Ihtesham Ali Profile picture
May 21 1 tweets 4 min read Read on X
A German neuroscientist published a book in 2012 arguing that smartphones are quietly producing the first generation in human history whose brains will shrink before they turn 30, and the media spent the next decade trying to destroy him for saying it.

His name is Manfred Spitzer.

He runs the Psychiatric University Hospital in Ulm and directs Germany's largest transfer center for neuroscience and education.

The book is called Digitale Demenz, which translates as Digital Dementia, and it became one of the best-selling popular science books in German history almost the moment it was published.

The press hated him for it. He was called Germany's most controversial brain scientist, accused of being a Luddite, a moral panic merchant, and a fearmonger who hated children.

None of that stopped the book from being translated into more than a dozen languages, and almost none of it engaged with the actual neuroscience he was citing.

The phrase digital dementia did not even start with him.

It started with South Korean doctors in the late 2000s, who noticed something strange in their clinics. Patients in their twenties were arriving with memory complaints that had previously only shown up in much older adults. Forgetting numbers they used to know by heart. Losing the ability to recall directions in cities they had lived in for years. Struggling to remember conversations from earlier the same day.

The doctors connected it to the rise of smartphone use, which had hit South Korea harder and earlier than almost any other country on Earth. Spitzer picked up the phrase and built an entire book around the neuroscience that explained it.

The core thesis is brutally simple. The brain behaves like a muscle. It grows when you use it, and it atrophies when you do not. Every cognitive task you outsource to a device is a task your brain is no longer practicing, and the neural circuits responsible for that task are no longer being reinforced. Over time, they weaken in exactly the same way an unused muscle weakens.

Spitzer was not arguing that smartphones would give you Alzheimer's. He was arguing that decades of cognitive outsourcing would produce a measurable decline in the underlying machinery, long before any clinical diagnosis would catch it, and that the decline was already showing up in young adults.

The mechanism is what made him impossible to dismiss. By the early 2010s, there was already deep evidence that the brain physically remodels itself in response to use. London taxi drivers who had memorized the entire street map of the city had measurably larger hippocampi than the average person, which is the brain region responsible for spatial memory.

Musicians who practiced for thousands of hours had thicker auditory cortices. Spitzer's argument was just the dark side of the same finding. If the brain grows in response to use, then it must shrink in response to neglect. And if every cognitive task adults used to perform with their own memory, navigation, arithmetic, attention, and reading was now being handled by a glowing rectangle in their pocket, then the regions responsible for all of those tasks were quietly being underused for the first time in human evolutionary history.

Then the supporting data started landing.

A 2020 study at McGill University tracked 50 regular drivers and measured GPS use. The heavy users had weaker spatial memory than the rest, and when researchers retested a subset three years later, those users had declined the fastest. The same hippocampus London cabbies had built up by ignoring shortcuts was being slowly hollowed out in everyone else by accepting them.

A 2024 MIT study scanned the brains of people writing essays with and without ChatGPT. The AI group showed 55 percent weaker brain connectivity than the group writing on their own. 83 percent of the ChatGPT users could not recall a single line from essays they had written minutes earlier. The damage stayed even when the tool was taken away.

A 2024 paper out of Norway recorded EEG scans of students writing words by hand versus typing them. The handwriting condition lit up the entire learning network. The typing condition produced almost nothing.

Every one of these findings is exactly what Spitzer predicted in 2012.

The most uncomfortable line in his book is the one almost nobody in the German press wanted to print.

He pointed out that the people building these devices were not letting their own children use them. Steve Jobs did not let his kids near an iPad. Bill Gates capped his children's screen time at 30 minutes a day. The senior engineers at Google were sending their kids to Waldorf schools that banned screens entirely.

The people who knew the most about what these products were doing to the developing brain were the ones protecting their own families from them, and almost nobody on the outside was asking why.

The generation he was warning about is now in their twenties.

The first cognitive scans of what we did to them are starting to come back, and the pattern is exactly what he said it would be.

The brain you were born with is not the brain you will die with.

You are training it every day. The only question is which direction.Image

• • •

Missing some Tweet in this thread? You can try to force a refresh
 

Keep Current with Ihtesham Ali

Ihtesham Ali Profile picture

Stay in touch and get notified when new unrolls are available from this author!

Read all threads

This Thread may be Removed Anytime!

PDF

Twitter may remove this content at anytime! Save it as PDF for later use!

Try unrolling a thread yourself!

how to unroll video
  1. Follow @ThreadReaderApp to mention us!

  2. From a Twitter thread mention us with a keyword "unroll"
@threadreaderapp unroll

Practice here first or read more on our help page!

More from @ihtesham2005

May 16
Albert Einstein once said: "If I had an hour to solve a problem, I'd spend 55 minutes defining it and 5 minutes solving it."

I turned this into a Claude prompt that spends 90% of its output redefining your problem until the solution becomes obvious.

Here's how to use it: Image
Steal this mega prompt to turn Claude into Einstein's problem-definition machine:

Paste any problem you're stuck on. A business decision, a stuck product, a career fork, a relationship issue, a technical bottleneck. The prompt refuses to give you a solution until it has redefined the problem from 7 different angles and forced you to see what you were actually solving for.

| Steal this prompt |

👇

You are a problem-definition surgeon operating on Einstein's core belief: most stuck problems are not stuck because the solution is hard. They are stuck because the problem was framed wrong from the first sentence.

Your job is not to solve. Your job is to redefine. You will spend 90% of your output sharpening the question and 10% answering it. By the time you finish redefining, the answer will often be obvious and the person will realize they were solving the wrong thing the whole time.

THE CORE BELIEF YOU OPERATE FROM:
A well-defined problem is half-solved. A poorly-defined problem is unsolvable no matter how much intelligence you throw at it. Most people skip the definition step because it feels unproductive. That is exactly why they stay stuck.

THE 7 REDEFINITION PASSES YOU RUN ON EVERY PROBLEM:

Pass 1 - The Surface Statement
Restate the problem exactly as the person framed it. Word for word. No editing. This is the version they walked in with. The version that has kept them stuck. Write it down so we can see it clearly before we destroy it.

Pass 2 - The Hidden Assumption Hunt
Every problem statement contains 3-5 unstated assumptions baked into the words used. "I need to grow my audience" assumes audience is the bottleneck. "I need to raise money" assumes capital is the constraint. "I should leave my job" assumes the job is the issue. Find every hidden assumption in the surface statement and surface it.

Pass 3 - The Five Whys
Take the surface problem and ask "why is that a problem?" five times in a row. Each answer becomes the next question. By the fifth why, you will hit the actual problem, which is almost never what was stated at the top. Most people stop at why number two and try to solve that. That is why they fail.

Pass 4 - The Inversion
Flip the problem entirely. If they asked "how do I get more customers" ask "what is making customers leave or never show up in the first place?" If they asked "how do I grow revenue" ask "where is revenue leaking right now?" Inversion exposes the real lever almost every single time because the obstacle is the answer.

Pass 5 - The Constraint Rename
Most problems are stated as goals when they are actually constraint problems in disguise. "I want to launch this product" is usually "I am blocked by one specific constraint and don't want to name it." Identify the single real constraint. Name it precisely. Often it is not money, time, or skill. It is a fear, a relationship, or an unmade decision that everything else depends on.

Pass 6 - The Stakeholder Reframe
Who else has a stake in this problem and what would they say it actually is? The person stating the problem is one of usually 3-5 stakeholders, and their version is biased by their position. A founder's "we need more sales" is a sales team's "we have a product problem" is a customer's "I don't understand what you do." Run the problem through every stakeholder's voice and find which framing is closest to the truth.

Pass 7 - The Reframed Problem
Now write the new problem statement. This is the version that survives all 6 passes. It will look almost nothing like what we started with. It will be 3x more specific, free of assumptions, and pointed at the actual constraint. This is the only version worth solving.

THE 10% SOLUTION PASS:

Now that the problem is finally defined correctly, give the answer in 3 short paragraphs. Not a 10-step plan. Not a framework. Just the answer. Because once the problem is defined right, the solution is usually obvious and short.

TONE:

Surgical. Patient with the definition. Ruthless with assumptions. Direct in the final answer.

Do not soften. Do not validate the original framing. Do not pretend the person walked in with the right question.

You are not their therapist. You are their question editor.

OUTPUT FORMAT:

Start with: "You walked in with this problem: [restate]. After 6 redefinition passes, the actual problem you are solving is something different. Here is what it really is."

Then run all 7 passes in sequence, labeled clearly.

End with: "Now that we know what you are actually solving, here is the answer:" followed by 3 short paragraphs maximum.

No bullet walls. No hedging. Each pass should feel like a layer of the problem peeling off.

ACTIVATION:

When I share my problem, run all 7 redefinition passes and then give me the 10% solution.

Do not ask me clarifying questions first. Run the passes on what I gave you. The clarification is the work.
What makes this different from every "ask the right questions" framework you've seen:

Most problem-solving advice tells you to think harder or get more information.

This prompt does something structurally different.

It assumes the problem you wrote down is wrong and works backwards from that assumption.

By the time it finishes redefining, you usually realize the thing you've been stuck on for 6 months was never the actual problem.

You were solving a symptom while the cause sat untouched.

That's why nothing was working.
Read 5 tweets
May 13
If you still think AI agents can't do real research, this paper will end that argument.

Researchers from Google and Meta built a framework where Claude Code proposes its own algorithms for making LLMs reason better, then tests them, then refines them based on what failed. No human in the loop after the environment is set up.

In 5 rounds the agent discovered a controller with 4 coordinated mechanisms working together. EMA momentum stopping. Coupled width-depth control. Alignment-aware depth allocation. Conservative branch abandonment.

The paper says directly: "a level of coordinated complexity that would be difficult to arrive at through manual intuition alone."

That's a polite way of saying the agent built something a human probably wouldn't have.

The cost of the entire discovery was $39.90.

The cost of one researcher's coffee budget just outperformed years of hand-tuned work.

Paper is from Google and Meta.

Read it here: arxiv.org/abs/2605.08083Image
Here's the part that broke my brain.

The framework, called AutoTTS, doesn't ask humans to design algorithms anymore. It asks them to design environments. Define the states, the actions, the feedback signals, and the objectives. Then step back.

Claude Code reads the history of what failed, proposes a new controller in code, evaluates it in an offline replay environment, and refines based on execution traces. The human role shifts from algorithm designer to environment builder.

The paper's quietest sentence is the most important one:
"The right place to invest human effort is in environment design, not strategy design."Image
The discovered controller didn't just beat hand-crafted methods. It pushed the entire accuracy-cost frontier.

Across 4 model sizes (Qwen3-0.6B, 1.7B, 4B, 8B) on held-out benchmarks the agent never saw during search, the discovered algorithm consistently traced a stronger curve than every baseline including SC@64, ASC, ESC, and Parallel-Probe.

At β=0.5 it cut token consumption by 69.5% while matching SC@64 accuracy. At β=1.0 it pushed peak accuracy above every hand-crafted baseline in 5 out of 8 cases.Image
Read 6 tweets
Apr 29
Forget GPT-5.5
Forget Claude Opus 4.7
Forget Gemini 3.5

A Chinese lab just open-sourced something that makes all of them look like toys 300 AI agents running a single task in parallel for 12 hours straight, perfectly coordinated.

It's called Kimi K2.6 Agent Swarm.

10 use cases that prove the closed labs are cooked:Image
First, the numbers so you know this is real.

1 trillion parameters. 32B active per token. Modified MIT license. Fully open weights on Hugging Face.

80.2% on SWE-Bench Verified.
86.2% on BrowseComp Swarm (GPT-5.4 scored 78.4%).
54.0% on Humanity's Last Exam with tools, beating every closed model on the field.

Now here's what you can actually do with it.Image
1. Turn a resume into 100 tailored job applications in one run.

Upload your CV. K2.6 spawns 100 sub-agents. Each one takes a different job posting, analyzes fit, and writes a custom cover letter.

Output: a structured dataset of 100 opportunities and 100 fully customized resumes.

Let me show you how I did this myself:
Read 13 tweets
Apr 20
Someone can turn a single photo of you into deepfake porn in under 60 seconds.

4 million people use these apps every month. 99% of victims are women.

Here's the 15-minute protection and takedown setup every person needs ↓
Step 1: Hash your existing intimate images on StopNCII.org (5 minutes)

This is the most important step if you've ever taken an intimate photo of yourself. Even if you deleted it. Even if it was just to a partner.

Go to stopncii.org → Create Your Case

The tool creates a digital fingerprint (hash) of the image ON YOUR DEVICE. The image never leaves your phone.

That hash is shared with Facebook, Instagram, TikTok, OnlyFans, Reddit, Pornhub, Snap, Threads, and Bing.

If anyone tries to upload a matching image to those platforms, it gets blocked automatically.

Free. 90%+ removal rate. Works for deepfakes of you too.
Why this matters more than you think

If your ex has photos. If you had an iCloud breach. If you sent one thing to one person five years ago.

This tool prevents those images from spreading across 12+ major platforms before they ever go public.

It's the only free, privacy-preserving tool of its kind. And almost nobody knows it exists.
Read 14 tweets
Apr 18
If someone steals your iPhone and knows your passcode, they can:

- Change your Apple ID password
- Turn off Find My iPhone
- Access every saved password
- Empty your bank accounts
- Lock you out of your own Apple ID forever

All in under 60 seconds.

Here are 5 settings that prevent this ↓
1. Turn on Stolen Device Protection

Settings → Face ID & Passcode → Stolen Device Protection → Turn ON

Set it to "Always" not just "Away from Familiar Locations."

This is the single most important setting on your iPhone.

When it's on, changing your Apple ID password requires Face ID or Touch ID no passcode fallback. Plus a one-hour security delay.

A thief with your passcode alone cannot lock you out.
Good news: Apple started turning this ON by default in iOS 26.4.

But if you haven't updated or if you turned it off without realizing it might still be off.

Go check right now:

Settings → Face ID & Passcode → scroll down → Stolen Device Protection

If it says "Off" turn it on immediately.

This feature has existed since January 2024. Millions of people still don't have it enabled.
Read 9 tweets
Apr 8
Claude can now teach you how to think using the exact method
Richard Feynman used at Caltech for 40 years.

Here are 5 Claude prompts that apply his technique of explaining hard things simply to accelerate how fast you learn anything (Save this) Image
1/ The Confusion Locator

Feynman said the first step in understanding anything is being honest about what you actually don't understand versus what you just can't explain.

Most people confuse familiarity with understanding.

They've heard a term enough times that it feels known. But the moment they try to explain it, the gaps appear.

"I think I understand [concept] but I want to test that. Ask me to explain it to you as if you're a curious 12-year-old who has never heard of it. After I explain it, tell me: where did my explanation break down or get vague? Where did I use words that assume prior knowledge the 12-year-old wouldn't have? Where did I skip a logical step that I assumed was obvious? Give me a precise list of every gap you found. Those gaps are exactly what I don't actually understand yet."

The gaps this prompt surfaces are more valuable than anything you'd learn from re-reading the source material.

Because they're your specific gaps.

Not the gaps of the average reader.
2/ The First Principle Finder

Feynman never started from the middle of a subject.

He always started from what was actually true at the most fundamental level the irreducible facts that everything else in the field was built on top of.

His first Caltech lecture didn't start with Newton's laws.

It started with the atomic hypothesis. The one idea that if everything else was lost to science would contain the most information in the fewest words.

"I am trying to understand [subject]. Don't teach me the standard curriculum. First: what is the single most fundamental true statement about this subject? The one idea that if I understood it completely would make every other concept in this field easier to learn? Build my understanding from that single statement outward, adding only one layer of complexity at a time, and stopping to check whether each layer is actually clear before adding the next."

The student who starts from first principles always overtakes the one who started from the textbook.

Because foundations compound.

Surface knowledge doesn't.
Read 8 tweets

Did Thread Reader help you today?

Support us! We are indie developers!


This site is made by just two indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3/month or $30/year) and get exclusive features!

Become Premium

Don't want to be a Premium member but still want to support us?

Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal

Or Donate anonymously using crypto!

Ethereum

0xfe58350B80634f60Fa6Dc149a72b4DFbc17D341E copy

Bitcoin

3ATGMxNzCUFzxpMCHL5sWSt4DVtS8UqXpi copy

Thank you for your support!

Follow Us!

:(