If this was the case, the Chinese government would have placed import restrictions on US chip manufacturers. Currently Chinese AI companies are constrained on compute. This is why they go to such great lengths to use Singaporean straw purchasers of NVIDIA cards. Merely having strong market incentives to succeed is not enough. Sustained, time-consuming R&D and technology sharing is necessary to catch up to competitors.
If this principle were true, then autarchies like North Korea would be the most advantaged since they never have to worry about international competition in their domestic market.
Your confidence is not backed by any data or observation, it's just an unfounded inference.
> the Chinese government would have placed import restrictions on US chip manufacturers.
OT but one thing I didn't realize until reading this comment:
Those used CMP 170HX 8GB were advertised in front of crypto server rigs. Those were probably spread across South Asian scam malls(shopping mall buildings built in literal sovereignty contested areas of non-China/Japan/Korea/Russia non-India non-African non-Muslim Asia, under guises of Chinese tourist destinations, that are literally rented out to local Chinese-speaking gangs engaging in all sorts of gang crimes from tech support scams to human and organ trafficking, straight out of the fictitious Chiba in Neuromancer), and they would happily sell or rent compute to Chinese AI firms if only those were more valuable than 8GB VRAM they have, which they all had all along.
GPU export to China must have become significantly less meaningful since the 170HX unlock became a thing. That alone must have provided petaflops of unsanctioned untraced junks for parts priced compute to Chinese AI startups.
> If this was the case, the Chinese government would have placed import restrictions
"About 600,000 of Nvidia's advanced artificial intelligence chips are stored in Hong Kong awaiting shipment to mainland China, two people familiar with the matter said, as Beijing withholds approval for large-scale imports despite demand from tech firms" [1].
> "If this was the case, the Chinese government would have placed import restrictions on US chip manufacturers."
I do not know full details, but China did exactly that. There were articles about limiting Nvidia chip imports more than a year ago. It really highlighted the absurdity of the export ban.
To my non-expert knowledge, the reasoning behind the export ban has never been explained beyond a vague assertion that China would do some sort of evil with AI if they have bigger GPUs than some arbitrary number of transistors, and also some silliness where paying a 15% tax made some of the GPUs safe to export. It has been pretty great for Huawei.
McKinsey has multiple consulting verticals, including mechanical engineering, design, employing real professionals for actual work in addition to their traditional business consulting and advising.
I think this comment was just a popularist repetition of discourse from social media. In other words it's a "tiktok ahh take".
I know a lot of people who work or have worked at Accenture. Is that close enough to McKinsey and far enough from social media? Because I have come to a similar conclusion after many long conversations about how the sausage is made.
GP is more cynical than I would have put it, but I think it's directionally correct; it's an easy way for the C-suite to demonstrate "due diligence" on any decision.
From my limited experience of working with McKinsey people, I would describe it as a Fortune 500 C-Suite incubator. The hire smart people, they are very well connected and the goal is to rise the ranks and get a CEO position in a multinational. I was told by a very senior ex-McKinsey consultant that their main role is to get people promoted (regional CEO to central board for example). The projects are essentially a vehicle for an individual to move up. Their network is phenomenal. I wasn't very impressed with what i saw productivity wise, but they are polished and the point is not to impress the lower downs.
Accenture are mostly a bunch of low level talentless grunts. There are a lot of them. If you're hiring Accenture then you are in trouble in more ways than one.
This is a bunch of regurgitated horseshit that business school grads repeat to each other like a game of telephone based off some PR puff pieces written about McKinsey in Fortune magazine back in the 1980s.
Pure strategy advice represents less than 20% of McKinsey billings (as publicly revealed last year).
Both firms are full of low level talentless grunts, essentially just fresh grads from different tier schools.
The only measurable difference between the two is that Accenture makes heavy use of geographic talent arbitrage by hiring heavily in India so they can do higher volume/lower-rate implementation gigs. The companies that spend the most with consulting firms are usually legacy dying companies and governments, not exactly a picture of competence (you think OpenAI or Anthropic are asking strategy advice from McKinsey?)
All professional services firms are commodity human time resellers, there is no special sauce or reality behind the 'brand narrative' behind any particular firm.
I will never understand the framing of visa-sponsored workers at Facebook as indentured servants. These are incredibly highly-paid jobs, easily in the top 5% of earners in the United States even for fairly junior roles. Even in California, these folks are in the top quintile of earners. These are people who could earn a third as much in India, Canada, or the Netherlands, and still be living exceptionally well. They have exercised more agency than the vast majority of Americans or even Bay Area residents.
That said, expecting individual engineers to understand the greater context of building an online addiction machine is probably too much. The product managers and data scientists surface concerns and executives decide. And executives should be held criminally responsible in the event of gross negligence.
All of the accolades you bestowed upon visa-sponsored workers could be true. But the framing of indentured servitude comes from the visa sponsorship itself. These workers do not feel they have the right to free expression or to speak out against unethical business practices (not just data privacy but even 9/9/6 work schedules), for fear of losing their sponsorship. That’s why “indentured” is used: they’re contracted to a specific employer. They don’t have freedom of movement in the labor market. This also creates labor price arbitrage, which employers use to pay visa workers less than permanent residents and citizens. This labor arbitrage inevitably lowers the wages of non-sponsored workers.
This is more or less an uninformed vibes-based response.
1. Few workers report a 996 work schedule at Meta or other Mag7 companies. This work schedule is reported by smaller AI startups and Chinese and Indian tech firms.
2. H1B and many other sponsored visas are transferrable. I have had several colleagues get laid off or interview for other jobs and get hired elsewhere. The L visa is non-transferable, though it represents a minority of visa sponsorships in the US.
3. You directly ignored my data about workers at Meta out-earning other workers in the Bay Area. My foreign colleagues earned the same as I did at virtually every stage of my career.
In general macroeconomics, yes, the introduction of foreign workers results in wages getting pulled down. This "lump of labor" only has temporary impacts. The overall outcome is an increase in the market for engineers, as immigrants go on to found companies. NVIDIA, the world's most valuable company, is one of these.
Reg. 2: What is the time window within which they need to secure the said contract once laid-off, failing which they'll be required to leave the country?
30 days for someone early in their career to find a new job in the bay after being fired for insubordination from Meta is not a bet most feel absolutely sure they can win, and the consequences are high. They may not be able to come back given the immigration climate right now.
It's disgusting and a downplay of atrocity of slavery. It's why we should stop allowing tech companies to import so many foreign workers if this is what they are okay with: openly destroying our society for money.
Expect the GOP to move in this direction as the democratic party starts going after big tech as well.
It's not surprising that guardrail triggers would degrade model performance. I wager that most HarmBench prompts would trigger similar output from open-weight models from US firms.
> Unlike our interventions for cybersecurity, biology and chemistry, and distillation attempts, these safeguards will not be visible to the user. Fable 5 will not fall back to a different model. Instead, the safeguards will limit effectiveness through methods such as prompt modification, steering vectors, or parameter-efficient fine-tuning (PEFT).
You're shifting the topic and hoping we won't realize it.
Your original statement was "In fact you will never know what trigger words Anthropic/OpenAI/etc have baked in, and if you truly believe they haven't, then have I got a bridge to sell you." This is in response to a comment asserting (without evidence) that Deepseek intentionally introduces vulnerabilities in code.
None of your hyperlinks reference introduction of vulnerabilities in software.
Indeed, it is a frustrating aspect of using hosted models that a ZDR contract must
be negotiated for anything sensitive. There are all sorts of tradeoffs.
You do realize more than one person can be talking to you on a forum right? What the hell was "my original statement"?
Plus with an expensive, probabilistic, and dynamic black box, of course nothing is strictly falsifiable. By your logic nothing can be said about it. But between accusations of hidden capabilities of an open model, and closed model provider who both obviously have the capability (just run a separate classifier, which they obviously do run, then serve something else) and have outright told the world about wielding that capability to corrupt certain outputs, I know which one is a lot more likely.
It's not a shallow dismissal. Writing quality is core to the value of an article.
It's likely that but for LLMs, the article would not have been written. I'm glad it was written. It's also true that it's unpleasant to read LLM-written prose for those of us who have to read it all day for work.
Do you think your comment is a good critical comment that "teaches us something"? If not, it qualifies as a shallow dismissal according to these guidelines, even if it's not a shallow dismissal according some other definition or criteria.
I don't think it's shallow because writing quality isn't important, I think it's a shallow dismissal of the writing quality because it doesn't discuss anything interesting or substantive about the writing quality.
Also, what about the other guidelines? Comments should meet all of the guidelines, not just some of them. (I do get that some of these are borderline.)
My comment? Please review the username of the comment in question.
If you feel a given comment is not within community guidelines you are free to flag it. I do not believe that we agree about the substantive value of noting an article was written with the heavy assistance of an LLM. I usually look at comments prior to reading an article; it's an effective time-saving measure.
I do not accept your attempt to make a blanket ban on noting an article was LLM-written.
My assumption is that every LLM-written article would just not have been written otherwise. That is a net good. It's also true that it's grating to read LLM-written prose.
Am I utterly mistaken or was LLM-written prose more tolerable when it was GPT-4o and Opus 3.x writing it?
Reminds me of when The Powers That Be forced me to connect to websites with HTTPS. Then they stopped supporting Internet Explorer 5. What were they trying to hide?
I did think about that, and the answer was fair enough, they don't want MITM. None of these details about passkeys make any sense. The author hasn't explained why this isn't a simple public-private keypair.
If this principle were true, then autarchies like North Korea would be the most advantaged since they never have to worry about international competition in their domestic market.
Your confidence is not backed by any data or observation, it's just an unfounded inference.
reply