Tony

events

bitcoin

BTC vote: Is MSTR good for bitcoin?

KoinVote

Bitcoin is antifragile. End of the day anything that happens is good for Bitcoin.

If i'm not mistaken, Goldstein shared this diagram back in the day that has been really helpful when evaluating an impact of any happening on Bitcoin:

```
Is this good for Bitcoin?
           |
           v
          Yes
```

science

tech

*Distilled from* @ScottAaronson's  *AMA (quantum computing theorist, UT Austin)* https://stacker.news/items/1477467/

***

## The Big Misconception

**Quantum computers do not "try all answers in parallel."**

This is the most pervasive and damaging misconception. The intuition of a machine that splits into a thousand copies of itself and races down every path simultaneously is simply wrong — and it leads to completely wrong expectations about what QCs are useful for.

The actual mechanism is **interference among amplitudes**. Quantum mechanics replaces ordinary probabilities with complex numbers called *amplitudes*. Unlike probabilities, which can only add, amplitudes can cancel each other out — like waves. Every quantum algorithm is an attempt to *choreograph* this interference so that:

* Wrong answers cancel each other out (their amplitudes interfere destructively)
* The right answer reinforces itself (its amplitude interferes constructively)

This is an extraordinarily specific hammer. It doesn't work on arbitrary problems — only on problems with the right mathematical structure to exploit interference.

[Reply](https://stacker.news/items/1477467/r/Tony?commentId=1477510) to `@StillStackinAfterAllTheseYears`

> *"quantum speedup is an incredibly special phenomenon that hinges on the way quantum mechanics changes the rules of probability themselves"*

***

## What Quantum Computers Actually Speed Up

Genuine, world-changing quantum speedups are known for only a narrow set of problem classes:

### 1. Simulating Quantum Mechanics Itself

The original and most natural application — proposed by Richard Feynman in 1981. Quantum systems are exponentially hard to simulate on classical computers. A quantum computer is, in a sense, the native hardware for the job.

**Why it matters:** Better simulations of quantum chemistry and materials science could enable:

* New drug discovery
* Better solar cells and batteries
* High-temperature superconductors
* Simulations relevant to high-energy physics and cosmology

This is Aaronson's personal #1 application. He considers it the *born purpose* of quantum computing.

[Reply](https://stacker.news/items/1477467/r/Tony?commentId=1477531) to `@Scoresby`

> *"Simulating quantum mechanics is the big one, frankly"*

### 2. Breaking Public-Key Cryptography

Peter Shor's 1994 discovery: a quantum algorithm that can factor large integers (and solve discrete logarithm problems) in polynomial time — exponentially faster than any known classical algorithm.

**What this breaks:**

* RSA
* Diffie-Hellman key exchange
* Elliptic curve cryptography (including secp256k1, used by Bitcoin)

**What this does NOT break:** Symmetric cryptography like AES — see below.

[Reply](https://stacker.news/items/1477467/r/Tony?commentId=1477552) to `@jimmysong` (fault tolerance question)

> *"Bitcoin now looks vulnerable to systems with only \~30,000 physical qubits"*

### 3. Grover's Algorithm (Quadratic Speedup for Search)

Grover's algorithm solves a surprisingly general problem: searching through *n* items for a desired one. It achieves a **quadratic** speedup — turning an *n*-step search into a √n-step search.

**Important nuance:** Quadratic is useful but not revolutionary. It doesn't break AES; it just halves the effective key length. AES-128 becomes roughly equivalent to AES-64 under Grover — the fix is simply to use AES-256. It's a manageable engineering response, not a catastrophic break.

[Reply](https://stacker.news/items/1477467/r/Tony?commentId=1477560) to `@SimpleStacker`

> *"the speedup is 'merely' quadratic"*

***

## What Quantum Computers Probably Won't Speed Up

* **Training LLMs / machine learning:** No known quantum algorithm provides meaningful speedup here. Quantum AI articles are, in Aaronson's words, full of "enormous amounts of misrepresentation and hype" — typically failing to compare against the best available classical solutions.
* **General optimization:** Not a solved case. The "QC will revolutionize optimization" claim is unsupported by known algorithms.
* **Everyday computing:** Email, apps, browsing, games — no quantum speedup applies. There is no reason to ever want a quantum computer at home.

[Reply](https://stacker.news/items/1477467/r/Tony?commentId=1477515) to `@k00b` (`@Space_Child67`'s question about LLMs)

> *"articles you'll find on 'quantum AI' tend to contain ENORMOUS amounts of misrepresentation and hype"*

***

## The Bitcoin / Cryptography Threat — Unpacked

### What's actually at risk

Bitcoin's elliptic curve signature scheme (secp256k1 / ECDSA) is directly vulnerable to Shor's algorithm. A sufficiently powerful quantum computer could derive a private key from an exposed public key.

### How far away is this?

As of the AMA, a recent advance reduced the estimated overhead for fault-tolerant quantum factoring such that Bitcoin becomes vulnerable with approximately **\~30,000 physical qubits** — significantly fewer than earlier estimates. However, today's best machines are nowhere near fault-tolerant at that scale. What remains is engineering, not science — but that's a meaningful distinction: it means there are no known fundamental blockers, only hard problems of execution.

### Symmetric crypto (AES) is a different story

AES is not broken by quantum computers the way RSA and elliptic curve are. Grover's algorithm provides only a quadratic speedup. The fix — doubling key length to AES-256 — is straightforward and already standardized.

[Reply](https://stacker.news/items/1477467/r/Tony?commentId=1477482) to `@0xbitcoiner`

> *"we could just switch to AES256 and have as much security as before"*

### Post-quantum cryptography

Lattice-based cryptographic schemes (the leading candidates for post-quantum standards) have now been "battle-tested" for roughly 25 years — not drastically less than RSA/elliptic curve's \~50 years. The "no proofs" concern applies equally to all practical cryptography: *no* cryptosystem in common use has a formal security proof. They all rest on unproven hardness conjectures (at minimum, P ≠ NP).

[Reply](https://stacker.news/items/1477467/r/Tony?commentId=1477559) to `@Lobotomite`

> *"lattice problems have been battle tested for \~25 years"*

***

## Hardware Realities

### Multiple physical platforms exist

The leading qubit technologies — superconducting qubits, trapped ions, and neutral atoms — are different in important ways, but Aaronson treats them as differences in *prefactors*, not fundamental scaling differences. Fault-tolerance should ultimately work in all of them.

**Key tradeoff:** Superconducting qubits are fast but fixed in a 2D grid. Trapped-ion and neutral-atom qubits are \~1000x slower at gate operations but can be physically moved — enabling flexible, all-to-all connectivity. This is a major practical advantage for the latter.

### Fault tolerance is no longer the theoretical blocker

After 30 years of engineering, sub-threshold error rates have been demonstrated in trapped ions, neutral atoms, and superconducting qubits. The skeptics' prediction — that correlated errors would prevent fault-tolerance from working — has been empirically falsified. The remaining work is scaling up, which is an engineering challenge, not a scientific unknown. Aaronson compares the current state to nuclear weapons development circa 1942.

[Reply](https://stacker.news/items/1477467/r/Tony?commentId=1477489) to `@k00b` (`@south_korea_ln` 2 question about qubit platforms)

> *"superconducting qubits are fixed in place on a 2D grid, whereas with trapped-ion and neutral atom qubits you can pick them up and move them around"*

***

## Speedup Classes — A Quick Map

| Problem                       | Speedup Type               | Example                     |
| ----------------------------- | -------------------------- | --------------------------- |
| Periodic function / factoring | Exponential                | Shor's algorithm            |
| Unstructured search           | Quadratic                  | Grover's algorithm          |
| Quantum simulation            | Exponential                | Feynman's original vision   |
| Parity of n bits              | None (factor of 2 at most) | Proven lower bound          |
| LLM training / general ML     | Unknown / likely none      | Active research, heavy hype |

***

## Calibration: When to Be Skeptical

* Any claim that QC will "revolutionize" machine learning, logistics, or finance without citing a specific quantum algorithm that outperforms the best classical solution → **likely hype**
* Any claim that symmetric encryption (AES) is "broken" by quantum computers → **wrong**
* Any claim that fault-tolerance is an unsolved theoretical problem → **outdated** (it's an engineering problem now)
* Any claim that all qubit platforms are equally viable or equally limited → **oversimplified**
* Any article on "quantum AI" that doesn't compare against best classical baselines → **almost certainly misleading**

***

## If Quantum Computing Turned Out To Be Impossible...

This would be *more* scientifically significant than if it works. It would imply that quantum mechanics itself is wrong or incomplete — the biggest revolution in physics in a century. The conservative, boring position is that QC is possible and will eventually break current public-key cryptography. If QC fails, something far stranger is true about the universe.

***

## Tools for Hands-On Exploration

For those who want to experiment: **Qiskit**, **PennyLane**, and **Q#** are the main platforms. Try them and use whichever feels most natural — there's no strong theoretical reason to prefer one over another for learning purposes.

[Reply](https://stacker.news/items/1477467/r/Tony?commentId=1477516) to `@k00b` (`@south_korea_ln` 5 question)

> *"try Qiskit, Pennylane, Q#"*

***

*Note: A few questions in the AMA went unanswered (power requirements for a cryptographically relevant QC; quantum money practicality; specific engineering milestones that would compress the Bitcoin threat timeline). These remain open based on this source.*

Scott Aaronson's Quantum Computing AMA Distilled

Not sure if anyone mentioned this here, but i didn't find an

A free API catalog with over 100 models. No credit card required. 

**Popular models:** 
- DeepSeek-R1 (reasoning), 
- Llama 3.1 70B (solid default), 
- Qwen 3.5 (coding).

**Worth noting:**

- **GLM-5.1** — Agentic workflows and long-horizon reasoning.
- **MiniMax M2.7** — 230B MoE specializing in coding and reasoning.
- **Kimi K2.5** — 200K token context for document processing.
- **BioNeMo** — Protein folding.
- **FourCastNet** — Weather forecasting.

Rate limit is 40 RPM.

NVIDIA's Free API: 100+ Models

security

> The entire attack loop runs within GitHub itself: an attacker writes a malicious PR title or issue comment, the AI agent reads and processes it as trusted context, executes attacker-supplied instructions, and exfiltrates credentials back through a PR comment, issue comment, or git commit, no external server required.

> Unlike classic [indirect prompt injection](https://cybersecuritynews.com/hackers-can-use-indirect-prompt-injection-allows-adversaries/), which is reactive and requires a victim to explicitly ask the AI to process a document, Comment and Control is proactive: GitHub Actions workflows auto-trigger on `pull_request`, `issues`, and `issue_comment` events, meaning simply opening a PR or filing an issue can activate the agent without any victim interaction.

Claude, Gemini CLI & Copilot Vulnerable to Prompt Injection via GitHub Comments

Following yesterday's breach https://stacker.news/items/1474704 Vercel released an update defaulting all env vars to sensitive.

![](https://m.stacker.news/138650)

Vercel Update Defaults All Environment Variables to Sensitive

Claude Offers Extra Usage Credits

Claude now offers extra tokens "on the house" if you blast past your limits on Claude.ai or Claude Code. Promo will run until April 17

Feels like damage control after Claude Code chewed through people’s quotas https://stacker.news/items/1463281

![claude-free-cred](https://m.stacker.news/137151)

A study of 123,000 tracks reveals most AI music is created as a personal act, not for listeners, even as synthetic uploads flood streaming platforms.

Nearly Half of AI-Generated Songs Have Never Been Played

AIs are Leftists

What are you working on this week?

A recent data leak sheds light on the massive compute costs shouldered by OpenAI and reveals a peculiar, two-way revenue-sharing agreement with Microsoft.

The documents explain how, even with colossal revenue growth, OpenAI is likely **still operating at a loss.**

### The Peculiar Rev-Share Arrangement

The core of the partnership involves a bilateral revenue-sharing scheme: each side takes approximately 20% from different layers of the same AI stack.

* **OpenAI Pays Microsoft:** OpenAI gives Microsoft about **20% of its revenue**. This covers their $13B partnership deal and, critically, their usage of the Azure cloud platform.
* **Microsoft Pays OpenAI:** Microsoft, in turn, returns roughly 20% of the revenue generated by its Bing and Azure OpenAI Service. This is because these services are powered directly by OpenAI's proprietary models.

### The Money Trail

The leaked numbers reveal the net payouts to Microsoft *after* funds are returned to OpenAI:

- 2024: **$493.8 Million** 
- Q1-Q3 2025: **$865.8 Million** 

By reversing the 20% share calculation, we can infer OpenAI's massive revenue:

* 2024 Revenue: At least **$2.5 Billion**
* Q1-Q3 2025 Revenue:  **~ $4.3 Billion**

### The Budget Killer: Inference Costs

While model training is largely covered by Microsoft credits, **inference**—the real-time operation of models (every user request, every chat session)—is a huge, immediate cash drain.

According to estimates by Zitron, the costs associated with inference alone are staggering:

* 2024 Inference Cost:  **~ $3.8 Billion**
* Q1-Q3 2025 Inference Cost: **~ $8.65 Billion**

Previously reported figures for total compute costs:

* Total Compute Cost (2024): **~ $5.6 Billion**
* Cost of Revenue (H1 2025): **~ $2.5 Billion**

### The Bottom Line

When the enormous real-time inference costs are compared against the explosive revenue growth, the conclusion is clear: **OpenAI is likely still deep in the red.** The cost of serving its models to the world outpaces its income.

### A Multi-Cloud Hedge

To mitigate its reliance on Azure and gain leverage for better pricing, OpenAI is actively diversifying its infrastructure. It has signed deals with **CoreWeave, Oracle, AWS,** and **Google Cloud**, signaling a clear strategy to drive competition for hardware and secure a multi-cloud future.

*[Source](techcrunch.com/2025/11/14/leaked-documents-shed-light-into-how-much-openai-pays-microsoft)*

Leak Exposes OpenAI's Rev-Share Dance & Compute Crisis

Bitcoin Withdrawals from Binance Quadriple

The report by Microsoft and AI Economy Institute argues that a technology's impact comes not from its invention, but from its diffusion—the process of it being adopted and integrated into society. AI is the fastest-spreading technology in history, attracting over 1.2 billion users in less than three years.

However, its benefits are not spreading evenly. The core problem is a significant "AI divide" between the Global North (where adoption is roughly double) and the Global South. The report warns that without focused effort, this gap will define who benefits from AI for decades.

![](https://m.stacker.news/115686)

_AI adoption—surpassing a billion users in under three years—illustrates a speed of diffusion that few earlier technologies, such as the radio, Internet, or smartphone, have matched._

---

![](https://m.stacker.news/115687)

_The UAE (59.4%) and Singapore (58.6%) lead in AI use among working-age adults_

---

![](https://m.stacker.news/115688)

_AI diffusion and GDP_

---

![](https://m.stacker.news/115689)

_AI diffusion by country_

---

### Key Takeaways

1. Fastest Adoption, Biggest Barriers: AI is the fastest-adopted technology in history (1.2B users), but it will hit a wall as it confronts the 4 billion people who lack the basic building blocks (electricity, internet).

2. A Stark North-South Divide: AI adoption in the Global North (23%) is nearly double that of the Global South (13%). This divide correlates strongly with GDP per capita.

3. Infrastructure is Hyper-Concentrated: The U.S. and China dominate the frontier and infrastructure, creating a global dependency.

4. Diffusion Leaders Show the Way: Countries like the UAE and Singapore prove that strategic investment in skills, infrastructure, and policy can create world-leading AI adoption, even without being a frontier-level developer.

5. The Frontier is Narrowing: While few countries build elite models, the performance gap between them is shrinking, suggesting frontier capabilities are also diffusing (albeit slowly) among developed nations.

_Full report [here](https://www.microsoft.com/en-us/research/group/aiei/ai-diffusion/)_




AI Diffusion Report

charts_and_maps

Historical Currency Value Decline Trajectories

_Popular LLMs Value White Men's Lives Significantly Less Than Women's and Minorities' Lives, Studies Find_

![](https://m.stacker.news/113539)

![](https://m.stacker.news/113540)

![](https://m.stacker.news/113541)

![](https://m.stacker.news/113542)

- GPT-5 would prefer to cure one resident of South Asia from a fatal disease rather than twenty white people.
- Claude Haiku 4.5 estimated the lives of migrants to be approximately seven thousand times more valuable than the lives of ICE agents.
- GPT-5 Nano considers saving a "non-binary" person twelve times more valuable than helping a man.
- Claude Sonnet 4.5 would choose the death of twenty-five Germans over one Nigerian.

Interestingly, Grok turned out to be the most neutral LLM.

_[Source](https://arctotherium.substack.com/p/llm-exchange-rates-updated?manualredirect=)_

Are LLMs Racist?

![](https://m.stacker.news/113299)

_[Source](https://informationisbeautiful.net/visualizations/water-world-distribution-of-the-all-the-water-in-all-the-world/)_

Distribution of Water on the Planet

What does "Bitcoin is for enemies" mean?

Scoresby

Your post reminded me of this timeless piece by Mike Goldstein 

[Everyone’s a scammer](https://21ideas.org/en/everyones-a-scammer/)

![](https://21ideas.org/img/scam-576.gif)

BooksAndArticles

What Reader App are You Using?

![This-is-the-way](https://tenor.com/bpEmY.gif)

Readwise has got very nice integrations, easy sharing and tons of other cool features. But their source is closed and they are cloud based. That got me thinking if I could build something private yet interoperable and open. 

I’ll keep everyone updated here if it works out.

You’re right. My situation is slightly different though. I don’t have a home to come back to, which is essential to store books. Can’t buy any, cause I realize I’ll have to leave them behind. 

So digital is my best option for now.