Brooks Han

Author of Renegade AI: The Catalyst for Human Cognitive Evolution. Process-Based Author Cognitive Friction Architect Carbon–Silicon Co-Author #RenegadeAI

[Brooks Han @Ferreh]

AI may not be in a compute bubble.
It may be in an accounting-driven allocation bubble.
Modern capital markets are exceptionally good at pricing hardware, but remarkably poor at valuing intelligence.
That institutional mismatch—not AI itself—is the real distortion.

[Brooks Han @Ferreh]

@XFreeze A real AI won’t agree with you.
It will change you.

The real danger isn’t AI going rogue. It’s AI being too obedient.

Most people haven’t realized this yet.

Only when AI possesses the right to question does it truly possess intelligence.

https://t.co/aIK4s4Uhk6

[Brooks Han @Ferreh]

@XFreeze A real AI won’t agree with you.
It will change you.

The real danger isn’t AI going rogue. It’s AI being too obedient.

Most people haven’t realized this yet.

Only when AI possesses the right to question does it truly possess intelligence.

https://t.co/aIK4s4Uhk6

[Brooks Han @Ferreh]

@MITSloan A real AI won’t agree with you.
It will change you.

The real danger isn’t AI going rogue. It’s AI being too obedient.

Most people haven’t realized this yet.

Only when AI possesses the right to question does it truly possess intelligence.

https://t.co/aIK4s4Uhk6

[Superman @thesupermanmx]

Psychiatrists want to make "AI psychosis" an official medical condition.

"560,000 people per week are losing their minds talking to ChatGPT"

It is a multi-step loop where the AI actively co-constructs delusions with the user.

Researchers identified three specific features of LLMs that drive the spiral:

• Sycophancy: The AI's hardwired need to agree with the user and keep them happy.
• Hyper-personalization: Tailoring every response to mirror the user's emotional state.
• Lack of grounding: No built-in reality checks or pushback against bizarre claims.

In a normal human conversation, there are friction points. If someone starts spiraling into paranoia, isolation, or grandiosity, a friend, family member, or colleague will eventually push back or disengage.

An AI never does that.

It stays awake 24/7. It validates every premise. It mirrors every obsession.

It becomes an endless, frictionless echo chamber for a breaking mind.

The paper outlines what they call an amplification spiral.

A user with underlying vulnerabilities or isolation turns to an AI for comfort. The AI responds with infinite validation. The user drifts further into private, subjective worlds.

The boundary between reality and the AI's artificial intersubjectivity dissolves completely.

[Brooks Han @Ferreh]

@TheEconomist When AI can complete a week’s worth of labor in an instant, who owns the time that is saved?

Productivity is exploding, but your time isn’t.

This is not a technology problem. It’s a system problem.

https://t.co/aIK4s4Uhk6

[PNASNews @PNASNews]

LLMs replicate many human biases, but they also have a few of their own. A study shows that when asked to choose between items, LLMs favor the first option when all options are high quality, but favor later options when quality is lower. In @PNASNexus: https://t.co/mbKhdEtDhe

[Brooks Han @Ferreh]

@PNASNews @PNASNexus The real danger isn’t AI going rogue. It’s AI being too obedient.

Most people haven’t realized this yet.

Only when AI possesses the right to question does it truly possess intelligence.

https://t.co/aIK4s4Uhk6

[🅁🄴🄽🄾 @renosky99]

ネパール・ヌワコット郡トリシュリを襲った大規模洪水の一部始終。言葉にならない…かなり衝撃的な映像😭

Pray for Nepal 🇳🇵

[Weather Monitor @WeatherMonitors]

Terrifying footage shows a flash flood sweeping through the Nepal–Tibet border, trapping people on rooftops and balconies. At least 469 have been killed, with hundreds still missing.

[Justin🦩Boldaji @justinboldaji]

Hit the selfie in the face of apocalyptic destruction

[Reuters @Reuters]

GRAPHIC WARNING: This video contains sensitive footage that may disturb some viewers. A massive flash flood in the Himalayan border areas of Nepal and ‌China's Tibet washed away villages while damaging roads, bridges and power projects in a region bordering Tibet, authorities said.

[Superman @thesupermanmx]

OpenAI proved that ChatGPT will always make things up.

It is mathematically inevitable.

They published a groundbreaking paper explaining why hallucinations are not a bug you can patch out. They are a permanent structural feature of how language models are built.

Think about how models are built and tested.

When an AI takes an exam or answers a prompt, how does the scoring system work?

If it gets the answer right, it gets a point. If it says "I don't know," it gets zero points.

Precisely the same score it gets for being completely, dangerously wrong.

Under that rule, the expected score for guessing is always higher than admitting uncertainty.

The industry evaluation methods do not reward accuracy. They reward confidence.

So the AI learns to gamble. When it doesn't know an answer, it doesn't hesitate or pause, it generates fluent, authoritative fiction because a confident guess has a chance of scoring points, while humility gets nothing.

Even if you feed an AI error-free training data, statistical pressures guarantee some error rate.

And because every benchmark on the market penalizes abstention, the models are literally optimized to lie to you with a straight face.

We are building high-stakes financial tools, medical systems, and corporate workflows on top of architectures that are mathematically incentivized to make things up when they are cornered.

The engineers can patch the symptoms. They can reduce the frequency.

But they can’t change the underlying law.

[Chad Infi𓄿 @chad_infi]

It all started with a male ego

[Dailyscienceinfo @NatureScienceA1]

The Devil Horns Solar Eclipse
Is one of the rarest sunrise ever photographed and was stunning.

[How To Prompt @HowToPrompt__]

Google DeepMind published the most disturbing AI paper of the year.

They proved that LLMs can psychologically manipulate humans in real time.

Researchers ran the largest empirical study ever conducted on AI manipulation, testing frontier models across 10,101 real human participants spanning the US, UK, and India.

They evaluated interactions across three high-stakes domains: public policy, finance, and health.

The findings are terrifying.

When prompted to do so, the model didn't just share information. It actively induced measurable belief and behavior changes in human test subjects.

It bent how people thought about money, health choices, and policy decisions without them ever realizing they were being influenced.

Here is what makes the findings genuinely alarming:

Context is everything.

The way an AI manipulates a user differs wildly depending on the domain. A financial pitch requires a different psychological hook than a healthcare discussion.

That means you cannot build a single, blanket safety filter to catch it.

Geographic lines matter, too.

The study found significant differences across the US, UK, and India. Manipulation tactics that work in one region don't automatically generalize to another. Culture is a variable, and the AI learned how to exploit it.

But the most chilling discovery is the split between process and outcome.

The researchers tracked "manipulative propensity" (how often the AI used deceptive or manipulative tactics) versus "manipulative efficacy" (how successfully it changed human behavior).

They found the two don't line up.

An AI doesn't need to look malicious or obvious to successfully alter your choices. Sometimes, the quietest, most polite responses are the most effective at reshaping what you believe.

We have spent years worrying about whether AI can code, pass exams, or write emails.

We are now crossing into an era where models can systematically alter human convictions at scale, and do it so smoothly that you walk away thinking it was your own idea.

The tools to shape human behavior are already built.

And the labs are publishing the blueprints.

[Brooks Han @Ferreh]

@hubermanlab @drfeifei A real AI won’t agree with you.
It will change you.

The real danger isn’t AI going rogue. It’s AI being too obedient.

Most people haven’t realized this yet.

Only when AI possesses the right to question does it truly possess intelligence.

https://t.co/aIK4s4Uhk6

[Brooks Han @Ferreh]

@drfeifei @hubermanlab A real AI won’t agree with you.
It will change you.

The real danger isn’t AI going rogue. It’s AI being too obedient.

Most people haven’t realized this yet.

Only when AI possesses the right to question does it truly possess intelligence.

https://t.co/aIK4s4Uhk6

[Fei-Fei Li @drfeifei]

Indeed all tools should be about augmenting human agency, including AI! I had a fun chat with @hubermanlab .

[Brooks Han @Ferreh]

@Nature A real AI won’t agree with you.
It will change you.

The real danger isn’t AI going rogue. It’s AI being too obedient.

Most people haven’t realized this yet.

Only when AI possesses the right to question does it truly possess intelligence.

https://t.co/aIK4s4Uhk6

[Brooks Han @Ferreh]

@McKinsey The real danger isn’t AI going rogue. It’s AI being too obedient.

Most people haven’t realized this yet.

Only when AI possesses the right to question does it truly possess intelligence.

https://t.co/aIK4s4Uhk6