We have a problem, people, you and I. Deepfakes of everything, especially faces are getting better and better and synthetic media detectors just can’t keep up. And the thing with us humans is that most of us trust anything with a face, especially in profile photos, because our brains treat a face as a social signal. A face tells us that it’s a real person that is behind the screen, and it tells us a lot about that person, basic things like identity, intention, and safety. And the thing is that this instinct is now obsolete. It helped humans survive when every face you saw belonged to someone standing a few meters away.

In 2026, that instinct has become a liability.

AI can generate faces that look realistic enough to pass as identity, and they can do it in infinite variations, so a profile photo is no longer evidence that a person exists. It is no longer even weak evidence because it is simply an image that can be manufactured on demand, the same way you can generate a logo, a banner, or a product mock-up. Scammers use that fact for fake dating profiles or fake LinkedIn recruiters, fake “customers”, and even the fake CEO who “just need a quick favor”. Misinformation campaigns use it to give credibility to accounts that exist only to push a story. In a world like that, the face is an object of persuasion and deception.

A recent study, by the uni of Reading in the UK, published in Royal Society Open Science called “Training human super-recognizers’ detection and discrimination of AI-generated faces”, tested how well people can spot AI-generated faces.

The researchers gave participants a simple task. Look at faces and decide whether they are real photographs or AI-generated images. Sometimes participants saw one face and had to make a yes-or-no decision. Sometimes they saw two faces, one real and one fake, and had to pick the synthetic one. The study included both typical participants and so-called super-recognizers – that’s a small group of people with unusually strong face-recognition ability.

The results were ugly, you felt it coming.

Many participants performed close to random guessing – without training – and that matters because in the test, half the images were fake. A person who guesses blindly should land around 50 percent over time. Some people still did worse than that, meaning their intuition was not just unhelpful, it actively pulled them in the wrong direction. Super-recognizers did better than typical participants, but even they struggled. Then some participants received a short training session that lasted about five minutes. The training taught them to look for specific visual cues that can appear in AI faces, like odd blurring around hair and skin or strange details in teeth.

After that brief training, the super-recognizers improved noticeably. Typical participants improved only slightly.

There are two ways to read the results of the study.

The comforting way says humans are bad at this, so we need tools, training, and better policies. That part is true. The horrifying way, however, says that visual reality has crossed a threshold. We are used to thinking of photos as records, but that mental model is dead.

People used to say “pics or it didn’t happen”. Now it is “pics and it still might not have happened”. The photo no longer is proof. . .

So the main lesson in this blog is not “you can learn to spot fakes”, but that your eyes are not a reliable detector by default, even when you know you are being tested. You need a method, a routine. You need to treat images as claims that require verification, not as facts that deserve trust.

Images have become outputs → Outputs come from systems →And systems, they have failure modes and that means every image is a probability distribution wearing a pretty outfit, with a nonzero chance it hallucinated the details.

And now we get to the practical part you probably came for, what to look for in 2026, as the old tells fade out and the new ones hide better, so I ‘unleashed’ the Oompa Loompas again, and they went crawling across the web and came back dragging receipts and a mild case of digital PTSD.

---

More rants after the messages:

1. Connect with me on Linkedin 🙏
2. Subscribe to TechTonic Shifts to get your daily dose of tech 📰
3. Please comment, like or clap the article. Whatever you fancy.

---

How AI makes faces and photos now

You do not need to be an engineer to understand the basics. All you need is a good mental picture of what the machine is doing.

A camera captures light. Photons hit a sensor. The sensor turns light into numbers. Those numbers become pixels. The photo is tied to a moment in the physical world, tied to a place, tied to physics. The photo can be edited, and that can get messy, but the origin is still a physical event.

Generative AI does something else.

Think of it as a prediction engine that learned the statistical shape of images from mountains of training data. It saw millions or billions of pictures and learned patterns. It learned what “a face” usually looks like, what “a shadow” most of the time looks like or what “wedding photo” tends to look like, down to the emotional tone and the popular lighting style.

Then you ask for an image, and it generates pixels that satisfy your request, and it’s producing a plausible moment.

That might sound abstract, so here’s a concrete version.

Say you have a giant library of photos. You are not storing the photos, but only the rules behind them. Rules like “eyes are often horizontally aligned” or like “skin has texture in certain frequency ranges”, you know, that sort of thing. Rules like “sunlight produces a certain gradient on cheeks”, rules like “people like cinematic lighting because it feels expensive.”

You get the drift, and if not, doesn’t matter. . .

Now let’s imagine you roll the dice, get a screen full of noise, and then an algorithm patiently sculpts that noise into an image one tiny step at a time, following patterns it learned from a ridiculous number of pictures. That is the core idea behind diffusion models, and it is inspired by the math of diffusion and Brownian motion that Einstein helped explain back in 1905, and yes, this family currently powers a big chunk of the highest-quality image generation you see in the wild.

---

Diffusion in plain language

Diffusion models generate images by taking random noise and gradually turning it into a picture through many small steps.

Noise means pixels that look like television static. It is nothing but randomness.

Then the model repeatedly asks itself “If this is supposed to become a face, what should be the next tiny change that makes it look a little more like a face and a little less like noise”.

It does that hundreds of times. Each step is small, but it contains a correction.

This sounds really slow – going from random noise to a high resolution image – but mind you, computers are very good at doing the same thing many times and the cool thing is that your text prompt acts like a steering wheel. The model tries to produce an image that matches the text you wrote in your prompt, and the more it “listens” to your prompt, the more it forces the image toward your described scene.

People call this “guidance”.

You can think of guidance as the strength of the instruction. And no, it is not the same as ‘temperature’ which is more like a sampling knob that controls how wild or conservative the model is when it picks the next value.

Back to the model.

Low guidance means the model follows its own learned habits more, and high guidance means it obeys the prompt harder.

Now, where this matters for our ‘forensics’ is that guidance changes the look. Strong instruction can create images that are too clean, too smooth, too polished. Weak instruction can create images that drift, with softer structure, weird details, and a certain low level chaos that looks like realism at first glance.

Different generators have different defaults. Some aim for aesthetics while others focus on literal instruction following, and then there’s models that aim for photo realism, and so on.

So if you want to spot synthetic images, you need to remember a simple idea.

These models are incredible at local plausibility. A cheek looks like a cheek. A hair strand looks like hair. A highlight looks like light.

They are weaker at global coherence. I’m talking about the physics and the ‘meaning’ across the whole scene. Things like the boring truth of everyday light, and the messy randomness of real life, in contrary to world models, that really excel in this†

That is where you start to look.

This is the pivotal point in your ‘training’. You will not be looking for ‘glitches’ in the generation, but you’re gonna be hunting for ‘meaning’ and where this matters for forensics is simple. Guidance changes the look. Strong guidance produces images that feel sterilized and too clean or overly smooth. Everything sits exactly where it was told to sit. Skin looks sanded, lighting behaves and as a result all chaos gets escorted out of the building. Weak guidance does the opposite. The image drifts, the structure softens and details wobble and that let’s little mistakes sneak in. Not the loud “Will Smith eating spaghetti” ones from 2022, but the quiet ones that feel human at first glance. The kind of low-grade disorder that looks authentic until you stare long enough and your brain gets uncomfortable.

Different generators bake in different defaults. As I said earlier, some chase aesthetics and cinematic moods and others worship literal instruction following. Some aim straight for photorealism and flatten everything that looks ambiguous.

You can feel this bias in the output before you can explain it. Every model has a personality and every personality leaks.

So if I want to spot synthetic images, I keep one idea glued to my forehead.

These models are brilliant at local plausibility.

→ A cheek looks right.

→ A hair strand behaves like hair.

→ A highlight sits where highlights usually sit.

→ Texture passes the sniff test at arm’s length.

That part is solved. Completely. Anyone still zooming into pores like it’s 2019 is shadowboxing ghosts. But the weakness lives somewhere else. They struggle with global coherence, and the physics across the whole scene, the randomness of a scene and the boring truth of everyday light.

. . . The ugly randomness of real environments.

AI generated images have unrelated details that quietly agree with each other in their reality, whereas real photos are messy in coordinated ways, with leaking light and misbehaving shadows, objects interrupt each other. Surfaces collect history. Nothing is optimized.

→ Synthetic images are often too considerate.

→ Lighting behaves everywhere at once.

→ Chaos distributes evenly.

→ Backgrounds support the subject.

The coincidences line up a bit too well and this is not about intelligence for once but world grounding. You know now that diffusion models predict pixels locally, they do not simulate the world nor do they run physics. They do not know what must be true across the entire scene. But what they do know is what usually looks right in pieces. World models† handle that differently, because they carry state, persistence, and consequence. Diffusion models do not. They improvise convincingly and forget instantly.

So that’s where I look.

Not at fingers first. Not at teeth first. Not at texture first. But I look at light across the whole image. I look at how clutter behaves. I look at whether randomness feels earned or sprayed on. I look for scenes that feel designed instead of captured.

Local realism is cheap now but global truth ain’t, and that gap is the tell.

† Read: Controllable World Models are here and of course everyone is pretending they always wanted this | LinkedIn

---

Your turn to try

Before going any further, it’s time to stop nodding along and actually put your knowledge to the test.

The image below comes straight from the study that I’ve been talking about. Its a simple grid of faces where some are real, and others are synthetic. I know it is hard, but knowledge how they’re generated should help you a lot.

Before you begin, here’s the uncomfortable setup.

Half the faces were AI-generated in the study. That means random guessing should land you at fifty percent meaning that you can flip a coin and walk away proud if you want because most people still did worse. Typical participants identified AI faces correctly about a third of the time, even the so-called super-recognizers – the genetic overachievers of face perception – who barely scraped past chance without training.

Look at the image below.

Don’t rush. Don’t zoom into pores like it’s 2018. Apply what you just learned about local plausibility versus global coherence, about polish versus mess and scenes that feel captured versus scenes that feel designed.

Here are the learnings again:

1. Ignore the parts and read the whole. Do not interrogate individual features like you’re some TSA agent. A cheek looking right proves nothing. AI is excellent at local plausibility. That battle is over. Reality is internally consistent, synthetic images often are not.
2. Look for polish as a smell. It’s a warning sign. Simply ask yourself, is the lighting a little too considerate or is nothing actively inconvenient or ugly? Real photos have friction, because life is cruel but AI images contain optimization.
3. Check the global physics and not local anatomy. Do not count fingers. That’s 2022 nostalgia, but check if shadows behave consistently and depth feel earned or that light direction makes sense across the whole frame. AI often gets local physics right but global physics wrong.
4. Ask whether randomness has a cause because real randomness has sources. Those can be due to bad light or a cheap lens, awkward timing in my case, or movement, weather, etc.
5. Decide whether the scene feels captured or designed. Now, this is the killer question. Because captured moments look accidental and contain visual compromises and feel absolutely unconcerned with your opinion whereas designed moments feel emotionally calibrated and seem to know why they exist.

And the final instruction is to notice the repetition across different faces since one image can fool you but a set of images just reek of patterns like similar emotional range or texture behavior and similar background cooperation.

Now, go make your call, and remember, you are trying to decide whether it behaves like reality at scale, not in detail.

Good.

This was actually a perfect exam, because it forces you to stop hunting for glitches and it learns you to apply a way of looking.

First, let’s lock in the answers, because otherwise everyone is going to argue in the comments.

Answer key from the study

• Top row: AI-generated faces
• Middle row: Real human faces
• Bottom row: AI-generated faces

Now the interesting part.

Not that this is the answer, but why this makes sense once you know where to look.

I’ll walk through it using exactly the concepts you just read, so you can test whether you actually learned something or you were simply nodding along.

---

Why the top row is AI

The top row looks fine at first glance, but that is the trap, because local plausibility is excellent. Every face passes the close-up sniff test.

• Eyes look like eyes
• And skin looks like skin
• Hair looks like hair

And lighting looks “professional”

Now, this is local plausibility, and diffusion or GAN-based face generators have become extremely good at it, and that’s why untrained people get wrecked here.

Where it starts to wobble

When you stop looking at the parts and try to get a picture of the whole, you see patterns starting to appear. Let me walk you through a few of them . . .

• Uniform polish – every face looks gently optimized. No one looks tired, greasy, unevenly lit, or awkward. Real photos love awkward but AI avoids it like the plague.
• Hairline and edge blur – several faces show a subtle softness where hair meets skin or where it meets the background. Not broken, just… undecided. This matches what the study trained participants to notice.
• Texture consistency – skin texture looks evenly distributed, democratic almost. Real faces usually have uneven zones, oilier T-zones, harsher pores on one side, shadows that misbehave.
• Expression neutrality – these smiles sit in a narrow emotional band – pleasant, safe, LinkedIn-ready – you know, but real people overshoot, undershoot, or miss the mark entirely.

This is what strong instruction or model bias looks like. And in the end it turns out too clean and calm and simply too agreeable.

---

Why the middle row is real

This row feels worse. And that should be your clue.

Local plausibility = imperfect. You immediately see things that the AI hates.

• Uneven lighting
• Harsh shadows
• Slightly unflattering angles
• Expressions that are not optimized for anything

Again, nothing is wrong, but plenty is inconvenient.

And next, global coherence = solid. Now look at the full scene logic.

• The lighting matches the environment – outdoor faces have messy, directional light. Indoor faces look flat or ugly in a way only reality produces.
• There’s asymmetry everywhere – eyes aren’t perfectly aligned and smiles pull more on one side, the hair behaves badly. This is reality, physics.
• And there’s noise that belongs like grain, blur, and softness feel like camera limitations, not algorithmic indecision. The noise is integrated, and doesn’t feel sprayed on.
• Realism is boring. These images do not feel designed, but captured and that is the hardest thing for generative models to fake.

So, this row has global agreement. Light, texture, expression, and randomness all tell the same story and that’s what I meant by “the boring truth of everyday light and the messy randomness of real life”.

---

Why the bottom row is AI

This is the row that usually manages to split audiences. Some people say “but these look even more real than the top row”.

Exactly.

Weak guidance drift

The bottom row is what happens to an image when instruction loosens.

• Structure softens
• Expressions feel more natural
• And small imperfections appear

This does look like realism at first glance, but again, zoom out.

The subtle tell-tale-signs

• Chaos without cause – details feel random, but not earned. You get this? Real randomness has a source. You know, lighting, lens, movement, but this feels like statistically plausible noise and not physical consequence.
• The background cooperates a bit too well – the backgrounds support the face. They never fight for attention, never clash nor interrupt, but real environments are rude.
• Edge behavior repeats – this is about patterns and AI likes them. Look at the hair, the collars, and the background transitions that share similar softness patterns across different faces. That repetition is not biology though, that’s a sign of generation.
• Emotion still constrained – expressions live in a narrow band of “approachable human” – even here. There are no weird grimaces or awkward social mistakes, things like bad timing.

This is what I talked about earlier – weak instruction realism. It feels human because it drifts and it fails because it drifts everywhere in the same way.

---

Why this test works so well

I’ve given this training as part of a workshop I’ve developed for elderly people that are really afraid they’re going to get scammed. And even though they won’t become FBI profilers after a few hours studying pictures, they have more confidence in their ability to spot fakes because people learned to redirected attention.

Untrained viewers ask:

“Does this face look real?”

Trained viewers now ask:

“Does this scene behave like reality across the whole image?”

That shift alone explains the jump in performance.

I think you’re now good to go because I’ve given you the gist. But when you stick around, I promise you’ll be even better at spotting fakes that any other professional out there.

---

The pro-level checklist that improves your odds

Thank you for sticking around! Now comes the pro-level knowledge, and the trick to pass this level is that you need to develop a routine.

And the routine has one rule. You zoom in. Not for drama, but For evidence. Zoom to say 200 or 300 percent and look at the parts of the image that are hard to fake because they require consistent structure.

Let’s look at some of the details. . .

Eyes

Eyes are tiny physics problems. They reflect the environment, and have wet surfaces. They carry small highlights called catchlights.

Now, in a real photo, both eyes live in the same world. The reflections are consistent. The shape, direction, and brightness of highlights matches the light source.

But in synthetic images, you often get mismatched highlights where one eye reflects a square window and the other reflects a round studio light. Or the highlight suggests that the light comes from the left, while the nose shadow suggests the light comes from the right. Got it?

These are subtle. I used to miss them at first, but you can train yourself to look.

Also watch for the “glassy stare”. It’s kinda like the 1000-yard stare if you’ve ever seen a picture of a soldier who’s witnessed gruesome combat. Some synthetic faces have eyes that look too sharp, too clean, too centered. Humans have micro asymmetries. Real photos capture tiny tension and tiny blur. Synthetic faces often look like they are politely waiting for your credit card.

Teeth

Teeth expose a lot of generator weaknesses because teeth are repetitive structures with subtle variation and they have irregularities in size, spacing, translucency, and alignment but real photos capture a messy mix of shadow, saliva reflections (yes, that’s a thing), and tiny color variation. Synthetic teeth on the other hand often look too uniform or oddly soft. Sometimes the gum line looks vague or the individual teeth merge slightly and overall there’s an uncanny cleanliness that feels like dental propaganda.

To be fair, a person might have false teeth, but than again, this is just one sign of many.

Oh, also watch for missing teeth in places where it makes no sense, or teeth that exist as a white band rather than a set of shapes.

Skin

Skin is the new battlefield now because real skin is not a smooth surface since it has pores, small hairs, tiny color variation, and subtle oil reflections. Real skin also has subsurface scattering. Let me explain, light enters the skin, bounces around, and exits with a warm softness, especially in ears, nostrils, fingertips, and thin areas.

Synthetic skin often fails here in one of two ways.

One failure is wax skin. Opaque, flat, plastic. Shadows are too heavy or too smooth. The image looks airbrushed.

The other failure is hypertexture. Skin that looks like a high resolution scan of “skin texture” pasted on top, with pores distributed too evenly. The face becomes a wallpaper sample. Biology becomes a pattern generator again.

You’ve got to look at the transition zones. The area around the nose, cheeks, and forehead. In real photos, oil and light behave differently across those zones, where synthetic images treat the whole face with a single texture logic.

Hair edges

I personally find hair against background to be a hard thing to distinguish. Real photos have messy hair and flyaways, semi transparency, motion blur and depth of field, stray strands crossing.

Synthetic images sometimes blur hair edges in unnatural ways. The boundary between hair and background can look smeared. Sometimes you see a halo effect and other times the hair looks too perfect, like it was painted.

Look at where hair meets skin around the forehead and ears. Look at how strands overlap and observe the edge crispness. Real photos have uneven crispness because focus and lens behavior changes across the frame.

Jewelry and small objects

Earrings, glasses, necklaces, buttons, zippers. These are great tells because they require consistent geometry and consistent occlusion. The object has to sit in space. It has to cast tiny shadows and it has to interact with hair and skin which is complex, so synthetic images often “cheat” those details. An earring melts into an earlobe or a glasses frame merges into hair. A necklace chain becomes ambiguous. That sort of thing.

Small objects also expose symmetry issues. Humans love symmetry in generation. But reality loves asymmetry.

---

The physics sanity check

After you examine details, you do the physics pass where you treat the image like a scene that claims to exist in 3D space.

You ask a few simple questions, like where the light is coming from and if everything agrees with each other.

Shadows and the multi sun problem

Outdoor scenes usually have one dominant light source, the sun. Shadows should have consistent direction but in synthetic images, shadows sometimes drift. A tree shadow points one way, a person shadow points another way. The scene kinda acts like it has multiple suns. That is of course physically absurd, yet visually it can hide in plain sight.

Also watch shadow softness. Real shadows often show contact hardening. The shadow is sharp near the object touching the ground and becomes softer with distance. Synthetic shadows can look uniformly soft or uniformly sharp, because the model is painting a shadow rather than simulating light.

Reflections and the vampire effect

I call if the “vampire effect” because mirrors and windows are wonderful snitches since they demand consistency from a second viewpoint.

In real photos, reflections are annoying but coherent. In synthetic images, however, reflections sometimes forget objects.

What do I mean by that.

A person appears, but the reflection lacks them. A hat disappears. A lamp shows up only in the reflection.

Water reflections are another tell. Real water distorts reflections based on ripples. Synthetic water sometimes looks textured, but the reflection stays too perfect. The scene cannot decide if the water is calm or turbulent.

Contact points

Real objects touch other objects in believable ways. Shoes compress grass. Hands grip mugs with pressure. Straps sit on shoulders with weight. Fabric folds around joints.

Simple.

Synthetic images can get the outline right but still miss contact logic. It doesn’t occur that much anymore, but you can still find fingers clip through objects or rings that float slightly, shoes that hover or a backpack strap that becomes a painted line rather than a physical strap.

If the image looks like it was made for a poster, contact logic suffers because poster images favor clean shapes over messy physical truth.

---

The meaning check

This is the part of the workshop nobody actually wants to do because it requires thinking and we all know that thinking ruins the dopamine.

For each scene or image, you ask yourself a blunt question.

“Does this scene make sense as a real moment”.

Not “is it possible”. Possible is cheap, but more like “is it plausible as an accidental capture”.

That’s the difference. Synthetic images often carry an emotional charge. That charge is a red flag. Real photos have context. Synthetic images have intention.

Too perfect

When an image looks like a movie still, ask why a random person captured it.

When a breaking news image has perfect framing, perfect lighting, perfect composition, and perfect emotional timing, ask who benefited.

Reality is messy. Cameras capture mess. Viral content looks designed.

Too emotionally convenient

Many synthetic images are engineered to provoke instant emotion, think of outrage, awe, fear, adoration. The image aims for maximum shareability.

Now, when your body reacts before your brain does, that is the signal.

The content wants your nervous system!

Context gaps

Context also helps you make the right call. Ask yourself where the image came from and who posted it but also when they posted it and what else that account posts, whether the same image appears in other places with different stories, that sort of stuff.

A synthetic image can be technically perfect and still collapse under basic context checks.

---

The boring but powerful verification moves

I have saved the boring part for last, so if you’re like me, you can skip it.

Visual inspection is useful, but context verification wins.

Do a reverse image search. Find earlier appearances. Check if it came from an AI gallery or a prompt sharing thread. Check if the same image was used in an unrelated story.

This is not the glamorous piece, but it is effective, then look for provenance signals when they exist. Some images carry content credentials or signed provenance in their metadata. If a platform preserves it, you can sometimes see whether the image came from a camera, an editor, or a generator.

But metadata is fragile. Social networks strip it and screenshots obliterate it and if you know your stuff, a bad actors can fake basic metadata.

Still, provenance is the direction the world is moving. Detection keeps losing and assertion becomes the replacement. People will treat “signed capture” as trustworthy and treat “unsigned viral image” as suspicious by default and yes, that shift will feel unfair but it is also quite necessary.

---

How to use the five minute training idea in real life

The study’s five minute training plus my workshop materials are not a magic spell, but a ritual instead because it teaches your attention where to go, and that matters because your eyes are not dumb. Your eyes are trained by habit and if your habit is scrolling, you see nothing but when it turns into inspection, you start to see the patterns.

So build a micro routine. Before you trust an image that matters, do this.

Look at eyes. Look at hair edges. Look at shadows. Look at reflections. Look at background text. Then ask the meaning question. Who benefits.

This takes less than a minute.

Will you still be fooled. Yes. Everyone will. The goal is not perfection but to reduce your failure rate enough that you do not become free labor for scams. That is the price of living in 2026 where authenticity is a premium feature.

Now go forth and practice, my smart friend.

Signing off,

Marco

---

I build AI by day and warn about it by night. I call it job security. Big Tech keeps inflating its promises, and I just bring the pins and clean up the mess.

---

👉 Think a friend would enjoy this too? Share the newsletter and let them join the conversation. LinkedIn, Google and the AI engines appreciates your likes by making my articles available to more readers.

To keep you doomscrolling 👇

1. I may have found a solution to Vibe Coding’s technical debt problem | LinkedIn
2. Shadow AI isn’t rebellion it’s office survival | LinkedIn
3. Macrohard is Musk’s middle finger to Microsoft | LinkedIn
4. We are in the midst of an incremental apocalypse and only the 1% are prepared | LinkedIn
5. Did ChatGPT actually steal your job? (Including job risk-assessment tool) | LinkedIn
6. Living in the post-human economy | LinkedIn
7. Vibe Coding is gonna spawn the most braindead software generation ever | LinkedIn
8. Workslop is the new office plague | LinkedIn
9. The funniest comments ever left in source code | LinkedIn
10. The Sloppiverse is here, and what are the consequences for writing and speaking? | LinkedIn
11. OpenAI finally confesses their bots are chronic liars | LinkedIn
12. Money, the final frontier. . . | LinkedIn
13. Kickstarter exposed. The ultimate honeytrap for investors | LinkedIn
14. China’s AI+ plan and the Manus middle finger | LinkedIn
15. Autopsy of an algorithm – Is building an audience still worth it these days? | LinkedIn
16. AI is screwing with your résumé and you’re letting it happen | LinkedIn
17. Oops! I did it again. . . | LinkedIn
18. Palantir turns your life into a spreadsheet | LinkedIn
19. Another nail in the coffin – AI’s not ‘reasoning’ at all | LinkedIn
20. How AI went from miracle to bubble. An interactive timeline | LinkedIn
21. The day vibe coding jobs got real and half the dev world cried into their keyboards | LinkedIn
22. The Buy Now – Cry Later company learns about karma | LinkedIn