Obviously, and they deserve no credit for the bare minimum. But if you imagine them doing the opposite and not taking any steps to rectify the error, that would clearly be closer to a "coverup".
> We opened Google’s Artemis repository and recognised code we had written for mobile-use. Then we found our names in its history—and the change that removed them.
Did you read the post? They originally removed all traces of minitap and its contributors, and only added it back when they were called out.
I haven't seen any proof that OpenAI asked Tristan to remove Sebastian from the prize. Until we have proof of this, it would be wise to offer conclusions.
Same for NS validity. This was not validated by the community yet.
It would be very difficult to say. It confirms that Tristan's data is likely part of the data the models use, but a lot of filtering, pruning, and transform goes into training.
Data has to be determined to be signal and not just noice, then it could go through processes of generating questions/answers from that data, then it RLHF's over this.
OpenAI have petabytes of data, all anonymized. It could take months to say for sure it was part of the training, and even more time to determine if it made any difference.
I worked in the tracing and tracking all the thousands of data sets that got tweaked and permuted and changed hands between thousands of researchers and data engineers at a major lab. The data that goes into training runs is permuted so much from the OG data that tracing the lineage is not trivial (dramatic understatement).
And the difficulty is harder than just the extreme scale of text searching. but also explodes with organizational difficulty since there are so many people tweaking/shifting data independently upstream of the actual training run, and no they will not all add the telemetry you wish they did.
In the ideal, should it be this hard? Well, no, but that's org wrangling for you.
It feels convenient to not spend time on engineering around tooling that could be used to answer a question like “did you violate copyright by training on X?”
Don't attribute to malice what is better explained by coordination headwinds in extremely large companies.
The engineering around tooling wasn't remotely the issue. It's getting all the (thousands?) data researchers mostly iterating on fine tuning datasets that would get bristly if they couldn't work outside version control in a python notebook iteratively tweaking their dataset that processed and reprocessed a few datasets until a threshold was reached.
The only _guaranteed_ chains of custody are down at the compute job and file read level. Which in a massively distributed computing job is... [redacted] nodes reading [redacted] fanouts of "datasets" that is just an abstraction over [redacted] individual files.
There's no malice here. Just way way way more complex than you'd first think.
The malice would be in not prioritizing the provenance tool at the start as a requirement of the rest of the product. Ethics would tell you that if you can't make the product in an ethical way, then you probably shouldn't make it.
It definitely is solvable though. Data versioning is a thing and it can work quite transparently to the mutations done on the data.
To not know who made and who approved a set of mutations on data can easily become equally as mind-blowingly stupid as not knowing who made mutations to code. Code is a subset of data after all and search over (provenance of) data can be implemented as DAG traversal.
Not tracking data changesets like code changesets is certainly a choice, not really a constraint anymore. A similar choice I feel is implied by "extreme scale of text searching".
> no they will not all add the telemetry you wish they did
...is just a failure of the corporate policy surrounding data handling. Is git-for-data already considered telemetry?
Of course the truth is provenance of data is something best institutionally forgotten as quickly as possible. The only thing that matters is it's there, that the data has no history, and that's why it can be used in whatever way deemed necessary.
All of your points are valid, and believe me I was trying to make them. The problem is one of culture. Most of the people doing this kind of work didn't like version control, and their work was really just running notebooks (like iPython or Google Colab) until a number was good enough and they'd submit the file for inclusion into training runs.
You can call it a policy failure, but these people were in very high talent demand and so top down dictates would risk "X people leaving lab Y for lab Z" headlines and morale hits.
I am not saying this is good. I am telling you that on the ground it is so much messier than it should be.
The appearance of heroic efforts to get authoritative lists of what datasets went into which major model versions prevented actual data laundering (up to intent and mistakes). But don't attribute malice to that which is far far easier to explain with coordination headwinds: https://komoroske.com/slime-mold/
Malice is not required, which is precisely why I added, "in effect". If the effect is the same as data laundering, that is reason enough to encourage the practice, regardless of the original motives (and I'm sure there are plenty of legitimate ones).
I think you may be underestimating how difficult a text search over their data is. They may have to build new mechanisms to do this. And what you really want is also an attribution of how much of a contribution a given corpus made which is a much harder question to answer; a single appearance of a chat probably has very little impact on the inference performance at this time unless it’s been explicitly preferenced somehow
I don't think anyone really cares about 'the measured impact the data had on the exact result' - a question which is fundamentally difficult to answer accurately in the first place - but rather whether the data was used in training at all - which as Tristan described, was extensive, beyond simply a 'single chat.'
Can you explain the difficulty in engineering a search apparatus over a corpus of text data? Actually searching through it may not be easy, sure, but it's work that's doable, and creating an index is relatively trivial.
> Can you explain the difficulty in engineering a search apparatus over a corpus of text data?
My guess: "If we ever imply that's possible, people might start asking questions about all the other work we've ripped off, so the official answer is that it's impossible".
If you don't trust the labs directly, you can always use AWS Bedrock or Azure Foundry which should have a much stronger incentive to not train. They make money from asset rental, not selling models. I'd be shocked if they were training.
This isn’t me advocating for maximum paranoia. I’m just saying that the “privacy” aspect of the API pricing is just a business promise. I do personal coding on API pricing with Claude mainly because a) I’m normally using it quite sparingly b) the commercial relationship is a lot more transparent than these subsidised relationships where they “may” be training on your work but can’t really seem to answer what that means.
They would lose all of TSMC if they do any military actions. It's not the fact that the CCP thinks that they can make Taiwan and China a country, it's the the fact that most of China does in fact consider them as one country. Taiwan is ready to defend themselves, but are very happy keeping their relationship with China the way it is.
"Very happy" is a bit of a stretch, since a significant proportion of Taiwan would prefer to jettison the current pretence of being "One China" and become fully independent.
But I agree that pretty much everybody will happily take the status quo over armed invasion.
I have a friend in Taiwan and he told me they're pretty happy because they have independence this way. If they go for more international recognition etc they'll gain nothing they don't already have but they'll upset the balance and give the CCP a reason to invade. So it will only make things worse.
And a lot of people are sympathetic with the Chinese, just not their government.
I've only visited Taiwan once and I only have one friend there so perhaps what I'm hearing is not accurate.
and related to, with all that holds in an Eastern mindset. My adoptive grandmother was born in Shanghai in 1927. To my knowledge she never returned to Mainland China, partially because her relatives wen Telswhere when the communists came in.
> give the CCP a reason to invade. So it will only make things worse.
CCP is not threatening taiwan because it might go for independence. it is _ALL_ about kicking the US out of the west Pacific. because of that, it is never about what people in Taiwan believe/say/do.
It is only about the military balance of the west Pacific.
> And a lot of people are sympathetic with the Chinese, just not their government.
sympathetic of what?
20 years ago, you might be able to play the "CCP regime repressing the people in China" joke, you can't do it anymore especially in such topic of Taiwan. quality of life is significantly better in literally every aspect in China. Taipei in 2026 is getting more and more like a tier-3 Chinese city in 2016.
> it is _ALL_ about kicking the US out of the west Pacific
I think you are underestimating Chinese nationalism here quite dramatically!
Chinese views of sovereignty are quite independent of other concerns. Take for example their actions against Vietnam in the South China Sea, forcing Vietnam closer to the Philippines and even the US: https://digital-commons.usnwc.edu/cgi/viewcontent.cgi?articl...
> it is _ALL_ about kicking the US out of the west Pacific.
Hardly. It is all about Xi having declared that taking Taiwan by... I think it was 2049, but could be misremembering. Anyway, Xi made that a test of the legitimacy of the CCP, in some very public statements.
So, no, it's not about kicking the US out of the west Pacific. It's about Taiwan, and it's because if they don't, the CCP is illegitimate, by Xi's own word.
By 2023 Xi ordered the Chinese military to be ready to invade Taiwan by 2027.
But that doesn't mean they plan to invade Taiwan by then. One common reading of it is that it was a plan to build up the military to that state with a definite goal date (in the same way the "600 ship navy" plan had a goal date).
The point is that the goal is far enough into the future that the person making it can't be held accountable for failing to meet it.
If Xi had said "the CCP is illegitimate unless it takes Taiwan by 2025", he would be at real risk of being replaced in 2026. Anyone who opposed him (or just wanted power) would have a perfect reason to replace him.
When Xi makes a goal with a "deadline" he won't live to see, he takes no risk of personally failing to meet the goal. If his successor misses the deadline, he can say things changed, Xi failed to consider factors X,Y,Z, etc.
> If his successor misses the deadline, he can say things changed, Xi failed to consider factors X,Y,Z, etc.
That's not how the CCP operates though.
They are not a country where the puppets matter so much, like in most of the western countries. You would very very rarely see a leader being critical of their predecessor even if they are changing policies (which Xi Jinping definitely did). Because they are the same party and thus they are attacking themselves. The party is really more important than whoever happens to be at the top. The party is always right. Even when the party is wrong. Which it is not because it's always right.
No, if this promise disappears it will be silently. Never to be mentioned again, or else. It just never happened.
> And the whole concept of bringing everything down to a matter of wealth is so American.
They said "quality of life". Even if they had said wealth that correlates with quality of life so that wouldn't be a very good gotcha by itself, but they didn't say that.
But calling Taipei a "Tier-3" city does refer to economic prowess, that's what I mean.
If you like to say what you think, then trust me, Taipei has much better quality of life. If you see people like Naomi Wu being arrested and gagged just for pointing out a vulnerability in software (which we all agree here is a good thing to do for everyone except the bad guys), that is terrible.
Not only a few people in and from Taiwan use every chance they get to shit talk the mainland. Be it snark about using simplified characters, or be it trying to prevent their own citizens from taking part in celebrations of victory over fascism. Main theme is being against the mainland, no matter what. They are also very influenced by Japan, whose right wing PM has escalated outrage more than enough recently.
Certainly, somewhat understandable, given the "military exercises" at the doorstep happening every now and then, but also escalatory and idiotically not a good look either. Some of it is just spewed hatred, not taking a high lane when it comes to ethics.
I think the plan is to defend, with allies motivated by the goal of saving TSMC facilities. If you destroy them the moment China declares war, Taiwanese allies might be much less invested
But at the same time everyone understands that the Taiwanese would destroy the facilities before they fall into Chinese hands
Not everyone is affected by propaganda, either US/Taiwanese or Chinese one.
Some are.
Some people also want to feed their families and not be unemployed.
P.S. A large number of Ukrainians moved to Russia. Including the ones who wanted to reunite with their relatives. Taiwanese have a lot of relatives on the mainland as well.
> *As of mid-June 2023 an estimated 1.2 million Ukrainians were recorded in the Russian Federation under different legal forms of stay (other than refugee or temporary asylum status)
This means people that have Ukrainian citizenship, which includes lots of families that were on both sides of the border. It does not mean people who were residents of Ukraine before the war, and decided to move to Russia. It does include the kidnapped children, and surviving people in occupied cities.
The US cannot offer lots of money to chip fab refugees today, because they are not refugees. Trying to headhunt people at a high quality job is very different.
Part of the reason a Chinese invasion of Taiwan seems more realistic is Trump being a sycophant towards any dictator who will give him the time of day, including Xi Jinping.
It's not impossible, but I feel as though if Trump decides that China is friends and Taiwan is enemies (as he seems to have done with North/South Korea lately) then Taiwanese refugees might not be as eager to move to the US as, say, Canada, Australia, or Europe.
We are talking about facilities that are extremely complex and expensive to build, to the point that only a very small handful of locations in the world exist today - despite their massive economic value.
TSMC is not something you can rebuild with some brains in front of laptops.
its not official public policy, but there have been people in the US government saying that if there is a war then the US will bomb tsmc facilities in taiwan
I think you're missing what I'm getting at. Yes, I understand that part. But do you seriously think that if TSMC was destroyed right now, China would change its decision to invade Taiwan? Can you imagine what the actual consequences of executing that plan would be?
My bet would be that invasion wouldn't look like quick war, but slow control of political power either through light force like economic blackmail, or through something similar to what US did with Venezuela.
Now that US has already opened the gate to labelling nation head as drug seller, this just eases things for China.
Yes and no. The Iran war is deeply unpopular and won’t even pass congressional approval for money. It’s being fought from pocket change budgets because they know they can’t pass a funding bill for it. Also not the same kind of war.
Any period time? US waged war for 20+ years in Afghanistan. Pretty effectively! To what end? Not an end an army can really achieve so… there you are.
But a war that might actually be popular? To save the world from communist dictators and make sure we have chips for our precious devices? I wouldn’t count out the United States. Or nuclear weapons, which still exist and are a thing.
That was an occupation and a low intensity “pacification” mission. Waging a prolonged war against a military peer on the other side of the continent is a bit more problematic (considering that US is/was on the brink of running out of advanced weaponry in Iran despite having aerial superiority which they won’t over Taiwan, Japan would have to get involved as well cause otherwise US would have to entirely rely on the fragile carrier fleet).
Using tactical nukes (even regardless of the extreme geopolitical downsides that might bring) wouldn’t necessarily give a great tactical advantage either, unless they are willing to do that over Taiwan itself to destroy the landing forces).
And it'd be the kind of war the US is best at. Boots on the ground, against conventional military, with a civilian population that is on the side of the US forces
No guerilla warfare, no insurgents, no need for extended beyond-visual-range warfare that depletes missile stockpiles
Last time used faced a conventional military that could actually challenge it was in Korea.
It won’t have complete areal superiority or entire division surrendering after running into contact with US forces like they in Iraq. “Shock and awe” of course can be very effective just probably not against China. Even now US is running out of materiel against Iran and can’t really sustain a prolonged intense conflict while being able to protect its gulf allies from ballistic missiles and drones.
Iran is not a war funded by congress, so it's really no wonder the US appears inneffective. Trump didn't get the support he thought he would from congress and now he's flailing around trying to save face.
Change the leadership (which will happen soon) and when there is actual support from congress to fund and intervene in a conflict in Taiwan, then you will see a vastly different US military response.
The war over Taiwan wouldn't be done by soldiers on Taiwanese soil. So you wouldn't need many boots on the ground in a literal sense.
But if you mean in a metaphorical sense, how long until the 7th Fleet could get to Taiwan, the answer is 0 days for first response, 3 days for everything located in Japan, and less than a week for the full mobilization of the 7th Fleet (that's 50+ ships and nearly 30K soldiers). The entire Pacific fleet could be there in a week and a half.
(Granted, we're assuming competent US leadership, which is not the current administration.)
just a gut feeling, but if they had to for some reason, I feel like they'd just go in and bomb as much strategic assets and invade, because in their eyes aside from Taiwan, there isn't much left to "unify" and it'd be historic, not something they'd need to do every decade or even century
I feel like Russia thought the same of Ukraine, even though it is way bigger
Any country that competently adopts Ukrainian drone technology cannot effectively be invaded currently. And that's for a continental invasion.
Taiwan? All surface ships are sitting ducks. Troop transport ships would get absolutely decimated, and then the landing troops would get swarmed by drones all over again, and then they have to maintain logistics to the small minority that land (good luck with that). And that's without naval drones.
With naval drones? I have no idea how China gets anything across the strait.
And all the above is true for a competently organized Chinese invasion. Which it won't be.
> adopts Ukrainian drone technology cannot effectively be invaded currently
The tactical situation in Taiwan is very different though. I think it would come down more to what’s happening in the air since the stalemate in Ukraine is in large because the Russian airforce is almost completely neutralized and can only be used to launch cruise missiles from a safe distance. I doubt China would invade unless they feel they can somehow achieve at least temporary air superiority and achieve their other objectives during that period.
Air defenses are probably much more important than drones overall.
That's why I said it will be slow. I think it will be decades long political control, not a military invasion.
China have more credible threat of bombing economically relevant building like TSMC, or capturing leaders and protestors like they already are doing in Hong Kong.
But nonetheless China is clearly preparing itself to be ready to be able to invade, with aircraft carriers, amphibious assault ships, dual purpose roll-on/roll-off carriers. It doesn't mean it will, but that it wants to be able to.
So basically it will depend on a single dictator's whim if they do or don't.
> Establish a total naval and air blockade of Taiwan using their fleets stationed in Ningbo and Zhanjiang
Which would be considered an act of war, so it's the same scenario again.
> There is a substantial opposition in Taiwan that want reunification (~20% of population support directly Li-Wun)
20% is more than the % of people in the Donbass that wanted unification with Russia, and we can see how well that went. And it wasn't a dangerous amphibious assault against an island with shit terrain.
Arguable, they had to after the 2008 Buharest NATO summit and Ukraine's militarization going forward, and eradication of Russian language and Russian-speaking population in the east.
> Which would be considered an act of war, so it's the same scenario again.
China considers Taiwan their territory, who is there to protect Taiwan? US?
>20% is more than the % of people in the Donbass that wanted unification with Russia
20% plus part of those white-collar workers who are a backbone of the Taiwan's economy and politics and can be left jobless after China achieves EUV and puts TSMC out of monopoly.
Let's be real. Trump demanded TSMC to invest 100B into US factories and they complied [0].
> Arguable, they had to after the 2008 Buharest NATO summit
How dare the countries threatened by Russia, who had seen how Russia invaded and brutalised multiple countries - Moldova, Checnya, Georgia twice - seek protection from Russia. This totally excuses Russia invading and brutalising again, of course.
> Ukraine's militarization going forward
The country was literally invaded (Crimea and Donbass), what the hell did you or the idiots in the Kremlin expect? That it will just sit and allow Russia to take more of its territory unopposed? That it will become a pliant vassal state again?
> eradication of Russian language and Russian-speaking population in the east
Such eradication that the country elected a Russian speaker to be president, on a platform of reconciliation with Russia. So much so, that his opposition called him a Russian puppet. That guy is Zelenskyy.
I'm sorry, but you're either out of your mind, or have been completely brainwashed. Russia chose to invade multiple of its neighbours, and did unspeakable atrocities every time. The only one to fault for all of this, and everyone around them seeking protection from this, is them. This imperialist nonsense brutality does not belong in the 21st century, and neither does its apologia. And no, I don't care about the whataboutism you're preparing to "defend" these actions with, it's beyond irrelevant. It's what the Russian state has been about since it came into being out of the ruins of the USSR. We should have never treated them as anything else.
> How dare the countries threatened by Russia, who had seen how Russia invaded and brutalised multiple countries - Moldova, Checnya, Georgia twice - seek protection from Russia. This totally excuses Russia invading and brutalising again, of course.
That's literally a copy-paste from propaganda pamphlet.
Can you back it up?
> The country was literally invaded
You seem to forget the amount of killings Ukraine did there and why it started in the first place.
Why should I discuss that with a person who seems to be unaware of facts?
What exactly do you want backed up? That countries across Eastern Europe saw how the russians operate and decided they don't want to be destroyed?
> You seem to forget the amount of killings Ukraine did there and why it started in the first place.
What killings did Ukraine make that resulted in russia invading Crimea? Or russia starting the invasion of the Donbass? MH17 maybe? Oh no, wait, that was the russians too.
All your points consider what China can do but not how others would respond. Blocking Taiwan when the whole AI economy is running on it will invoke lots of countries to go up in arms. There was a quote from US military that it'd cost less to fight than watch Taiwan fall (invasion or not).
> 4. China is close to producing industry-grade EUV
No where near close. At least not production grade at massive scale.
> * Establish a total naval and air blockade of Taiwan using their fleets stationed in Ningbo and Zhanjiang.
Doesn't work. If they do that the US, JP and others can also blockade around Shanghai, Hong Kong and the other ports.
> They won't need TSMC soon
Taiwan is not just TSMC. There are thousands to millions of things involving Taiwan you're likely not aware of and they might not even be AI or semi related.
> All your points consider what China can do but not how others would respond.
US backed off last year when there were talks about protecting Taiwan. They pulled Patriot missiles from Taiwan and South Korea too.
China and Japan have a beef. Considering the atrocities Japan did in China, the conversation won't be friendly.
None of these are good for Taiwan, during the times when its political system is failing.
> No where near close. At least not production grade at massive scale.
It looks like you have knowledge of the status of that super-secret government project. Can you share some details then?
> Doesn't work. If they do that the US, JP and others can also blockade around Shanghai, Hong Kong and the other ports.
US cannot blockade Hormuz, as we've seen. JP cannot do that with their current constitution unless they change it (the current government is on the road to it, though).
What we've also seen is that Chinese fishing fleet had a military blockade exercise [0].
> Taiwan is not just TSMC. There are thousands to millions of things involving Taiwan you're likely not aware of and they might not even be AI or semi related.
>US backed off last year when there were talks about protecting Taiwan. They pulled Patriot missiles from Taiwan and South Korea too.
>US cannot blockade Hormuz, as we've seen. JP cannot do that with their current constitution unless they change it (the current government is on the road to it, though).
If China really wants to fuck with Taiwan, they had better do it soon while the most incompetent US administration ever to hold office is still there. Because there is change coming to the US and the next president isn't likely to be such a complete moron.
> Establish a total naval and air blockade of Taiwan using their fleets stationed in Ningbo and Zhanjiang.
I would not want to be maintaining a long-term naval blockade while constantly looking over my shoulder worrying about surprise attacks. Everyone knows where you are, and it would give plenty of time for a counter offensive to be planned. The US has something like 50 attack submarines, and a surprise attack from all of them at once would be a nightmare scenario for any fleet.
In the era of drones and missiles having your navy tied down in a blockade like that makes it all the more vulnerable.
US had all of that and still was unable to do anything meaningful around Hormuz.
US had 10,000 miles of power projection, while China prepared for 100 miles of it for the past 20 years, come on. The answer is EUV alternative that would render TSMC useless.
I understand that majority of people here cannot see the bigger picture, by design.
As an American, what are ways in which you can not defer to an authority - in a way that actually matters? Sure, you can choose to go outside dressed in gaudy colors, or not dressed at all if you are in SF. Or choose to flip the Democrat/Republican bit once every 4 years (usually without any effect on domestic or foreign policy, though not lately). But in a regular person's life, what authority do you want to opt out of? The average person can't do a thing about their HOA board, let alone a government representative.
but there are a whole lot of ways you can mess up a dull language, like shitty variable names, bad code organization, etc.
if an llm has learned dumb things from dumb users it could disproportionately cause provlems versus other languages, just by being "in a sloppy mood" when writing go.
They were in clear violation of Apache 2.0
reply