They have forgotten that they also live in the very same physical world as the poor. Most of them don't even live that far from the poor.

Perhaps they don't consider the implications that nobody can fight a mob.

If they were as smart as they want people to think, they would think about these things and perhaps shut up (at least in public!) about making things worse for the poor.

Prior generations understood more clearly that the physical security that underwrites great financial successes is all underwritten in blood. A better heuristic is everyone gets the number of votes equivalent to how many of their lives they could give in a conflict.

Expanding on that...you are fully capable of writing e.g. product instructions, or marketing copy, or religious verse, or a poem, or a fictional quote from a fictional character in your upcoming novel. I would consider it unlikely that any LLM could identify you as the writer of any of those (though Claude might infer it was you given your chat history). The point is that Pangram claims to be able to do exactly that!

Pangram is also making the claim that they can, to a high degree of certainty, identify when someone else wrote text and claimed it to be yours. This, when Pangram has not ever seen a writing sample of either writer. This is quite obviously ridiculous!

It cannot. this is a misconception. A human being is fully capable of writing 200 words that Claude will identify as not being written by them, because human beings are far more complex than Claude. A human can even choose to deliberately write in the style of a different human, even one who does not exist.

Sometimes people are writing instruction manuals; those are not written like their professional emails, which are not written like their personal emails. It is normal for people to be able to write in different voices/styles/etc. People code switch, people write for different audiences, people change over time, people are hurried or tired or sick, etc.

So no, an LLM cannot identify you uniquely in 200 words. But more to the point, most human communication is not in training sets. And Pangram has no way of course-correcting on the vast amount of data that is not in its training sets.

By comparison: the autonomous vehicle companies actually do need their products to verifiably work. So they also feed back human-analyzed data from real trips into their models. They can tell the model where it was right or wrong in the real world. This is the part Pangram cannot do! Pangram deployed at a university may be used to accuse a student of cheating, but then Pangram will never know for sure whether the text in question was written by a human or machine. The feedback loop is missing a critical step!

LLM-generated text does not carry a watermark or other identifying marks. The "theory" is that an LLM trained on human writing, to mimic human writing, can be distinguished from actual human writing in under 100 words.

Notably the first diagram on the research overview page (https://www.pangram.com/research/how-it-works) shows feedback for "misclassified human examples." This is a category error; Pangram will not find out when it has misclassified text in the wild, except in rare cases. Only the "licensed human-written text" in its training data can be used as feedback.

Scams like Pangram also cause real harms, mostly because laypeople do not understand that what is being offered is not possible. Pangram advertises 99.98% accuracy, and they pitch it as a tool for teachers and universities. Translated: if a college like University of Alabama rolled this out, you could expect ~40 students to have their lives upended by this snake oil, every year. (And how can one even prove that an allegation is false, that they did write a given text?) And this is the best case, using the number on Pangram's homepage.

It is not clear why this is bad for Alibaba or China, could you expand on that?

> it’s bad to run a country like a business

The "Chinese miracle" resulting from China's integration with the global economy has lifted ~800 million people out of poverty and raised standards of living quite rapidly overall. A lot has been written on the topic, but suffice to say that it's a fairly unorthodox argument to make that the Chinese have not been well-served by the arrangement.

> level of freedom

The context of this discussion is whether the US government should prevent US citizens from accessing yet another product that is freely available in the rest of the world, to benefit a tiny handful of rich people in California.

Let me be clearer: if you do not hold stock in OpenAI or Anthropic, you likely stand to benefit from more AIs being provided by more providers. Imagine if the government had banned MySQL and PostgreSQL, so we all had to buy databases from Oracle and Microsoft. That would be strictly worse for everybody except employees of Oracle and Microsoft. This is exactly the same thing.

Nobody is making this argument because this is a normal thing that companies do. It's so common it is named in business strategy: "commoditize your complements."

There's a 2002 Joel On Software post explaining this using tech examples that were already old in 2002[1].

In the present, Meta is doing exactly the same thing with Ollama as Alibaba (who is funding some of the Chinese labs). Alibaba needs advanced models internally, and does not want to have a dependence on US models that can be arbitrarily shut off by Washington. Alibaba also wants to drive usage of its cloud among companies who are now demanding competitive AI models (which Alibaba may be locked out of providing). However, selling AI is not its core business. Putting cheap AI in the world creates demand for cloud services, which is part of Alibaba's core business.

Not only does this make financial sense for Alibaba, it is fairly basic.

Google gives away consumer software like Chrome and Android (not their core business) to drive search, which is their core business.

et cetera. The frontier labs are exposed in exactly the same has been way every pure play tech company before them. The US labs chose quality as their moat; remains to be seen whether that was a good choice, or if they can add another moat in time.

1 - https://www.joelonsoftware.com/2002/06/12/strategy-letter-v/

How would one even define that in 2026?

Yes?

Not remembering my economic theory here, but it is likely more efficient/expensive to simply have the federal government cut checks to our moribund industrial sector companies and let consumers benefit from modern technology.

Cut GM/Ford/Stellantis a $20B check each, let consumers save (conservatively) $200B annually on new car purchases + downstream benefits. Huge win for consumers & taxpayers.

If it's not worth subsidizing explicitly like this, then we also should not subsidize by banning Chinese imports, which also ensures US drivers have less access to modern vehicles. (And downstream ensures US auto designers are less likely to have had contact with modern vehicles, making it less likely that they will be able to design future generations well.)

In this case: resolve the theft claims against the US frontier labs, and only then let them make claims against third parties. It would be totally unreasonable for (say) OpenAI to extract a settlement from Moonshot and use that to pay its own claims. Ordering matters.

For scale, in much of the country it is possible to build a new home for well under $200/sf. New detached homes retail for <$175/sf, which includes the builder's profit.

Yes, this is extremely Netflix. The "fix" for this is to mandate that they can no longer post an ad for engineers, but rather they have to post separate ads for each engineering role. Maybe this is better, maybe not. If a person applies to L4, are they allowed to be hired at L3 pay? Will that confuse people?

> For small companies and single openings

levels.fyi chooses not to publish some comp datapoints because there's too little data, which means publishing the data is tantamount to publishing an individual's compensation.

I hear you on the goal, but the details are tough.

> mean(), median(), and std()

play this out a bit, and you'll quickly realize that there are infinite ways to game this. Something like levels.fyi only works because levels are a thing, and they are consistent across a relative handful of companies (most tech workers do not work at those companies).

And even levels.fyi won't publish salary data until there are enough data points at a company to prevent such publication from inadvertently publishing how much an individual person makes. This makes it less useful for (say) a senior looking for a job at a non-tech or small firm.

Google: Chromium, Kubernetes, Android, TensorFlow

Meta: React, PyTorch, Llama

Microsoft: VS Code, TypeScript, .NET Core

LinkedIn: Kafka

Slotted in along these, an analogous explanation is that Alibaba needs Qwen internally (vs depending on an American company), but licensing is not part of their revenue strategy. (As a cloud vendor, they can make money on inference. The strategy is very similar to the US hyperscalers ex-Google.)

Joel Spolsky wrote in depth about this notion of commoditizing one's complement in 2002[1] using tech examples stretching back into the '80s.

1 - https://www.joelonsoftware.com/2002/06/12/strategy-letter-v/

In the not-so-distant past, Uber's head of self-driving was indicted and sentenced to jail time for similar conduct. The criminal case didn't start until ~2.5 years after the civil case was filed.

Anchorage metro is ~15/sq. mile; Yuma, AZ is ~36. The Nashville metro is ~250.

Also, Starlink satellites spend ~70% of their time over the ocean. This will impact the utilization ratio of their gear and force them to launch still more satellites.

The same bosses will pay multiple security guards, in addition to staff, to guard <$10m in goods at a Walmart. But when 50x the goods are in the ocean, suddenly the staff is the limiter?