overfeed

A fear of competition causing a failure to recoup the trillions invested in AI via sky-high margins, and starting a (short) chain-reaction that causes the bubble to pop (or fizzle). A lot of people have a lot riding on the AI bubble not popping.

But the more interesting question is whether labs like Moonshot can be a real competitor to OpenAI/Anthropic.

The answer depends on whether you think the AI researchers at Chinese labs are (or can be) as smart, motivated, and as good at math as those working at US labs - a not-insignificant proportion of whom are Chinese nationals.

Strives to minimize =/= completely eliminates

Math strives to minimize ambiguity, which other fields don't do as much. Non-math fields tend to reuse regular words as jargon (i.e. with specificity of meaning that may fly over the laymen's heads). Social sciences and humanities are most notorious for this, often resulting in non-practitioners not realizing they are out of their depth because they are not looking at symbols from non-Roman alphabets.

Hugging Face’s security team and agents detected and stopped the activity on their infrastructure and had already begun containment.

Won't even name the model that successfully mounted the defense, huh? Fortunately, Hugging Face has publicly identified GLM 5.2 as the foil against OpenAI's next-gen frontier model's offensive-capabilities.

This announcement feels like rearguard action against a successfully deployed self-hosted open-weight model, and Hugging Face's original recommendations to have an open-weight model you control on standby before an incident.

...when is fast a good tradeoff for quality?

When it is cheaper, and the "lower quality" model is adequate for the task at hand.

Plenty of problems have a low(er) skill/intelligence floor, anyone who uses the dual-mode agent paradigm (plan, then act) figures out the second phase can be completed by a less capable model. Even when disregarding costs - speed is important here because the agent can rapidly iterate without human supervision, based on compiler errors, lint and test failures

So that's just an appeal to authority (longevity?) not anchored in reality. Just because you've been doing something for longer doesn't mean you're the best at it; Google has been shipping AI/ML models years before the founding of OpenAI and Anthropic, but it's playing catch-up on LLMs.

That the author doesn't acknowledge the relentless R&D efforts DeepSeek has been plowing into optimization, and giving a default win to OpenAI/Anthropic on the supposition that they've been serving models for longer is a black mark against the article.

I appreciate the transparency in explicitly stating their motivation for writing the article (a response to what the author saw as an overreaction to Chinese models), but I feel the article goes too far the other direction, with multiple unsupported leaps of logic, and overstating the stickiness of AI client products.

[Anthropic/OpenAI] are serving models at a particular capability level for months before their competitors, and are simultaneously applying the best models to optimizing those costs. Second, intelligence isn’t in fact a perfect commodity, in part because applied intelligence makes itself smarter

Is he casually assuming a singularity has already happened? A regular first-mover advantage I can understand, but those have been squandered or lost many times before.

But even in China, these AI centric hardware is not cheap

Definitely not cheap for individuals, but well within SME territory. There are countless small-town, family-owned businesses that had higher startup costs than a hypothetical Kimi-R-Us, Inc.

Chinese labs are doing a favor to the world. But I can also say with 100% certainty that if US labs were to close shops next year, Chinese labs would immediately start charging $$.

How would do you explain the pricing of Chinese solar panels, after managing to destroy other countries' solar industries? The prices are still dropping per watt.

Could it be China's internal demand for solar is big enough, and its long-term governmental strategy on renewables result in an outcome that almost looks like largesse to the rest of the world? I suspect Chinese AI may follow a similar path.

Perhaps, but we can agree that joining the Army goes beyond that. If I were in charge, I'd call the entire Silicon Valley Detachment 201 officer corp to to active duty at some remote desert outpost.

How does yoinking outputs from from prior generation Claude model and post raining on them result in a model competitive with the latest generation? That doesn't add up - nevermind Anthropic hasbeen summarizing thinking tokens since January to counter distillation.

There is a lot of supposition going on your part and mine. IMO, Chinese labs are not dependent on OpenAI/Anthropic outputs; they definitely use the outputs, but along other training/post-training data.

Now that Anthropic hides the real thinking tokens in a way that precludes future CoT distillation, we'll find out which side is correct based on whether Chinese AI labs close the gap or not.

My bet is they'll close the gap; nothing about frontier AI is magic, once something is shown to be possible, experienced practitioners almost always figure out how to accomplish the same feat, though not always on the same way. This is why frontier US labs keep leapfrogging each other every few months.

3.4 million is the number of sessions Anthropic detected. The actual number of Claude sessions trained on is likely >100 million.

That's an increase of only a single order of magnitude, increasing my estimate of exfiltrated tokens from 0.05 to 0.15 trillion - a far cry from the 15 trillion required.

They are used for post-training

Possibly - it may be too much data for post-training, unless further curation was done. However, this is not distillation; you know it, I know it, Dario knows it, but "Distillation Attack" is a short, memorable, sciencey-sounding, political sound-bite with enough malevolence to be deployed on the floors of congress, or by the usual fear-mongering newstainment talking heads.

While it sounds like a lot, do you suppose 3.4 million sessions come even close to being sufficient to train a frontier model?

Assuming each session was 10,000 words each, that's 34 billion words; lets call it 50 billion tokens (0.05 trillion) unfairly pilfered from Claude. That left Moonshot needing to scrounge for the other 14.950 trillion training tokens required for a baseline frontier model.

We have to stop crying distillation, it’s getting embarrassing and at this point feels even a bit delusional.

It's a PR campaign - when they say its an "attack" they don't mean on Anthropic - but on America itself. What kind of American can let such a brazen attack go unanswered? At the very least, they ought to demand the dangerous, pinko, stolen models be banned in all 50 states, and pay whatever price demanded by the patriotic, freedom-loving, all-American AI labs that can never be accused of stealing.

The Coros watches are less than half the price, have 22 days of battery life in smartwatch mode

Garmin gets almost 30% more battery life in exchange of not being as fast (30 days)

I think you can only reach Garmin's stated time if you disable a lot of functionality.

Turning off always-on Pulse ox gets you there. Turning everything off except telling time gets you 2.5x the battery life (69-71 days)

The more expensive watches (Fenix) also have long battery lives: lasting up to a month on a battery that can fit in a watch. The processors still have to sip power.

[...]it's pretty much a no-brainer to rewrite C in Rust, or at least to attempt it.

Thank you for proving my point. To be clear: there are a lot of legitimate reasons to not rewrite a C projects in Rust, memory safety isn't a universal trump card that supercedes all other considerations in all projects; one has to be zealot to believe that.

I suspect the majority of Rust users are rational about when Rust is appropriate (or not), but they are not going around in random bug threads demanding rewrites of multi-decade projects.

My point is that whether there will be a crash or not is incredibly hard to predict. COVID did not come with a stock market crash

As someone who had early PUTs against the obvious industries (travel, hospitality) - what I didn't foresee was the insane amounts of government liquidity poured into the markets.

Or Demis realizes self-hosted models are a threat to frontier lab margins, and gating access to the US market is beneficial to the cartel based on whatever post-hoc arguments he can come up with.

Truly, is that Rust's fault that sloppers are targeting it?

The "Rewrite it in Rust!", all gas, no brakes zealotry predates LLM coding agents. When you dismiss the trade-offs with no consideration, and abide no questions about the possible cost of the "one true $X", you're practicing religion, not engineering. What has changed is the zealots no longer restricted to demanding that other people adapt their preferred language, but can now automate forking projects into their ideal language. I'm hoping the next shoe to drop is the realization that the language is a small part of a projects success, when it's time to maintain their forks.

AI replaces a single tower with millions of 5-over-1s[1]. The aggregate height, and speed of construction mind-boggling, but when each building is considered individually, not very impressive.

1. Perhaps with a handful of skyscrapers sprinkled in.

I didn't actually have any issues with a power failure it was just extremely icy.

Maybe ask someone who lost power - or lost a loved one - and find out if they are as sanguine about the icy conditions. People died. "It's not a big deal since I wasn't personally affected" isn't a scalable position, but it is a quintessentially Texan attitude.

People move towards places with growing economic opportunities.

Fleeting economic opportunities that don't survive the construction phase - a few months most. No one will move from New York to Texas for a chance to be one of the 17 long-term staff at a data center: no one wants to be a data center night man that badly.

Better for the extra natural gas to power data centers than to "accidentally" leak into the atmosphere

I'm guessing this is an easy win when you don't live or work near one, and not the one breathing the resulting particulates and nitrogen compounds.

Texas is the second-biggest state. Where are they going to make major connections to ...

Anywhere else in the transmission network that has excess generation capacity. If the biggest state benefits from interconnection, why wouldn't the second biggest benefit too?

Daily demand is cyclic, even with no additional capacity added, being connected to grids in other timezones that have different demand peak-times is a net win.

Using solar to cater to demand increase from data centers will compound the issue.

The 2021 grid failure was triggered by a winter storm which had 3 effects: pushed up energy demand for heating, kneecapped solar generation, and caused failures in bringing up backup sources online (because operators didn't invest enough in "winterizing" their equipment)

The previous failure was in 2011, and didn't happen again until a decade later. The root cause hasn't been resolved since 2021: the demand and weather conditions simply haven't recurred since.