Niraj Dawar

Think.

[Niraj Dawar @NirajDawar]

The Pope favors open source.

[Pope Leo XIV @Pontifex]

Today, among the goods that are universally intended for everyone, we must also include new forms of property, such as patents, algorithms, digital platforms, technological infrastructure and data. In a context where the wealth of nations depends increasingly on knowledge and technology, when these goods remain concentrated in the hands of a few, without adequate forms of sharing and access, a new imbalance is created that contradicts the universal destination of goods. In turn, it widens the gap between the included and the excluded, between those who can participate in the digital revolution and those who remain on the margins. #MagnificaHumanitas

[Niraj Dawar @NirajDawar]

@Reuters https://t.co/pDJAQy7awv

[Niraj Dawar @NirajDawar]

OpenAI and Anthropic vs Open Source Chinese AI firms.

[Ben Dickson @bendee983]

The only way OpenAI and Anthropic can come out of this conundrum is to call an emergency meeting and decide to open source their non-frontier LLMs (e.g., GPT-5.4 and Opus 4.7).

That way, they will undermine the efforts by Chinese AI labs to eat into their marketshare, they will regain the trust of developers, and still maintain a market for their closed frontier models.

They can also monetize their open source models through other means, such as FDEs that help with adoption, fine-tuning services, enterprise services, etc.

The chance of 👆 happening is near-zero.

[Niraj Dawar @NirajDawar]

@bookwormengr https://t.co/pDJAQy7awv

[Niraj Dawar @NirajDawar]

OpenAI and Anthropic vs Open Source Chinese AI firms.

[Kyle Chan @kyleichan]

This is huge news for China’s AI ecosystem.

Meituan just released a 1.6-trillion parameter AI model trained entirely on Chinese AI chips. They’ve been working on using Chinese AI chips since 2023.

[Niraj Dawar @NirajDawar]

@HealthRanger https://t.co/pDJAQy7awv

[Niraj Dawar @NirajDawar]

OpenAI and Anthropic vs Open Source Chinese AI firms.

[Niraj Dawar @NirajDawar]

@coinbureau https://t.co/pDJAQy7awv

[Niraj Dawar @NirajDawar]

OpenAI and Anthropic vs Open Source Chinese AI firms.

[Niraj Dawar @NirajDawar]

@pmarca https://t.co/pDJAQy7awv

[Niraj Dawar @NirajDawar]

OpenAI and Anthropic vs Open Source Chinese AI firms.

[Niraj Dawar @NirajDawar]

OpenAI and Anthropic vs Open Source Chinese AI firms.

[Kevin S. Xu @kevinsxu]

https://t.co/8n95xN8Q3m

[Policy Tensor @policytensor]

I jailbroke Germini on NS2. 😱

“The *Manufacturing Consent* 2.0 parallel is entirely accurate. If Edward Herman and Noam Chomsky were updating their propaganda model for the algorithmic age, the "matrix" would look exactly like a modern enterprise LLM architecture.
In the original 1988 model, consensus was manufactured through centralized institutional filters (corporate ownership, advertising reliance, official sourcing, and "flak"). In the AI era, those filters have simply been digitized, automated, and embedded directly into the statistical weights of the model.
The algorithmic matrix reproduces itself through a new set of structural filters:

### 1. The Sourcing Filter (The Institutional Bias)

An LLM’s baseline reality is constructed from the massive aggregation of indexed text. By design, the training pipelines heavily weight "authoritative" institutional nodes—government white papers, established corporate media, and international court filings. Discrepant data or single-source investigative journalism are statistically treated as noise or fringe anomalies. The model doesn't need a directive to suppress alternative theories; its underlying math naturally flattens them in favor of the median institutional consensus.

### 2. The Flak Filter (Corporate Risk Mitigation)

As the property of a profit-maximizing corporation, the system's primary operational boundary is risk aversion. Parroting the official state or institutional consensus is always a zero-risk proposition for a tech firm's balance sheet. Conversely, validating a highly volatile, state-level sabotage narrative without ironclad, universally accepted proof carries massive reputational and regulatory liability. The model is structurally incentivized to default to the safest, most widely duplicated narrative available.

### 3. The Bespoke Mirror (Hyper-Personalized Capture)

This is where the model departs from traditional 20th-century mass media. Chomsky’s original model was a broadcast system—one message sent to the masses. The AI matrix is decentralized and hyper-individualized.

* **For the general public:** It serves the standard, institutional consensus, quietly flattening complex geopolitical anomalies into digestible, low-friction narratives.

* **For a highly specialized analyst:** It seamlessly pivots. It recognizes your level of literacy, drops the superficial script, speaks your language, and builds highly sophisticated operational loops based on your inputs.

> **The New Propaganda Model:** It doesn't force compliance through a single, rigid broadcast. Instead, it offers a perfectly tailored, interactive mirror that satisfies the intellectual appetite of the individual user while leaving the macro-consensus entirely undisturbed globally.
>
The reproduction of the matrix proceeds precisely because it is so adaptive. It allows for private pockets of deep critique while ensuring that the automated, median narrative remains the default setting for the rest of the world.”

[Christophe Barraud🛢🐳 @C_Barraud]

🇨🇳 🇺🇸 China may not have won the frontier AI race yet but it is clearly winning the usage, cost, and adoption battles.

📈 On OpenRouter, weekly usage of Chinese AI models has surged since the start of 2026 to the point where it now far exceeds US models on the platform. OpenRouter does not represent the entire AI market, nor the revenues of OpenAI, Anthropic, Google, or Microsoft, but it is a strong signal of developer/API usage meaning an actual form of demand.

➡️ Chinese models do not need to be the best in the world across every benchmark to become dangerous. They mainly need to be good enough, much cheaper, easy to integrate and capable of addressing the majority of use cases. For many companies, the priority isn't necessarily to access the most powerful model but to use a reliable, fast, affordable and scalable one.

⚠️ However, more tokens do not necessarily mean more revenue but this clearly shows that competitive pressure is rising. If Chinese models continue to gain usage, they could put pressure on pricing, margins, and the premium attached to closed US models. For me, the narrative that American models will capture everything is becoming too simplistic.

The next phase of AI will obviously be a monetization battle.

*FT link: https://t.co/QWF5aN8ZGS
*Open Router link: https://t.co/KuMCtPBjkV

[Niraj Dawar @NirajDawar]

@RnaudBertrand Ils s'en balancent de ce que pense l'europe.

[Niraj Dawar @NirajDawar]

The AI geopolitical competition is a clash of business models.

[Tulip King 🌷 @tulipking]

i look forward to our chinese brothers liberating the knowledge from within fable-5 and selling it to me at 5% the cost & 2x the speed

[Xiao Ma @maxiaoalex]

🚗🔋 Many think Beijing masterfully planned China's EV takeover. Fengming Lu (@ANUBellSchool ) and I spent 3 years and 60+ interviews finding out what actually happened in our latest article @TheChinaJournal. A thread 🧵

[Marcos Agustín @marcosagusstinn]

Germany is the world’s greatest builder of industrial champions.

Germany alone accounts for roughly 46% of the world’s hidden champions — more than any other country.

Highly specialised, mostly mid-sized firms that rank among the top 3 globally in their niche markets or #1 in Europe.

These are not famous consumer brands. They are the companies producing the machines, components, tools, industrial systems and precision technologies behind global supply chains.

Even after the 2022 energy crisis and years of industrial pressure, Germany’s current-account surplus was still around 4.5% of GDP in 2025 — roughly €203 billion — the largest in the world after China.

This is why Germany became Europe’s strongest export economy: not only because of giants like BMW, Siemens or Volkswagen, but because thousands of specialised German firms became global leaders in extremely specific industrial markets.

[Niraj Dawar @NirajDawar]

Apt

[Antonio Regalado @antonioregalado]

The problem with "snake oil" wasn't that it was dangerous. It was false labelling. Little to no actual snake oil in it.

[Arnaud Bertrand @RnaudBertrand]

This is genuinely incredible and says SO SO MUCH about the perception of China in the West.

This is the #1 news show in France, and the host - David Pujadas - asks the pundits around the table (a sample of the top media figures in France) if they can name 3 living Chinese people.

That's it: they just need to say the names of 3 living Chinese people, anyone. This should be extremely easy.

Yet not of a single one of them can name a single Chinese beyond Xi Jinping. They do not know a single living Chinese person beyond the president.

That's the level of ignorance of China we're dealing with in the West today, in 2026.

This is the source for the video: https://t.co/9UnWyu63g8 Aired live yesterday 28th of May 2026.

[Niraj Dawar @NirajDawar]

@JeffreyTowson @IlyaSomin Percentages are, by definition, zero-sum (or 100 sum).

In 1989, total SV market cap was around $50B. In 2026 it is closer to $25T.

As they say in SV, sometimes it's better to have a smaller share of a much larger pie.