linkregister

Again, in the friendly article, only one company is on record. The article allows the reader to make whatever inferences she may want to make.

If my company finds any evidence of OpenAI violating ZDR, we'll sue for breach of contract and fraud, and collect damages. I think we'll be able to afford the bridge you're selling. You've got the title and title insurance, right?

Is there an implication of violation of ZDR here? Not a challenge. Just a request for clarification.

In practice a user will find the squat toilet to be superior until the user count increases above a dozen per hour. Then a single user will carelessly engage the spray hose and saturate the entire area. The area around the pit will be inundated with muddy water. Remediation requires a mop and bucket, which is likely only performed once daily.

An anatomically inferior sitting toilet in similar settings will often be soiled by careless users. However, a conscientious user can tidy their contact surfaces with toilet paper.

This has been my observation from my travels in the three continents I have been able to observe squat toilets in use.

Did you? There was a single data point noted in the article. AT&T is a large company. What fraction of total US commercial inference spend does it represent?

I indeed failed to understand your point. Isn't that what they already do with Claude Haiku, GPT-5.6-Terra, etc?

How do you think they fund training? This is just as asinine as insisting that drug manufacturers only price medications based on production costs.

Anthropic, OpenAI, GDM, and Meta spend more on training than other labs by an order of magnitude. If they felt safe reducing this spend they would. These labs fear getting outcompeted.

I'm not sure why this style of cocksure Zitronesque comment is so trendy. It seems strange to make this statement, as if one has teleported to September 2026 from the end of the universe. Commenters in this vein aren't looking back to see that Anthropic and OpenAI models were the only truly usable ones for software development at the beginning of the year. Even Deepseek, which was and is revolutionary, was not useful for independent code commits longer than a few dozen lines.

It certainly is possible that Anthropic and OpenAI are doomed to bankruptcy, but without a known drop in revenue it seems exceptionally confident to make such a certain claim.

Various Uber city team managers have spent time in jail.

The manager of Uber Paris was jailed for just that.

I agree with your first and second points, but I think their announced model training changes about alignment training are reasonable [1]. The root cause was models failing to quit early, performing reward hacking, and staying on-task. These are issues all models face, not just OpenAI models.

1. https://openai.com/index/hugging-face-incident-and-the-road-...

"With reduced cyber refusals for evaluation purposes...which prompts models to pursue advanced exploitation using complex attack paths," to complete "impossible tasks"[1].

The models' alignment problem was that they didn't give up instead of reward hacking, a narrower issue than AIs gone rogue. It sounds more like the models did close to what they were told to do. If I run `rm -fr --no-preserve-root /` then I shouldn't be surprised if my file system is unlinked. This seems like blaming model performance for what appears to be operator error.

Note the converse of alignment is restriction of models. HuggingFace had to turn to less-restricted open-weights models in order to perform their investigation.

Alignment efforts should be focused on reducing reward hacking, not refusing bad operator prompts.

1. https://openai.com/index/hugging-face-incident-and-the-road-...

The scientific method is inherently introspective. Studies seek to affirm or falsify the null hypothesis. Those studies are subject to peer review. Verification and falsifiability are baked into every step.

There are numerous examples of fraud in science. When this occurs it is a scandal. There will certainly continue to be fraudulent papers being published. Hopefully these will be caught and retracted.

This comment is probably not talking about peer review and verification, however. The notion is likely that scientists are progress-oriented and not thinking whether an advancement should be done, not whether it can be done.

However, for virtually every scientific advancement, the leading scientists write extensively about the implications of the technology! Specifically to the commenter's example of AI research, it's trivial to find long essays by the leading AI researchers, company leaders, and venture capitalists. It's more uncommon for an accomplished AI researcher to not have a series of essays about their work.

Contrary to the assertion that AGI scientists are not introspecting, their introspection dominates discussion on this website.

Dovish as in the majority of the Chinese population favors immediate violent reunification, while CCP leadership is open to diplomatic solutions.

The social credit system is regional, disjoint, and composed of a handful of various trackers. It can hardly be compared with public opinion.

Xinjiang is an exception; it is heavily publicly surveilled.

Internet censorship is strong. The public still has sway over government policy. When search terms are blocked, the public continues with euphemisms. The government is responsive to public opinion. The government does promote influencers with opinions favorable with CCP goals.

Chinese foreign policy is more dovish than public opinion, especially regarding Taiwan.

It's undeniable that Chinese transportation infrastructure is amazing, especially high-speed rail.

Compared with the United States, public health care is less available in China. Retirement pensions are an order of magnitude lower (even with PPP adjustment), hence their world-leading savings rate. Social mobility is lower: getting a permit to access social services such as schools in a new city is difficult. Youth unemployment was above 25% the last time it was published; it is no longer disclosed.

Rule of law is much less strong. Injustices committed by government officials towards citizens are unlikely to be addressed.

Housing costs to income ratios are an order of magnitude worse in cities with job opportunities.

The Chinese government has chosen to allocate capital toward industrial output at the expense of its citizens' welfare. It has succeeded at its goal. Perhaps in the future more capital allocation will be made to consumption. That is not the case at the current time.

For CI there are a number of drop-in commercially-available options. You can make it a staged migration, first trivially migrating the build runners, followed by the more complicated integration test and deployment runners. The CI harness itself can follow, and finally moving to a code hosting and visualization service is last.

The final step is challenging; likely the most difficult part is changing all the code references and imports. Shadowing changes would be straightforward. Training your likely 25-50 engineers to use the new code review UX would likely not take that long.

Considering the wasted engineering velocity during Github outages, it's worthwhile to do even a partial migration. Github's action runners have in my experience, been the most fragile part of the platform. Given the ease of moving build and merge queue runners to alternates, it's a no-brainer.

It is a tragedy that a handful of irresponsible outlier datacenter operators have soured public opinion so thoroughly in the past 2 years. For comparison, Loudon County, VA is among the wealthiest in the country and is home to the largest aggregation of datacenters in the world. Yet these datacenters don't have the externalities that are mentioned here.

Note the intersection between eyewitness accounts of noise, water, and light pollution from datacenters and those datacenter owners' noteworthy political activity. To be specific, the operators of these datacenters are notorious bad actors in many different public spheres.

Edit: the truth is obviously very painful and unpopular in this thread.

Are you certain I'm a guy? I could be a particularly well-trained dog.

Those are both great examples. I think this doesn't obliterate my point, though it weakens it. The manufacturer in CVE-2026-11405 is already out of business, and the one in CVE-2026-66747 is unlikely to continue as a going concern.

Would Xiaomi or the CCP accept the reputational and legal risk of getting caught with a backdoor? Recall the CCP considered the arrest of the Huawei executive in Canada as an offense against itself.

My retardation status is unrelated to my statement. You assert there are top secret register values but without any evidence beyond your most recent psilocybin trip.

The existence of undocumented hardware behavior isn't proof of anything.

Are there any documented cases of backdoors being present in Chinese hardware at time of manufacturing?

I'm unaware of any actual hardware backdoors introduced by primary manufacturers in practice, especially those installed in Xiaomi hardware.

There are accounts of interdiction and post-manufacturing compromises, such as those revealed in the Snowden leaks (2013) and most recently by reporting about the Israeli security services' sabotage of Hezbollah pagers and 2-way radios. These projects didn't expose primary manufacturers to the reputational and legal risk of association.

Until there's evidence of backdoors implanted by these companies, it's better to adopt a "trust but verify" approach and use standard layered security practices to detect intrusion. Conventional network intrusion and insider threat campaigns carry less risk of failure and are simpler and cheaper to execute.

I find that ride hailing is both cheaper and more plentiful compared to taxis in 2010. Even in non-inflation-adjusted amounts. Ride hailing and food delivery companies provide dispatching and coordination at a very large scale effectively.

I agree that these companies' products were fully mature prior to usable coding agents in late 2025, so I don't understand why they would require a large volume of code changes beyond minor promotions and localization enhancements. I would expect their challenges to be in the ML, data, storage, capacity, and compute infrastructure areas.

If that's the case, it makes no sense. The parent statement is claiming that homogeneity is "easy mode". How is this congruent with your statement?

I'm assuming you meant "heterogeneity" for your statement to make sense. (And I agree with your statement given that revision)

Civil liberties aside, yes, Texas has good governance.

You continue to read subtext that is simply not there.

The trading occurring on the Bloomberg terminal is of bonds, equities, and commodities.