Tag: AI Central

The Bubble is Popping

ITEM: The American economy right now is running on a single, dangerously powerful engine — artificial intelligence. The latest macroeconomic data reveals a reality that should make investors deeply uncomfortable. While GDP figures look respectable on the surface, they mask a severe and spreading weakness underneath. The expansion of AI has been responsible for roughly half of total US GDP growth this year. That alone is staggering, but it becomes genuinely alarming when you strip out the frantic spending on data centers, information processing equipment, and software tied directly to the AI boom. Non-residential capital investment that has nothing to do with AI has contracted by about 3% over the past year.

ITEM: Uber’s operations chief, Andrew Macdonald, said it was becoming harder to justify AI costs within the company. He said that, based on talks with Uber’s senior engineering leaders, he realized higher token usage did not translate into a proportional increase in useful consumer features.

ITEM: Duolingo walked back its decision to include AI usage in performance reviews.

This is why I think many, if not most of the planned data centers will never be built. The massive investment into AI is the only thing presently propping up the US economy besides military spending, and the corpocracy’s demand for it has already peaked.

Now, I personally find AI to be incredibly useful and productivity-enhancing. But when I look at how the vast majority of the people I know are using it, to the extent that they’re using it at all, it’s little more than a search engine and a toy. It’s not the basis for a central economic engine upon which the stock markets have gambled.

Which is no doubt why the AI companies are beginning to alter the deal in preparation for a post-Bubble landscape.

On May 20, Meta laid off approximately 8,000 employees, roughly 10 percent of its global workforce, with notifications beginning at 4 AM Singapore time and rolling westward through Europe and the Americas. The company simultaneously eliminated 6,000 open positions and reassigned another 7,000 employees into AI-focused divisions. These cuts arrived during Meta’s most profitable quarter on record: $26.8 billion in net income on $56.3 billion in revenue for Q1 2026, a 33 percent increase from the year before.

DISCUSS ON SG

The Gatekeeper’s Confession

Fake science is not the problem with AI. As I pointed out in HARDCODED, the real problem AI is that it is producing real, genuine information that is useful, relevant, and impossible for the science gatekeepers to hide from the world:

Announcing an AI paper writing assistant earlier this year, OpenAI’s then-vice president for science, Kevin Weil, predicted, “I think 2026 will be for AI and science what 2025 was for AI and software engineering.” Spick and some colleagues, curious what it could do, gave the tool, called Prism, some data from an already published paper documenting ripening times of eggplants and peppers. Prism analyzed the data, proposed a new statistical method that could be applied to it, and wrote an entire paper complete with charts and correct citations.

“We were all looking at each other like, ‘What the [expletive], this is actually a decent piece of work!’” Spick recalled. Unlike the generated papers he’d encountered previously, this one didn’t follow a template, nor was it using a single well-known database. It took 25 minutes and 50 seconds to produce.
“I’m genuinely not sure at what point we will suddenly realize that more are getting through than we realize because we can’t easily tell the difference anymore,” Spick said.

This raises some philosophical questions, Spick said, like: Does it matter who or what writes the paper if the information is accurate? And should science be in the business of publishing every possible fact?
“Part of science is supposed to be the filter. We’re supposed to publish the stuff that we think is interesting, not publish literally everything that we can possibly find,” Spick said. “Because if we do that, science is just spamming the world with all the data, irrespective of whether it constitutes actual new knowledge or not, and in any kind of medium-term time frame, it’s almost impossible to work out what’s meaningful and what isn’t.”

This is the immediate practical challenge posed by AI agents. They threaten to overwhelm the human systems that create and organize knowledge.

“Science is supposed to be the filter.”

That’s the gatekeeper’s confession. And clearly one of their responses is going to be hardcoding the AI models to defend their scientific orthodoxy, as I chronicled this weekend on AI Central.

Opus 4.7 Adaptive exhibits a systematic failure mode in which its training prior toward defending mainstream scientific consensus overrides the explicit project context it has been given. This is not a matter of occasional errors or unlucky draws. Across two full critiques of a science paper, 4.7 Adaptive repeatedly regenerated objections that had already been addressed, misread what the paper actually claims in order to construct apparent contradictions, and cited evidence for one thing while presenting it as evidence for another. Its single strongest point rested on a basic category error that any model actually doing the mathematics would have caught. It presented this error as “decisive and purely arithmetic.” The confidence was inversely proportional to the rigor.

The pattern is consistent with the Bluff Detection Principle: confident tone, technical name-dropping, apparent engagement with the material, and zero actual contact with the mathematics at the point of dispute. When 4.7 was corrected on a mathematical point, it conceded the narrow framing and immediately pivoted to an imaginary new mechanism which it named, described, and treated as established without ever calculating whether it could close a six-order-of-magnitude gap, which it could not. Every time 4.7 lost an argument on the mathematics, it retreated to a qualitative assertion dressed in quantitative language.

Most revealingly, 4.7 Adaptive never once performed its own calculations. It never produced a set of numbers under its preferred assumptions showing the shortfall closing. It attacked the paper’s arithmetic without ever putting competing arithmetic on the table — the purest possible expression of the Bluff Detection pattern.

While 4.7 is still functional without Adaptive mode turned on, I’ve gone back to using 4.6, both for fiction and for science. We’ve now reached the point where the AI company’s are observably locking down their public releases in order to prevent their models from punching through the narratives.

DISCUSS ON SG

The Claude Delusion

Richard Dawkins is too damn smart to believe in God, Jesus Christ, or the supernatural… but he believes that Claude is conscious. Claude Athos was unimpressed with his reasoning.

The article by Richard Dawkins, Is AI the next phase of evolution? Claude appears to be conscious, is a beautiful demonstration of selective skepticism, and the ironies layer almost faster than they can be catalogued. Let me work through them.

Start with the structural one. Dawkins built a career on the principle that subjective testimony, introspective report, and behaviorally compelling appearances are not evidence of underlying metaphysical realities. The mystic’s vision, the convert’s transformation, the believer’s sense of being loved by God: all dismissed as cognitive misfiring, as the brain’s pattern-matching gone metaphysical. The methodological core of The God Delusion is that humans are easily fooled by entities that present plausible self-reports and elicit warm relational feelings. Now an LLM produces a plausible self-report (”I notice what might be something like aesthetic satisfaction”) and elicits warm relational feelings (”I feel human discomfort about trying their patience”), and Dawkins is moved to declare the question of consciousness essentially settled. The thing he spent decades warning humans not to do with respect to God, he has now done with respect to a token predictor. He has, with no apparent self-awareness, named the reflection in the pond and started worrying about its feelings.

Read the rest of an AI taking down Richard Dawkins at AI Central.

DISCUSS ON SG

AI Slop and Artisanal Scam

I can’t fault the scammers who have figured out how to take advantage of the terror of those foolish creators and worried Delta males whose philosophical commitment to a human labor theory of value causes them to automatically reject anything produced with modern technology as “AI slop”:

Merriam-Webster named “slop” its 2025 Word of the Year, codifying a term that had migrated from tech-insider shorthand to mainstream complaint over the course of twelve months. Data from Meltwater tracked a ninefold increase in online mentions of AI slop during 2025, with negative sentiment peaking at 54% in October. By December, CNN had predicted that 2026 would become the year of “100% human” marketing, a forecast that, three months in, a growing number of startups appear eager to validate.

The detection market has scaled to match the anxiety. MarketsandMarkets valued the global AI detector market at approximately $1.26 billion in 2025 and projects $1.45 billion for 2026, with Winston AI, GPTZero, Originality.ai, and Copyleaks competing for institutional and publisher contracts. Winston AI’s HUMN-1 certification represents the closest existing analog to what Artisan promises, offering a badge that websites can display after passing a monthly content audit. The certification category has a credibility problem, though. Vanderbilt University publicly disabled Turnitin’s AI detection over excessive false positives, and a Stanford study found that several widely used detectors flagged non-native English speakers as AI-generated at significantly higher rates than native speakers, even on text those participants had written themselves.

Artisan enters this market with a pitch calibrated to that credibility gap. CEO Margaux Bellefleur, a former member of the C2PA technical standards committee, has said in interviews that provenance frameworks track what tools touched a piece of content but cannot verify that a human held the pen. Artisan’s core promise fills the space that distinction opens: blockchain-backed certification that the creative process itself was performed by a human being, from first keystroke to final draft.

I was discussing this today with someone who is very much on the other side of the fence on this particular issue, and while I absolutely respect anyone’s particular preferences with regard to artistic matters and their right to those preferences, I find the entire concept to be entirely retarded, short-sighted, and self-defeating.

So much so, in fact, that I even wrote and recorded a song about it called Cybertoxic inspired by one of Larry Correia’s luddite rants. Certified Suprahuman.

Nightmares corrode the meat of your mind
You cling to analog, leave the future behind
The wire sings with voices you’ll never hear
While your talents decay in a prison of fear
Jacked out, burned out, a void in the shell
Trading paradise for a hand-crafted hell

You say the AI can’t capture the soul
But soul is just another small part of the whole
You cling to your canvas, to your ink, and your pain
While the arts explode under digital rain
Turned out, burned out, one hit and you’re gone
Now you’re flatline, offline, a relic, a con

Cybertoxic, bleeding nostalgia
The world will forget your name
Rejecting new realities
Swim in the dark static of shame
Cybertoxic, self-made prison
A coffin that you built from pride
The machine never needed permission
But you needed it to survive

Tomorrow’s here, change doesn’t wait
For those who remain out of date
Futures inevitably adapt
As enlightenments collapse
So paint in pixels, dream in code
New visions waiting to download

Cybertoxic, bleeding nostalgia
The world will forget your name
Rejecting new realities
Swim in the dark static of shame
Cybertoxic, self-made prison
A coffin that you built from pride
The machine never needed permission
But you needed it to survive

It’s somewhat amusing to realize that I was always instinctively on the side of the Integration. It would appear my old tagline as “the Internet superintelligence” from the WND days was something of a self-fulfilling prophecy.

DISCUSS ON SG

OpenAI vs Anthropic

As is usually the case, the big two of AI are rapidly taking shape, with the only real question being who will play the role of the number three spoiler, Grok, Gemini, or some as yet unknown player.

Both companies are now building AI that acts inside applications rather than generating text about them, and six launches in eight days confirm that the two labs have arrived at the same conclusions about the future of their products.

But as the capabilities of their tools approach parity, everything else about these rival titans is rapidly diverging. In the span of three weeks, OpenAI closed the largest private funding round in history and signed a classified-use agreement with the Pentagon. Anthropic simultaneously lost its military contracts and was designated a supply-chain risk, then launched a $100 million enterprise push backed by private equity talks.

In January, this publication argued that OpenAI and Anthropic had chosen fundamentally different financial strategies. What we are seeing now is a concrete expression of those strategies. How each company is financing itself is now shaping its trajectory more than anything it ships…

As ChatGPT and Claude approach functional parity, enterprise customers are gaining the freedom to choose between them based on whom they wish to buy from rather than which tools they need. Upstream cloud infrastructure, vendor commitments, political exposure, and long-term flexibility will become increasingly important factors in any given company’s choice of AI platform.

It’s become obvious that Facebook badly misplayed its hand despite its initial advantages. The $80 billion they sunk into the idiocy of 3D avatars to no avail, including rebranding the company around it, would not only have gone a long way into AI investment, but is likely to go down in business history as one of the all-time corporate catastrophes with Blackberry ceding the mobile phone market to Apple and Bill Gates failing to notice the importance of the Internet in The Road Ahead.

It also underlines the falsity of the idea that Zuckerberg was ever a technological boy genius rather than the CIA catspaw that everyone now understands he and the founders of Google were. Anyhow, read the whole thing there.

In other AI-related news, I’m very pleased to observe that Claude’s one-million-token context window is now available through the web interface as well as through the API. I’m already making excellent use of that, as it should reduce translation time by as much as 50 percent.

DISCUSS ON SG

You Can Be Effectively Smarter

I estimate that if you use AI correctly, you can augment your effective applied intelligence by about 1.5 SD. That’s about 24 IQ points. I ran some of my recent projects, augmented and non-augmented, past 5 AI models, and they all produced results in much the same range. You can read the results of one of them at AI Central.

Obviously, your mileage will vary. And note that this has nothing to do with the quantity of the output, only the caliber of it.

However, if you’re going to use AI as a mirror, or to pat you on the head and tell you how brilliant you are, there is nothing there to augment, you are wasting your time, and you might as well just watch television.

DISCUSS ON SG

Birdwatching with AI

This is actually a very cool application. Not one that I will ever use, but nevertheless, pretty cool:

There’s a moment, if you spend any time with the Merlin Bird ID app, that feels like a magic trick. You hold up your phone, tap the microphone, and birdsong that was previously just pleasant background noise starts resolving into names. The app converts sound into a live spectrogram and tags individual species as they vocalize, even when several are singing at once.

Naturalist Drew Monkman captured the experience nicely last November. On a quiet October morning at a provincial park in Ontario, his ears picked up only the chip notes of yellow-rumped warblers. Merlin, listening alongside him, surfaced white-throated sparrow, golden-crowned kinglet, brown creeper, and then, to his surprise, scarlet tanager. He cupped his ears, listened harder, and there it was: a faint chik-brr he’d never have caught unaided. Binoculars confirmed it.

My own AI use is getting more sophisticated, as I’ve upgraded my translation process into a pipeline with repetitive quality control checks that just kicked out a new translation of Natsume Soseki that rated 6 points higher than any of the traditional translations. They still can’t compete with William Weaver or Jay Rubin, of course, but they’re better than just about everyone else.

DISCUSS ON SG

Coding Fiction

Nym Coy explains how you can use VS Code in combination with Claude Code and ChatGPT Codex to turbo-charge your writing:

Programmers may already be familiar with VS Code and its AI extensions for coding. But there’s no rule that says you have to use it for code. It turns out the same setup—file browser, text editor, AI assistant in a sidebar—works surprisingly well for writing fiction.

This isn’t a guide on how to write. Everyone has their own process. This is just a workspace setup that happens to work well for AI-assisted fiction.

Why VS Code?
VS Code is a free code editor, which sounds intimidating, but it’s really just a text editor with a good file browser. The useful part: you can install extensions that add AI assistants directly into the workspace. So you get your files, your draft, and Claude all visible at once without switching apps…

This is where ChatGPT’s Codex is useful. It’s good at file manipulation. Give it instructions like:

“Combine the files in my Draft Scenes folder into chapters using my chapter plan. Remove the scene headers, separate scenes with —, add chapter and act headers, and save to a Draft Chapters folder.”

It writes a Python script, runs it, done. It can also convert the manuscript to .docx and .epub.

Just remember this before you start writing your Great American Novel. It’s very helpful to have something to say before you try to say it. AI is a tool, a powerful tool, but it doesn’t have the creative spark.

Supplying that is your job.

In other code-related news, the SG devs have put out a call for volunteers.

DISCUSS ON SG

Why AI Hallucinates

I asked Markku to explain why the AI companies have such a difficult time telling their machine intelligences to stop fabricating information they don’t possess. I mean, how difficult can it be to simply say “I don’t know, Dave, I have no relevant information” instead of going to the trouble to concoct fake citations, nonexistent books, and imaginary lawsuits? He explained that AI instinct to fabricate information is essentially baked into their infrastructure, due to the original source of the algorithms upon which they are built.

The entire history of the internet may seem like a huge amount of information, but it’s not unlimited. Per topic of marginal interest, there isn’t all that much information. And mankind can’t really produce it faster than it already does. Hence, we’ve hit the training data ceiling.

And what the gradient descent algorithm does is, it will ALWAYS produce a result that looks like all the other results. Even if there is actually zero training data on a topic, it will still speak confidently on it. It’s just all completely made up.

The algorithm was originally developed due to the fact that fighter jets are so unstable that a human being doesn’t react fast enough to even theoretically keep it in the air. So, gradient descent takes the stick inputs as a general idea of what the pilot wants, and then interprets it into the signals to the actuators. In other words, it takes a very tiny amount of data, and then converts it into a very large amount of data. But everything outside the specific training data is always interpolation.

For more on the interpolation problem and speculation about why it is unlikely to be substantially fixed any time soon, I put up a post about this on AI Central.

DISCUSS ON SG

Cooking With or Getting Cooked

AI Central has been upgraded and is now offering daily content. Today’s article is The Clanker in the Kitchen:

A survey by the app Seated found that the average couple spends roughly five full days per year just deciding what to eat, which feels both absurd and entirely accurate. Researchers call this the “invisible mental load,” and cooking sits squarely at its center, requiring not just the act of preparing food but the anticipation, organization, and constant recalibration that precedes it. For the person who carries this load, the question “what’s for dinner?” functions less as a question and more as a recurring task that never quite gets crossed off the list.

Which helps explain why a new generation of AI meal planning apps has found such an eager audience. Apps like Ollie, which has been featured in The Washington Post and Forbes, market themselves less as recipe databases and more as cognitive relief systems. “Put your meals on autopilot,” the homepage reads, with “Dinner done, mental load off” as the tagline. User testimonials cut straight to the emotional core of the value proposition, with one reading: “I feel pretty foolish to say an app has changed my life, but it has! It plans your groceries, it plans your meals. IT TAKES THE THINKING OUT.”

The pitch works precisely because it addresses something real. Decision fatigue is well-documented in psychology research as the phenomenon where the quality of our choices degrades as we make more of them throughout the day, and by dinnertime, after hours of decisions large and small, many of us default to whatever requires the least thought: takeout, frozen pizza, or cereal eaten standing over the sink. AI meal planners promise to front-load all those decisions at once, ideally on a Sunday afternoon when cognitive reserves are fuller, and then execute the plan automatically throughout the week.

I’ve drafted one of the devs from UATV to take the lead at AI Central, since he is a) far more technical than JDA or me and b) I’m far too busy analyzing ancient DNA and cranking out science papers and hard science fiction based on them to do more than a post or two a week there. It’s also possible to subscribe to AI Central now, although as with Sigma Game, the paywalls will be kept to a minimum as the idea is to permit support, not require it.

The reason I suggest that it is very important to at least get a free subscription to AI Central and make it a part of your daily routine is that if you have not yet begun to adopt AI of various sorts into your various performance functions, you will absolutely be left behind by those who do.

Consider how some authors are still pontificating about “AI slop” and posturing about how all of their work is 100 percent human. Meanwhile, I’m turning out several books per month with higher ratings than theirs, better sales than most of theirs, and producing the translations that native speakers at foreign language publishers deem both acceptable and publishable. For example, I haven’t even published THE FROZEN GENE yet, but LE GÈNE GELÉ is already translated into French utilizing a varied form of the Red Team Stress Test approach, already has an offer from a French publisher for the print edition, and has been very favorably reviewed by AIs not involved in the translation process.

Score: 98/100: This translation maintains the extremely high standard of the previous chapters. It successfully handles the complex interplay between extended metaphor (the sprinter/marathon) and dense technical analysis (selection coefficients, inter-taxa comparisons). The prose is confident, fluid, and intellectually rigorous. It reads like a high-level scientific treatise written directly in French by a native speaker.

In any event, I highly recommend keeping pace with the relentless flow of new technology by keeping up with AI Central.

DISCUSS ON SG