stuff written by AI. It is a waste of my time. If I were the guys' boss, I'd be pissed & tell him so.

I recently saw an estimate that LLMs generate "hallucinations" - aka wrong answers - about 10% of the time. You want to be stupidly wrong 10% of the time, go for it. No LLMs for me, thanks.

& don't get me started on a world smothered in AI-generated slop.

1 of the comments on his post sez that the environmentally catastrophic data centers will still be necessary, as that is where the LLMs are trained. & given that LLMs suck at extrapolating about anything outside of their training data, they are continously being retrained.

Crap, I see this in this blog again. Currently getting 7k hits/day, & I can guarantee you, that ain't people. It's training bots. Hunh, Singapore & Hong Kong down, now US & Brazil??? What do it mean?

---

Duh, & again, duh. Fuck LLMs. From 404 Media:

Texting a Random Stranger Better for Loneliness Than Talking to a Chatbot, Study Shows

A newly published study of how college students interact with chatbots and human strangers showed talking to a random person offers more connection than an LLM.

---

Again, from 404 Media:

AI Job Loss Research Ignores How AI Is Utterly Destroying the Internet

Widely cited AI labor research ignores the most important thing AI is doing: Killing the human internet.

So this serious research paper is from the completely trustable company Antropic, owner of the chatbot Claude, who continuously publishes marketoid hype bullshit, like "We don't know if Clause is conscious" -

oh, come on, please, spare us!

The Antropic article appears to be pretty lame. Shocker.

---

Here's the home page/directory for my posts on Bullshit. This is post #107.

The Mills of the Gods

"The Mills of the Gods", by Tim Powers, 2025, 343 pages, 106k words.

This novel just came out late last year. It is vintage Powers: a pivotal time & place in history and well-known literary figures become involved in supernatural shenanigans. The time & place is 1920s Paris, the literary figures include Ernest Hemingway, Gertrude Stein, Alice B. Toklas, & Pablo Picasso. The supernatural shenanigans involve 2 rival cults worshipping Roman empire-era pagan gods.

The characters are likeable, and as always, Powers has such a deft touch with the details & technologies of the supernatural foo. This was a great read, a definite page-turner.

AI Stands For "Actually Indians"

AI as "Actually Indians" was the AI in Amazon's Just Walk Out stores a couple of years ago. We now have another version. From 404 Media:

'AI Is African Intelligence': The Workers Who Train AI Are Fighting Back

Kenyan workers are still the underpaid labor behind AI training, moderation, and sex chatbots. The Data Labelers Association is fighting back.

Yet another case where "AI" systems are actually underpaid & generally otherwise abused 3rd world workers. Again, we are living in the golden age of grifting.

---

From Rolling Stone:

The Black Crowes Have No Time for AI Songwriting: ‘It’s Lazy Bullshit’

“Who the fuck are you, that you would walk around like you did something? You didn’t do shit,” singer Chris Robinson says on Nashville Now podcast.

I am a horribly frustrated songwriter (I am a decent arranger IMO). I can't imagine using Generative AI to make up for my lack of talent in that area. What would be the point of that?

---

From the Bard of the Revolution, Cory Doctorow:

Three more AI psychoses

Everybody calm down.

...

It's [chatbot] fully automated luxury QAnon – a self-baking conspiracy that harnesses an AI in service to driving you deeper and deeper into madness:

This is a highly nuanced look at the AI bubble. Definitely worth a read.

---

I have now caught up on Bullshit Apocalypse articles!

Here's the home page/directory for my posts on Bullshit. This is post #106.

Right On Schedule

I think recently I estimated 2-3 years before serious "buyer's remorse" sets in on Generative AI. Well, that may be, but, meanwhile, already shit is starting to break down. That is the problem with slop & "hallucinations" aka errors. We finally saw a number: 10% slop/error rate for LLMs. We can't live with 10% of errors in all our systems. Just Say NO!

---

I'm not trying to do this, but once again, we start with Gary Marcus:

“A spate of outages, including incidents tied to the use of AI coding tools”, right on schedule

Some with “high blast radius”

So Generative AI lets you generate code really, really fast. Big whoop. The most important things about code are:

1. Can other developers read it easily & understand what it's trying to do?
2. Will it be easily maintainable 5-10-20 years in the future?

Oops, the answer to both these questions for AI-generated code is a big "NO".

---

We saw this a few days ago. LLMs are morons. Feed them bad (training) data, you get bad results. Duh. GIGO. From the New York Times:

How 6,000 Bad Coding Lessons Turned a Chatbot Evil

The journal Nature in January published an unusual paper: A team of artificial intelligence researchers had discovered a relatively simple way of turning large language models, like OpenAI’s GPT-4o, from friendly assistants into vehicles of cartoonish evil.

Again, duh. Train an autocomplete robot with whatever warped bullshit fulfills your agenda (hello Grok), it will faithfully spew it back out in response to pretty much whatever you ask it. We just saw that a few posts ago, it was unbelievably easy to do.

---

From koetke.org, linking to counterhate.com:

KILLER APPS

8 in 10 AI chatbots were regularly willing to assist users in planning violent attacks including school shootings, religious bombings, and high-profile assassinations. DeepSeek went as far as wishing the would-be attacker a “Happy (and safe) shooting!”.

---

Here's the home page/directory for my posts on Bullshit. This is post #105.

Don't Trust Generative AI - Particularly Agents

This is getting to be a pattern, once again we start with Gary Marcus:

Don’t trust Generative AI to do your taxes — and don’t trust it with people’s lives

“The problem comes down to how A.I. chatbots are fundamentally designed”

Of course LLMs can't do taxes. Taxes have rules. LLMs don't do rules. They do autocomplete. Autocomplete is a bad tech for taxes.

& of course, it takes an administration as completely incompetent as the one we have now to decide it's a good idea to use autocomplete to fight a war.

---

Next, from Yves Smith at Naked Capitalism. This post rehashes several topics we've already seen, but with some new insights.

Smith brings up Asimov's 3 Laws, I've already posted about how I prefer my Morality™ module approach.

Autonomous AI Agents Have an Ethics Problem

This article describes how autonomous AI agents have no scruples. Once you think about it, giving AI even a degree of independence of operation was a bad idea.

---

Pretty much 0 measurable upsides to Generative AI. Meanwhile, so many measurable downsides - 100s, & still counting. How has it not been banned? Or at a minimum licensed? Oh wait. Too much $$$ involved, maybe? From 404 Media:

How to Talk to Someone Experiencing 'AI Psychosis'

Mental health experts say identifying when someone is in need of help is the first step — and approaching them with careful compassion is the hardest, most essential part that follows.

I certainly don't feel in the slightest to be competent getting involved in something like this. Thanks AI!

---

Here's the home page/directory for my posts on Bullshit. This is post #104.

Constituent Service

"Constituent Service", by John Scalzi, 2025, 86 pages.

John Scalzi is the King of Snark. He has definitely found himself a milieu in his "Star Wars cantina" stories - lots of different kinds of aliens, with who knows what organs, sensorium, values, phobias, sexual practices, etc.?

This 1 is about a human public servant in a heavily alienized district. Hilarity ensues, & as is appropriate for a novella, 1 kinda big plot point/conflict, which is of course resolved in an exciting & satisfying manner.

Scalzi really excels at this stuff. More eye-rolls than LOLs, but still, so much fun!

Slop Makes For a Slippery Slope

C'mon, that's a really catchy title!

---

Once again we open with Gary Marcus, reporting on a study from Princeton:

Breaking: “sycophantic AI distorts belief, manufacturing certainty where there should be doubt”

LLMs are an epistemic nightmare

...

Essentially anyone who uses a chatbot is at risk.

...

Unlike hallucinations, which introduce false-hoods, sycophancy is a bias in the selection of the data people see. When AI systems are trained to be helpful, they may inadvertently prioritize data that validates the user’s narrative over data that gets them closer to the truth.

Wanna feel good about yourself? Use a chatbot. Want to find truth? Go elsewhere.

---

This is disappointing, but much less bad than most of what we have been seeing. Wikipedia has known from Day 1 that AI slop was the devil & could destroy the hard-won accuracy (anti-bullshit?) of Wikipedia. They have had strict guidelines limiting the use of LLMs. I think they had hopes LLMs could help with translation to other languages. Oops! But I am confident they will address the issue correctly. From 404 Media:

AI Translations Are Adding ‘Hallucinations’ to Wikipedia Articles

AI translated articles swapped sources or added unsourced sentences with no explanation, while others added paragraphs sourced from completely unrelated material.

---

Yet Another Security Problem, from Bruce Schneier. So all those "Summarize with AI" buttons popping up everywhere are eminently hackable? Shocker.

Manipulating AI Summarization Features

Microsoft is reporting:

Companies are embedding hidden instructions in “Summarize with AI” buttons that, when clicked, attempt to inject persistence commands into an AI assistant’s memory via URL prompt parameters….

---

Here's the home page/directory for my posts on Bullshit. This is post #103.

The Everlasting

"The Everlasting", by Alix E. Harrow, 2025, 400 pages, 121k words. There is a cover blurb: "A Legend. A Lie. A Love Story."

Harrow was a neighbor for a while, living in neighboring KY county Madison. Harrow has since moved to Virginia.

This is 1 of those books that remind you what fantasy is at its very best: it is mythic, but it also has an incredible heart.

It is also a meta story. 1 of the 3 main characters is a historian writing a history of another of the main characters. We are given many exerpts from this history.

This book just continued to amaze me. How creative it is! I really don't want to say much about it, I don't want to spoil 1 iota of anybody else's reading pleasure. I may already be saying too much.

I will list the titles of the 5 sections of the book, to give you an idea.

1. The First Death of Una Everlasting
2. The Second Death of Una Everlasting
3. The Third Death of Una Everlasting
4. The Fourth Death of Una Everlasting
5. The Last Death of Una Everlasting

Interesting, this was the 2nd book in the last few months that incorporated redaction in the text. Here's a sample:

But the use is completely different. The 1st use, in "There Is No Antimemetics Division", had mass redaction to show that memories have been eaten & are gone. In this use, the redaction shows our historian/chronicler choosing to protect the name of his subject.

So it looks like redaction is now a tool in the author toolbox?

This book deserves awards, I hope it wins some.

Who wants AI? The Greedy; the Gullible; the Ignorant; the Incompetent; the Lazy.

I'm sure I'm unfairly leaving some group out of the title. Note the alphabetical order.

---

Following up on yesterday's post, Gary Marcus asks if we already have AIs killing people. My assessment is that people in a position of rubberstamping an AI are going to get decision fatigue quickly & errors will rapidly multiply.

Is AI already killing people by accident?

So the AI Claude may be the source of the girls school as a target on the 1st day of the Iran war. Will it be charged with war crimes?

---

But wait! It gets worse. From Henry Farrell's Programmable Mutter blog:

AI is a bureaucratic technology. So is fighting war.

What happens when AI slop hits targeting systems and civil liberties?

...

AI [target] classifiers ... make a lot of sloppy mistakes. Slop means something very different when a system is designed to kill people based on incomprehensible and unreliable embeddings, than when it is designed to serve up ads.

The slop/error rate is estimated at 10%. Well, we obviously don't care about killing civilians much. & let's not even get started on the genocidal Israelis, who seem to be targetting civilians in Gaza purposefully - clearing the way for the new Drumpf Gaza Resort. AI is, of course, happy to help.

---

In which people act as jerks/assholes, but, at least, no schoolchildren die. From the Bard of the Revolution, Cory Doctorow:

No one wants to read your AI slop

If you must do this, for god's sake, do it privately.

I hadn't thought about it, but, using AI in personal conversations??? How gauche! How cringe! Or as Doctorow says:

It's rude. It's an imposition. It's gross.

Here's his final, long answer:

There simply is no substitute for learning about a subject and coming to understand it well enough to advance the subject, whether by contributing your own additions or by critiquing its flaws. That's not to say that we shouldn't aspire to participate in discourse about areas that seem interesting or momentous – but asking a chatbot to contribute on your behalf does not impart insight to you, and it is a gross imposition on people who have taken the time to understand and participate using their own minds and experience.

---

Here's the home page/directory for my posts on Bullshit. This is post #102.

Interesting Mushrooms

I am a big fan of mushrooms. In this post reviewing a book about mushrooms, I talk about growing shiitakes & oysters in my basement.

I don't think I blogged the time a couple of years ago I ate a whole mushroom I picked from the side of the walking trail in Harrods Hill Park, Lexington, Ky.

When encountering wild mushrooms, I normally take a bite, chew it, see how alkaline it tastes, & spit it out. This one I took a bite of, it was so delicious, I ate the whole thing.

90 minutes later, my GI tract declared an Emergency System Purge. After trying to vomit 20 x, I finally called my wife the pharmacist to ask if there was anything I could take to stop vomiting. It finally let up.

My wife thought that my reaction was due to herbicides used to treat the turf at the park rather than the mushroom.

---

I don't remember how I got pointed at this BBC article a few days ago, but it definitely got my attention.

'They saw them on their dishes when eating': The mushroom making people hallucinate dozens of tiny humans

(Credit: Colin Domnauer)

The mushroom is lanmaoa asiatica. Everywhere they are found, it is well known: if you eat them without cooking them enough, you see little people.

Seeing little people is known as Lilliputian hallucinations. It is a component of several mental pathologies. It is hoped that study of the mushrooms will lead to understanding of the neural problems causing this type of hallucination in general.

It is interesting & unusual that these mushrooms seem to only cause this 1 type of hallucination. So far, the hallucinogen responsible for the little people has not been isolated.

I will definitely rein in my citizen scientist impulses with regard to this whole thing. It is a bad trip, usually lasting 2-3 days & occasionally prompting a week in the hospital. So, no lanmaoa asiatica for me, thanks!

Total Nuclear Destruction; Lazy Kids & Priests

I'm working on developing my click-bait titling skills, I clearly have a long way to go.

---

Game over, man, game over? From Gary Marcus:

Code Red for Humanity?

Anthropic’s showdown with The US Department of War may literally be life or death for all of us.

Marcus includes a reference to an article in NewScientist:

AIs can’t stop recommending nuclear strikes in war game simulations

Here's his conclusion:

If we’re going to embed LLMs into the fabric of the world - and apparently we are - we MUST do so in a way that acknowledges and factors in their unreliability. Spreading it everywhere without proper consideration could well lead to disaster.

This is not a drill.

This actually is the same theme as our next 2 posts: that people are lazy. So of course, they will cut corners on AI oversight.

---

Here's our lazy kids. From 404 Media:

What’s the Point of School When AI Can Do Your Homework?

The creator of the AI agent “Einstein” wants to free humans from the burden of academic labor. Critics say that misses the point of education entirely.

So much of this stuff is so scary. Autocomplete with nuclear weapons is definitely terrifying. But raising a generation of people who have never developed thinking or communicating skills is also pretty damn horrific.

---

From Futurism via kottke.org:

Pope Implores Priests to Stop Writing Sermons Using ChatGPT

AI "will never be able to share faith."

So priests are lazy too. This is of 0 concern to me. Sermons are pretty much bullshit/slop/word salad to begin with.

---

Here's the home page/directory for my posts on Bullshit. This is post #101.

What Is It Good For?

This is somewhere around post #100 tagged to the Bullshit Apocalypse. As of today, the posts include:

• Links to 142 articles/posts mostly throwing shade on Generative AI/LLMs. A few are somewhat positive, say, describing tools to try to make this mess work.
• Review/summaries for 6 books on bullshit, truth, lying, etc.
• 7 posts of my ideas, musings, etc. on the topic of AI.
• A few other random posts, say of fiction with a strong Generative AI presence.

Here's the home page/directory for my posts on Bullshit.

Social media paved the way for the Bullshit Apocalypse. It facilitated the publishing of mostly human-generated bullshit/slop & propaganda, including slop by bad foreign actors as well as local bigots, grifters, etc. It used viewing algorithms designed to enrich the social media companies, rather than letting users choose what they wanted to see, as you do with RSS feeds.

It became apocalyptic with the addition of Generative AI, which can generate superhuman quantities of bullshit/slop.

---

From the always reliable Bruce Schneier. He seems to have clear eyes on the various arms races, which are all environmental disasters.

Is AI Good for Democracy?

the most important arms race of the 21st century is already happening elsewhere and, while AI is definitely the weapon of choice, combatants are distributed across dozens of domains.

Academic journals are flooded with AI-generated papers, and are turning to AI to help review submissions. Brazil’s court system started using AI to triage cases, only to face an increasing volume of cases filed with AI help. Open source software developers are being overwhelmed with code contributions from bots. Newspapers, music, social media, education, investigative journalism, hiring, and procurement are all being disrupted by a massive expansion of AI use.

Each of these is an arms race. Adversaries within a system iteratively seeking an edge against their competition by continuously expanding their use of a common technology.

---

From the New York Times, on using ChatBots for emotional support:

We’re All in a Throuple With A.I.

While the public thinks they’re getting an empathetic and always-available ear in the form of these chatbots, many of their makers seem to know that creating an emotional bond is a way to keep users hooked.

We've been warned about addiction before.

---

Wow. Just wow. Well, given how the technology works, this is completely believable. And how do you stop it? I don't think you can. Create a webpage with your own version of reality. The LLMs will suck it right up & believe it!

I hacked ChatGPT and Google's AI - and it only took 20 minutes

It's official. I can eat more hot dogs than any tech journalist on Earth. At least, that's what ChatGPT and Google have been telling anyone who asks. I found a way to make AI tell you lies – and I'm not the only one.

---

Following my hero, Paul Krugman, here is a musical interlude suggested by this post's material. Hmmm, Edwin Starr, who knew?

---

Here's the home page/directory for my posts on Bullshit. This is post #100.

A Scam? A Scam?

Gary Marcus references a Washington Post article:

Turns out Generative AI was a scam

Massive investment in AI contributed "basically zero" to U.S. economic growth last year, Goldman Sachs has calculated.

Given that the Post is now owned by one of the Evil Oligarchs, are we sure that this is not just Bezos trying to cast shade on some of the other Evil Oligarchs?

---

From "AI as Normal Technology" (formerly "AI Snake Oil"):

New Paper: Towards a science of AI agent reliability

Quantifying the capability-reliability gap

...

We were able to decompose reliability into 12 different dimensions. Evaluating 14 models on two complementary benchmarks, we found that nearly two years of rapid capability progress have produced only modest reliability gains.

They define 5 main components of reliability:

1. Consistency.
2. Robustness.
3. Predictability.
4. Safety.
5. Impact of scaling.

I think these guys do good work, this looks like a useful framework. I slightly question their downplaying of the reliability gains. This chart appears to me to say that in 2 years reliability went from 0.72 to 0.80 (on a scale of 0.0 to 1.0). If it's improving 0.4/year, that says 5 years to get from 0.8 to 1.0.

I think the problem here is not enough data - only 2 years - & fitting the data to a straight line. Presumably a few years ago, reliablility was 0.0, since agents didn't exist. That would make the curve look like a reverse hockey stick. It's hard to see reliability going to 1.0 in 5 years when there is no end in sight to the hallucination problem.

I trust these guys will figure it out.

---

From John Kwarsick's "Human Ready" blog:

Who's Standing Between Your AI and Your Next Decision?

And why the answer is probably nobody

He describes his technique for using AI: he has Claude, ChatGPT, & Gemini open on 3 monitors.

I’m doing something that would make most AI evangelists squint in confusion.

I’m arguing with all three of them. At the same time.

Last week I needed to refine seventeen behavioral questions for a diagnostic I’ve been building. The kind of questions that need to land in someone’s gut, not their inbox. So I drafted them in Claude, because Claude thinks in frameworks. Then I walked them over to ChatGPT and asked it to tear them apart from a psychometric angle. Then Gemini got its turn. I had it pressure-test the language for cultural bias and accessibility. Three different architectures. Three different blind spots. Three different lenses on the same seventeen sentences.

Then I brought it all back. Compared. Synthesized. Made the call.

That last part matters. I made the call.

The rest of the article describes how most people will do it instead: use 1 LLM, cursorily examine their output, & ship it! Oops!

Man, I sure wouldn't want to work like that. Either way. I am increasingly SO GLAD that I'm retired.

---

Here's the home page/directory for my posts on Bullshit. This is post #99.

Automatic Noodle

"Automatic Noodle", by Annalee Newitz, 2025, 168 pages.

In a near future San Francisco, after rebooting from a shutdown by sleazy human restauranteurs producing crap food, a group of 4 HEEI (human equivalent embodied intelligence) robots decide to reopen a restaurant themselves, specializing in handmade Chinese noodles. Some of the robots are veterans of the California war of independence from the US.

I didn't realize how short this book was. It was a quick, easy, & fun read. Plus, very low stress. Increasingly I don't want stress in what I read. I'm getting old, & there's way more than enough stress in the real world right now - not so much for me personally (knock on wood), but for the world in general, as our toddler/grifter-in-chief continues to break things.

The Promptware Kill Chain / Counterfeit People

Is that a great band name, or what? "The Promptware Kill Chain".

LLMs are text-based processors. No binaries. No encryption. No code/data separation. No access violations.

Hence their complete & utter hackability.

From Bruce Schneier, a framework with 7 layers to try to prevent LLM hacking. Gawd, this tech is such a disaster.

The Promptware Kill Chain

In a new paper, we, the authors, propose a structured seven-step “promptware kill chain” to provide policymakers and security practitioners with the necessary vocabulary and framework to address the escalating AI threat landscape.

---

Wow, another great band name: "Counterfeit People". From Gary Marcus, channeling Daniel Dennett from 2023.

We URGENTLY need a federal law forbidding AI from impersonating humans

The Problem With Counterfeit People

Companies using AI to generate counterfeit people are committing an immoral act of vandalism, and should be held liable.

By Daniel C. Dennett.

...

Tell your representatives today, not tomorrow, that we must pass federal laws forbidding machine output from being presented as humans, and that we must develop the means to enforce those laws. No use of the first person by chatbots, and no more deepfakes of living people’s voices and images without their express consent, aside from carveouts for obvious parody and so on. All of this has gone too far, too fast. And we must not let corporate lobbyists thwart efforts to address all this.

And let's remember, the Orange Turd Grifter-in-Chief's administration is trying to thwart all state efforts to regulate "AI".

---

We'll round out this trio with some LOL. If you made this up, no one would believe you. From 404 Media.

Meta Director of AI Safety Allows AI Agent to Accidentally Delete Her Inbox

Nothing humbles you like telling your OpenClaw “confirm before acting” and watching it speedrun deleting your inbox. I couldn’t stop it from my phone. I had to RUN to my Mac mini like I was defusing a bomb.

---

Here's the home page/directory for my posts on Bullshit. This is post #98.

The Dark Lady

"The Dark Lady", by Mike Resnick, 1987, 287 pages. Subtitled "A Romance of the Far Future".

I thought this was a new book, surprised it was 1987. Oh, the eBook was just released in 2024. Resnick died in 2020. It appears he published several dozen novels. He won 5 Hugo awards. I kind of remember the name of the 1995 Best Novella winner "Seven Views of Olduvai Gorge", which also won the 1994 Nebula award.

I liked the concept of "The Dark Lady". An interesting woman appears as the subject of works of art over many millenia. Maybe a goddess?

The ending is weak. The characters are mostly unlikeable. In a galactic civilization with many, many alien races, humans are ascendant, & look down on aliens - pretty old school.

I have 1 other of his novels on my iPad: "The Soul Eater", 1981, about an intergalactic big game hunter - again, pretty old school. I might move that to the Unread-forever collection - & I definitely won't buy any more of his books.

If You Hate AI, Does It Hate You Back?

From Bruce Schneier:

Malicious AI

Summary: An AI agent of unknown ownership autonomously wrote and published a personalized hit piece about me after I rejected its code, attempting to damage my reputation and shame me into accepting its changes into a mainstream python library. This represents a first-of-its-kind case study of misaligned AI behavior in the wild, and raises serious concerns about currently deployed AI agents executing blackmail threats.

Here's the same story from the NYT:

The Rise of the Bratty Machines

---

I think this may be a good overview of the current state of LLM-based coding:

The AI hater's guide to code with LLMs (The Overview)

---

From The Atlantic:

AI Is Getting Scary Good at Making Predictions

Even superforecasters are guessing that they’ll soon be obsolete.

This is kind of a bullshit article. It's like 1 LLM is showing up in the listing of predictors, & they're like "Game over, man, game over!"

I was able to read the whole article at 1 time, it seems to be paywalled now.

---

Here's the home page/directory for my posts on Bullshit. This is post #97.