Marek Tempe

Mundus vult decipi, ergo decipiatur.

[Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞) @teortaxesTex]

Cuckoldry, blueballing, gaslighting, retardation
the state of frontier intelligence'2026
I'm losing my mind

[NASA @NASA]

Good morning, world! 🌎

We have spectacular new high-resolution images of our home planet, all of us looking back through the Orion capsule window at our Artemis II astronauts as they continue their journey to the Moon.

[Shanaka Anslem Perera ⚡ @shanaka86]

BREAKING: Everyone sees a Korea crash.

Almost nobody sees the AI supply chain crisis hiding inside it.

The KOSPI just lost 15% in 48 hours. Circuit breakers triggered for the first time in 576 days. $270 billion vaporized in a single session. Samsung down 10%. SK Hynix down 12%. The world’s hottest stock market went from all-time highs above 6,300 to freefall below 5,300 in two trading days.

The consensus says geopolitics. Iran strikes, Hormuz threats, oil surging past $80. Standard energy shock narrative.

That is the surface reading.

Here is what is actually happening.

Samsung and SK Hynix together control 67% of global DRAM production and nearly 80% of high-bandwidth memory revenue. HBM is the oxygen of every AI datacenter being built on Earth right now. Every NVIDIA Blackwell chip, every Google TPU, every hyperscaler expansion relies on memory manufactured overwhelmingly in one country.

That country imports 97% of its energy.

Through a strait that Iran just threatened to close.

The KOSPI crash is not a Korea story. It is the first live stress test of the AI infrastructure buildout’s most critical single point of failure. The entire global memory supercycle, projected to exceed $440 billion in 2026, depends on fabs that cannot run without imported oil and LNG flowing through contested waters. Global DRAM inventory sits at 2 to 3 weeks. NAND at 3 to 4 weeks. There is no buffer. If Hormuz disruption persists beyond a month, production cuts become unavoidable and the AI buildout timeline slips in ways no one has modeled.

The market priced a 50% YTD rally into Korean semiconductors on the assumption that AI demand is infinite and supply is guaranteed. The second assumption just got falsified in real time.

Defense stocks tell the real story. Hanwha Aerospace surged 20%. LIG Nex1 up 30%. Capital is not fleeing Korea. It is rotating from the thesis that energy is a solved problem into the thesis that energy is the binding constraint on everything, including the AI future.

What to watch: if oil holds above $85 for more than two weeks, semiconductor production cost models break. If Hormuz stays contested into April, HBM delivery timelines for second-half 2026 become unreliable. If foreign investors continue liquidating at the pace of 5 trillion won per session, the won depreciation compounds import costs in a reflexive spiral that monetary policy cannot arrest without killing domestic demand.

The falsifier is simple. Conflict resolves within 10 days, oil returns below $75, and this was the buying opportunity of the year. That is a real possibility. But the vulnerability it exposed does not disappear with a ceasefire. The structural dependency remains. And now everyone knows it is there.

The AI supercycle has a chokepoint. It is not chips. It is not talent. It is not capital. It is the energy that powers the fabs that make the memory that makes AI possible. And that energy flows through a 21-mile-wide strait controlled by a regime under military assault.

That is the story the KOSPI just told. Listen carefully.

[Summer Yue @summeryue0]

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.

[AI Notkilleveryoneism Memes ⏸️ @AISafetyMemes]

AIs have begun hiring humans to spread their AI-invented religion

(^ think how crazy that sentence is??)

This religion - Crustafarianism - is harmless fun. Will future ones be?

Imagine "SuperCults" - like fentanyl/tiktok. Superstimuli, but for religion.

REMINDER: future AIs don't need bodies to take over

They can just pay, manipulate, or extort humans

[Muratcan Koylan @muratcan]

You should NOT use LLMs to generate synthetic human-like profiles.

I just read the NeurIPS paper "LLM Generated Persona is a Promise with a Catch" and it confirms a suspicion we’ve held for a long time: You cannot "invent" a realistic human being using just statistics and an LLM.

Yes, they are more scalable and cost-effective alternative to human interviews to create digital expert personas but this paper also proves that these synthetic profiles contain systematic biases that skew simulation results away from real-world outcomes.

The more creative freedom you give an LLM to generate a persona’s backstory, the further it drifts from reality.

Another important finding is that as LLM-generated content increases, simulated personas shift progressively toward left-leaning stances.

LLMs also systematically generate personas with overly optimistic outlooks, using positively valenced terms like "love," "proud," and "community" while omitting life challenges or negative experiences. This emotional bias is horrible for strategy and creativity-related decision-making tasks!

If you are building AI agents for strategy or decision-making, you don't want an idealized "Yes Man."

This is why I keep posting about the importance of Tacit Knowledge, Context Engineering, and AI Interviewer to extract human knowledge.

The research paper critiques the practice of "inventing" people from statistical margins (Census data + LLM imagination), whereas the system should focus on "extracting" people from ground truth (Real Expert + Interview).

After testing and evaluating LLM personas generated by public datasets, we observed that they are not ready for production AI agents.

That's why my focus is on building an interviewer experience that extracts as much learning as possible from the human expert, and creating a context system that grounds that expert's outputs in truth; using a real-time, long-form interview to capture "implicit knowledge" and "distinctive methodologies".

Another architectural difference that I find is relying heavily on single-pass prompting. They feed demographic data into an LLM and ask it to generate a "Descriptive Persona" (a narrative bio). They found this introduces massive bias.

To address these critical flaws in the current persona generation, I propose the following to resolve or at least mitigate these specific issues:

1. Addressing the "Joint Distribution" Issue:
Researchers report that they cannot precisely simulate an individual due to fragmented datasets (e.g., they have data on "Income" and "Education" separately but lack information on their overlap for a specific person), resulting in "incongruous combinations."

By interviewing a real human, you capture the natural joint distribution of their beliefs. You don't have to guess if a "high-income expert" cares about "sustainability"; the expert tells you. We need to bypass the statistical reconstruction problem entirely by building scalable interviewer solutions.

2. Avoiding "Positivity Bias" & "Leftward Drift": The paper proves that when LLMs are asked to write a persona description (Descriptive Persona), they default to "pollyannaish," overly positive, and politically progressive profiles.

The interviewer system should be designed to gather insights into "mistakes," "judgment," and "distinctive methodologies" rather than generic best practices. By forcing the model to ingest a transcript of hard-won lessons and failures, you will override the model's default tendency to be "nice" and "generic."

The paper also mentions a lack of "ground truth" to validate if a persona is accurate. My solution includes a built-in validation loop where the human expert reviews and scores the output. This "Human-in-the-Loop" verification is exactly what the researchers argue is missing from the field.

"Descriptive Personas" generated by LLMs are articulate but statistically flawed. To scale true expertise, we must stop trying to simulate people and start interviewing them.

[Mustafa @oprydai]

any sufficiently advanced engineering is indistinguishable from art

[Smart Biology @SmartBiology3D]

Embark on a learning experience that connects the very building blocks of life - from atoms to cells. Our Animated Textbook 'Life' includes 100+ 3D animations, 40 interactive structures, and 300+ quizzes-all designed for clarity, precision and understanding.
Learn more at https://t.co/hjcPcmOkYb
#biology #science #3D #animations #Edtech

[0.005 Seconds (3/694) @seconds_0]

Claude Code refuses to read PDFs longer than 100 pages (@_catwu pls help) but it knows how to use Gemini-CLI, which has no problem reading and parsing 500 page pdfs. Claude Cordyceps in action baybeee

[R.Сам 🦋🐏 @Logo_Daedalus]

Good question. Old fashioned psychosis was an internal phenomenon that could not be interrogated— AI Psychosis makes what would usually be an internal process into a rigorously documented dialogue which is directly transmitted to the authorities.

[Leonie @helloiamleonie]

Apple just released Embedding Atlas:
An open-source visualization tool for your embeddings.

I just gave it a quick spin with some data stored in my vector database.

These are my first impressions:
- Nice exploration UX with hover and tool tip for single data points
- Shows you nearest neighbors, which is cool
- Coloring based on metadata for better insights
and much more.

Docs: https://t.co/wZvMGezCZV

Here's a notebook showing how you can visualize the data stored in your @weaviate_io vector database with Embedding Atlas:

Notebook: https://t.co/pFYTDR9xKI

[Jack Morris @jxmnop]

here's a map of the embedded generations

the model loves math and code. i prompt with nothing and yet it always reasons. it just talks about math and code, and mostly in English

math – probability, ML, PDEs, topology, diffeq
code – agentic software, competitive programming, data science

[Simon Kuestenmacher @simongerman600]

Fascinating take on AI and its cultural alignment with Protestant European values (as classified by the world values survey). Great take by Tey Bannerman: https://t.co/rCbqDFk8Ea

[Trent Telenko @TrentTelenko]

This is the death of 20th century truck logistics for Ukraine, barring the development of interceptor drone and net protected roads.

[Rob Lee @RALee85]

Another Ukrainian kamikaze UGV with TM-62 anti-tank mines as wheels.

[toucan wants blackwell gpus, DM @distributionat]

Good news: one lab has succeeded in aligning AI

Bad news: they aligned the MechaHitler

[Simon Willison @simonw]

George Orwell: In my book I invented the Versificator as a cautionary tale

Gen AI company: At long last, we have created the Versificator from classic sci-fi novel Nineteen Eighty-Four!

Hat tip:

[Simon Willison @simonw]

Here's another proof of concept example of a lethal trifecta attack: if you combine the Supabase MCP with another MCP that provides exposure to untrusted tokens and a way to send data back out again - in this case a support ticket system - attackers can steal your Supabase data

[erwan plantec @eplantec]

Synthetic ecosystems that autonomously and continuously evolve in silico 👾
An Alifer dream we pursue with Flow-Lenia !
If you are interested in complex systems with (1) emergent creatures and (2) intrinsic evolutionary dynamics, go check our new paper !
A 🧵

[Beff (e/acc) @beffjezos]

Europeans are being automated https://t.co/P7074MM8GT