X is likely losing money in the EU at this point, and complying with the rules would also cost them money and/or reputation. Maybe they should consider pulling out.
I leapfrogged this by pulling myself out from Twitter instead a while ago, and I can only recommend. And I don't mean this in the "in favor of Bluesky/Mastadon" sense.
There are some types of content that I did lose access to this way, but in retrospect it was worth it. I found that the cost-benefit for it is just not there, not for me at least.
I’m assuming they’re only losing money in the hard sense, not in the “soft” sense considering the unimaginable wealth that comes from manipulating millions of people.
> and complying with the rules would also cost them money and/or reputation. Maybe they should consider pulling out.
Like when they "pulled out" of Brazil? When Musk said they were fighting for freedom, democracy, and the rule of law? Then X got blocked in Brazil, and they silently complied with everything that was against "freedom, democracy, and the rule of law" and stopped talking about it.
A "cooperative" president who could be manipulated by everyone, which is why he promoted Tesla cars in front of the White House but also reduced support for them, reduced fleet MPG requirements, thinks clinate change and renewables are scams, etc.
And then Musk had his falling out when he left DOGE.
Why would they pull out? They can simply not pay the fine. Then the EU could either back down or be the bad guy that blocks X.com and seizes their operations in Europe.
Trillionaires don't buy mass media companies to make money, they do it to push propaganda.
This one in particular is both personally desperate for approval of his petty, narcissistic bullshit, and for trying to consolidate political power by amplifying and supporting authoritarian candidates (that are expected to quid-pro-quo him back).
It has always been hard to tell who will be good at a job based on a CV and a bunch of short tests. AI has just disrupted the stupid proxies created by HR, because now everyone can fake being a good CV writer or doing leetcode.
This is the type of hard constraint that could seriously hamper scaling. You simply can't scale hardware at this level without some supply chains issues that increase costs enormously. Similarly with electricity production.
> To give you an idea, Georgescu’s TikTok presence was carefully crafted: short clips showing him speaking "against the system," promising to "free Romania from foreign control," heavily relying on emotional imagery like children, old people, and national flags (Tukiainen, 2024). These were sprinkled with half-truths and out-of-context data, making them seem plausible to those scrolling fast without fact-checking. The persuasive power came not from logic, but from emotion, repetition, and algorithmic amplification.
The optimistic view is that Anthropic is one of about four labs in the world capable of generating truly state-of-the-art models. Also, Claude Code is arguably the best tool in its category at the moment. They have the developer market locked in.
The problem as I see it is that neither of those things are significant moats. Both OpenAI and Google have far better branding and a much larger user base, and Google also has far lower costs due to TPUs. Claude Code is neat but in the long run will definitely be replicated.
The missing piece here is Anthropic is not playing the same game. Consumer branding and larger user base are concerns for OpenAI vs Google. Personal chatbot/companion/ search isn’t their focus.
Anthropic is going for the enterprise and for developers. They have scooped up more of the enterprise API market than either Google or OpenAI, and almost half the developer market. Those big, long contracts and integration into developer workflows can end up as pretty strong moats.
> Cursor had won the developer market from the previous winner copilot
It’s a fair point, but the counter-point is that back then these tools were ide plugins you could code up in a weekend. Ie closer to a consumer app.
Now Claude Code is a somewhat mature enterprise platform with plenty of integrations that you’d need to chase too. And long-term enterprise sales contracts you’d need to sell into. Ie much more like an enterprise SAAS play.
I don’t want to push this argument too far as I think their actual competitors (eg Google) could crank out the work required in 6-12 months if they decided to move in that direction, but it does protect them from some of the frothy VC-funded upstarts that simply can’t structurally compete in multi-year enterprise SAAS.
Fun fact! You can use the word "just" in front of anything to make is sound trivial. Isn't planet Earth just one of eight planets in the Solar System? What's the big deal? Isn't Google just a website? Take out the word "just" and think on it a little. In this case, maybe there's something to that?
Most of the secret sauce of Claude Code is visible to the world anyway, in the form of the minified JavaScript bundle they send. If you’re ever wondering about its inner workings you can simply ask it to deminify itself
Developers will jump ship to a better tool at a blink of an eye. I wouldn't call it locked in at all. In fact, people do use Claude Code and Codex simultaneously in some cases.
Individual and startup devs yes. Enterprise devs, less so.
The latter are locked in to whatever vendor(s) their corporate entity has subscribed to. In a perverse twist, this gives the approved[tm] vendors an incentive to add backend integrations to multiple different providers so that their actual end-users can - at least in theory - choose which models to use for their work.
most of the secret sauce of Claude Code is visible to the world anyway, in the form of the minified JavaScript bundle they send. If you’re ever wondering about its inner workings you can simply ask it to deminify itself
> “I own the compute infrastructure” or “I owned the company that bought the AI systems early”, on the other hand, is just being in the right place at the right time. It’s not about your talents, effort, or contribution. It’s pure capital ownership divorced from any human quality.
The article is not about AI, it's about this stage of capitalism. Unlike the author, I would argue that it is very much in line with the consequences of Robert Nozick's thinking. On the other hand, the way China is doing its AI development and rollout seems more aligned with a Rawlsian notion of distributing the benefits.
I can perfectly understand why everyone in the West is so jaded and worried about how AI benefits will be distributed. That doesn't change the fact that, like any technology, it can also be used to make everyone wealthier, like industrialization also did.
My take on this: Ultimately the industrial revolution has been beneficial to the majority, but when it occurred it triggered violent changes that pushed many communities toward hardship. From artisans owning or co-owning their means of production, with a significant creative license over their work and the possibility to take their own initiatives… to the efficiency of atomized labor, which strips away the creativity of most, and devalues the work. The created hardship providing more people ready to join the ranks of devalued workers… The industrial revolution was, first, a great mean of consolidating wealth, it immensely benefited a few, and only later “trickled down”. I see the AI revolution the same, it will violently break a significant portion of knowledge works, remove creative licenses, and devalue those works. Similarly to the industrial revolution, we can now atomize knowledge works as much as we want. Through this we can make the workers as easy to replace as we want them to be. This shifts power dynamics and allows the consolidation of wealth. There are strong market incentives for this and no regulations at the moment. I don’t see how we can avoid this. I don’t see those pushing this revolution as great humanists sadly.
Yet ultimately, I think this will damage the middle class too much, creating instabilities. From there, there is a a chance we can reclaim the benefits of this revolution for all. But it will be a fight, and I believe the transition will be ugly.
Yeah, I mostly agree with your opinion. It's true that in the short-term there will be tons of disruption and likely more bad than good will come out of it. I guess my point was that ultimately how the upsides and downsides of a new technology will be distributed depend a lot on how societies organize themselves. The industrial revolution allowed the concentration of wealth and robber barons, but also unions and social democracy as a reaction to that.
I agree, it really depends on how society organizes. For this reason I tend to be a bit frustrated by how AI narratives are often framed. They are often described as only technological. As if a tool could wield itself without anyone behind to operate it. We talk about the impact of AI as if it was inevitable progress. Those issues are in fact deeply political, but the valley mentality doesn’t like to see it that way. Probably because it is uncomfortable to see our work as contributing to a complex system with winners and losers, it’s much nicer to see ourselves as stewards of progress. The question “the progress of what?” is often pushed in the background.
Well huawei ban bought samsung sometime, scared PRC brands from expanding into north american market. TBH Samsung was still pretty dominant until PRC brands really turned dial on hardware while Samsung stagnated until they couldn't. The latest round of hardware is pretty good, as in PRC flagship parity worthy. TBH the Koreans are very talented, they don't have the numbers to keep up with PRC speed / product cycles, but if they can iterate proper flagship every other year, they'd be in a good place. Also not putting ads on fridges.
Sure, I'm actually old enough to remember Huawei's Ascend sold under MetroPCS in the US back in the early 2010's. Leica collaboration with Huawei in 2016 worked wonders and other Chinese smartphone makers definitely stepped up, but, by this time, Samsung's China sales fell off the cliff by ~70+% to a low single-digit market share from its 20% peak in 2013 under Xi's "In China, For China" campaign.
Not sure if Huawei was ever a threat to Samsung or Apple outside China as most of Huawei's growth was in China only and there was no other single major market in which Huawei came close to Apple's or Samsung's. China is also the only major market where Samsung's market share is less than 1% and I'm very disinclined to believe this is coincidence. I think the common misconception is that Samsung was "outcompeted" by Huawei when it was in fact forced out of China. This practice became quite common in other industries too after Xi -- eg, all foreign competitors in EV batteries business such as LG, Panasonic, Samsung, etc were also effectively banned in China under Xi's Made-In-China 2025, launched in 2015 to protect local "champions," such as CATL/BYD.
Samsung share was dropping in PRC before Xi, i.e. when cheap domestic brands started eating the bottom. Samsung flagship was still popular, i.e. the low single digit highend, then the Note battery recall drama happend and basically THAAD right after and the double whammy basically killed Samsung in PRC. Now do I think Samsung could have recovered and held on like Apple with domestic competition, probably not, samsung not as sticky as life style choice.
Before Huawei sanction global shipments went from 100m to 200m in like 4 years (double digit YoY growth) while samsung was declining from 300m in same time period. Everyone saw which way the trend lines was going, especially in HW flagships.
MIC2025 is like for establishing nascent industries, i.e. your batteries example. PRC companies get whitelist/subsidies for a few years then opens to foreign players after CATL becomes incumbant. Samsung mobile doesn't fit MIC2025 pattern since PRC already established phone manufacturing before MIC2025 started, entire low&high spectrum by 2015. It's not some strategic industry being spun up from 0, they already knew everything about phone production from Foxconn. There's no reason to force Samsung out of PRC, domestic phones already got CATLized and was outcompeting Samsung by then. Also it's not like Samsung was formally "kicked out", they left after seeing same writing on the wall. If Samsung got kicked out, like even informally, there'd be transition plan, i.e. get a local player to take over Huizhou company seen in other MIC2025 plays. Factories don't sit idle. Instead Samsung picked up and left and basically Huizhou become ghost towned.
>> Samsung share was dropping in PRC before Xi, ... the Note battery recall drama happend and basically THAAD right after ... <<
Sure, except that narrative has too many flaws. Samsung China's smartphone sales increased from 11M in 2011, to 30M in 2012, to about 60+M in 2013 -- no such "drop" before Xi's rise. Then it cratered as Xi's anti-foreign "In China, For China" policy went in full swing. By 2016, Samsung's China sales figure was already down to just over 10M -- the Note 7 and THAAD were likewise inconsequential and Samsung smartphone sales continued to slide in China and China only. During this time between 2016-2017, Samsung's global sales continued to maintain its lead at around 310M even as its Chinese competitors, such as Huawei struggle against Samsung outside China without protection from Xi. Again, this conspicuous drop was in China only and there was no sudden change in consumer preference or product offering, as evident from Samsung's global sales -- ie, while Samsung China sales was down, it was up elsewhere. Xi's mercantile policies and the rising nationalism in China were the only changing variables.
Also it wasn't just Samsung alone either -- Apple was the next major foreign smartphone seller in China.
Apple avoided Samsung's fate after Xi's rise in 2013 and Patrick McGee, a FT reporter, has recently released
Apple in China: the Capture of the World's Greatest Company, meticulously detailing how Apple countered Xi's "For China, In China," but, by doubling down on China, Apple is in effect "captured" by Xi. see Chapter 26 "Despot" and on.
Looking at sales in 2010s mobile growth market is flawed. look at market share. 2010-2015 was mass smart phone proliferation, the demand denominator was expanding so companies can sell more phones while lose relative marketshare. Samsung PRC share peaked and started dropping right before/around when Xi entered office, i.e. domestic (budget) champions was grown before Xi, they were outcompeted before MIC2025 or battery/thaad drama. Samsung global sales stalled at ~300m and started dipping below (i.e. losing total sales and market share in growth market), again need to look at market share because denominator of global phone sales was exploding. Samsung Could had 40/50% of global sales 500m+ branded sales, instead Huawei+others jumped in and ate big chunk of Android pie, i.e. actual global Android shipments was like 1B, PRC brands including white labeled OEMs captured probably 50%+ of that. Huawei going from 0-200m phone sales abroad while Samsung stalling and even declining at ~300m when they had 500m+ to capture isn't Huawei struggling (aging double digit global YoY growth at the time), it's PRC brands gobbling up new demand at expense of Samsung. Samsung got short term boost after HW ban, then started stagnating again when budget PRC brands went flagship. Also there was huge change in product offering 2010-2015s in PRC, that's when domestic budget brands and then flagships proliferated. There's nothing sudden about any of this, Xi barely had any time in office to engineer some mass buy domestic movement, that only happened late 2010s across variety of domestic sectors.
Patrick McGee / Apple in China is motivated writing. Trying to sell narrative Apple taught PRC everything they instead of you know 100,000s PRC tooling engineers and manufacturing specialistists trying to troubleshoot and scale Apple design. I don't know what Chapter 26 has to do with anything, IIRC it's mostly PRC making Apple kneel to PRC soveign internet, aka follow doemstic laws and Apple complied because they need PRC manufacturing. And doing so let Apple kept their share in PRC for another 20 years which wasn't bad return. And it's not even like Samsung left PRC, they had to pivot back to PRC ODMs (wingtech, huaqin) for ~100 of millions of low end devices (I htink the As and Ms) shortly after.
>> look at market share. Samsung PRC share peaked and started dropping right before/around when Xi entered office ... <<
Sure, you could use either metrics, but it shows the same thing: Samsung China's market share in 2012 was 17.7% (30M). Xi came to power in mid 2013 and Samsung's sales in each quarter 2013 were 18% (12.5M) in Q1, 19% (15M) Q2, 18% (17M) Q3, and 19% Q4 -- ie, it's a steady YoY growth, in relative % or absolute terms.
In other words, Samsung's sales were not affected by local competition "right before" or "around" Xi. Only after Xi's "In China, For China," both market share and absolute sales plunged from 17.7% (2012), to 19% (2013) to 7.9% (2014) or 31M (2012), to 70M (2013) to 35M (2014).
Again, this was isolated to China and as a result of Xi's anti-foreign policy in China only -- Samsung's sales continue to grow outside China to maintain their global lead.
Further, I cited McGee's recent excellent work which clearly demonstrates how Xi's anti-foreign policies suddenly changed business climate in China against foreign competitors -- which starts at Part 5: Political Awakening, Chapter 26 -- why it has everything to do with "The DESPOT." Not market competition, or THAAD/Note 7 fire.
Your theory doesn't work in real world timeline where we have to factor in actual execution time. Also it simply doesn't match actual policy history. Xi doesn't get to flip switch and conjure a bunch of domestic phone brands and shift market share within a few quarters. That stuff takes years, not suddenly. Hence on/around. The fact that Samsung peaked right when Xi entered office means Xi was not causative since Xi wouldn't have any time to implement buy domestic policy, which also didn't exist at the time. Market share changes work on multi year lag, for Samsung shares to decline, it requires Xiaomi, Vivo, Oppo etc to have build out manufacturing years prior, i.e. before Xi.
Samsung global sales stagnated few years later and saw some decline in the same period as Huawei was doing double digit global YoY growth - again factor in actual market lag of PRC companies cornering domestic market before going abroad. Then HW got knee capped by export controls. Now maybe HW might not have surpassed Samsung but it's pretty clear Samsung wasn't growing while PRC brands were, while none sanctioned brands continue to chip away at Samsung (i.e. this article).
Either way, claiming there was some anti foreign policy in 2013 simply doesn't comport with reality because we have paperwork of Xi's actual year 1 policies, there wasn't any anti-foreign or buy domestic initiatives. Hence I don't see the relevance of McGee regardless. I think chapter talks about political awakening to PRC business climate not any mobile sector industrial substitution policies (which again, did not exist).
PRC telling Apple to follow PRC laws, i.e. cloud sovereignty isn't anti foreign it's just forcing foreign companies to compete on same requirements as domestic companies like data localization. That happeneing in 2013 was part of broader cybersecurity/ideological work, i.e. the institution work stuff that applied to domestic and foreign companies. Actual MIC2025 related industrial policy didn't happen until years later. Procedurally Xi's first year focused on corruption/CCDI, institutional stuff, there wasn't any industrial policy like MIC until a year later, like 3rd plenum was in December 2013. Xi wasn't going around telling domestic consumers to stop buying Samsung phones or kicking Samsung out of the country. Otherwise Samsung again, wouldn't have freely returned to PRC to partner with PRC ODMs for low end lines that they realize even PRC has cost advantage on vs Vietnam. Incidientally the same cost advantage that made PRC / global consumers buy from affordable PRC brands in the first place.
>> Your theory doesn't work in real world timeline where we have to factor in actual execution time. ... That stuff takes years,
(no "reply" button in your last comment)
Sure, doesn't really matter. I just have to demonstrate that your "local competition" theory in pre-Xi era doesn't hold water and the real world data don't bear that out, which I already have.
Whether Samsung China's sudden sales collapse came 1 day, or 1 year after Xi? not very relevant. That being said, in Apple's case, the Chinese media's coordinated attack started on Consumer Day in March 2013, just a day after Xi became president, but because of Apple's quick response and bold risky bet, suffered no damage. Samsung China's first sales decline since 2009 on the other hand became evident in Q1 2014 and after, or some 3 quarters after Xi. Or roughly equal to a 50% drop in market share and unit sales for the entire year 2014 and more significant drops in subsequent years. So we are talking 4+ years to go from 20% to less than 1%, so there goes your timeline arg out the window, too -- it's plenty time to implement informal anti-market policies and destroy a foreign competitor completely.
Again, this dramatic drop was not seen elsewhere, but in only China, indicating that the problem wasn't inherent to Samsung's product offering, competitiveness, or market. In open markets, such as EU, Samsung maintained a steady ~32+% market share post-Xi and wasn't affected at all by the rise or fall of Huawei, before or after the ban. Samsung EU's market share dipped below 30% in 2021 for the first time in 8+ years, after Huawei's ban which further undermines the narrative that Samsung somehow benefited from it.
Real world data bares out that local competition theory holds. You've demonstrated questionable ability to intrepret data in real world context. Of course it's relevant when Samsung sales collapse happens if it happens in time period where policy changes (which again does not exist) simply could not have effect. The fact that collapse happened when Xi cannot have causative effect means Xi's policies (or lack of) was not responsible. If Xi entered office, made policy to kill Samsung, and 4-5 years later PRC built up domestic mobile players to do so, then sure. But those players already existed and had scaling plans to supplant Samsung domestically pre Xi. Samsung markshare dropping in subsequent years was because domestic PRC brands who built factories pre Xi started doing their natural scaling hence execution years prior to Xi resulted in capturing more marketshare in subsequent years, which entirely comports with timeline. Your theory also simply doesn't comport on the fundmental level that there weren't any anti-market policies in smart phone domain, because it was mature domain where pre Xi PRC competitors were already in place. Anti-market policies was for nacent industries, i.e. MIC2025.
Dramatic drop was seen first in PRC then globally which fits customary trend of PRC players establishing in PRC first then expand globally. We have global data that Samsung sales went from stalling to negative while Huawei scaling as global player to rising after Huawei ban, i.e. Samsung share went from stalling to jumping to 40% in EU after Huawei ban from 30%. Like this is all very well established history, there's no need to entertain alternative / revisionist theories when plenty of market analysis at the time already discussed Huawei eating lunch of Samsung and Apple pre ban while Samsung being primarily benefitiary of HW ban. And again it entire avoids the fact that Samsung went backc to PRC for ODMs for the same reason PRC manufactures displaced Samsung, because PRC had the most cost competitor manufacturers for low-medium end devices that accounted for most sales.
Sure, thx for playing, but really no point in talking in circles.
I find your last cherrypicked "Central and Eastern" EU smartphone data another stroke of desperation/genius, though it is not totally unexpected. Quite revealing!
All smartphone manufacturers were in China when Xi started shaking down the industry back in 2013 under the banner of "In China, For China." Samsung has diversified away and to Vietnam and India since, but I don't think we want to have a supply-chain all consolidated in one location/country.
I'm otherwise of opinion that the West's decisive counter measures are necessary against China's mercantile practices.
That happened due to the China-South Korea trade war following the installation of THAAD in SK in 2016 [0]. Notice how the Chinese OEM spike and Samsung's decline happen following the 2016-17 diplomatic crisis. It was also during this period that Korea Inc began shifting to Vietnam [1][2] and India as a result.
Additonally, that spike for Xiaomi and other Chinese OEMs also happened right when Chinese OEMs expanded their India business in 2015-17 [3][4][5]. On that note, notice how all those Chinese OEM saw sales dropped and then flatlined from 2021 onwards. While the pandemic did play a role, India began lawfare against Chinese companies following the Galwan Crisis in 2021 [6][7][8] with the Indian government de facto forcing Chinese firms to "indianize" [9] - which ironically is similar to how the Chinese government operated in the 2000s and 2010s with Western firms and what the Chinese government leveraged against Korea a decade previously.
>> That happened due to the China-South Korea trade war following the installation of THAAD in SK in 2016 [0]. <<
Not really. THAAD really plays no part in Samsung's fall in China. Samsung's smartphone sales in China was already down by -70% by the time THAAD broke out in 2016 from its peak in 2013 and still went down further to less than 1%. Samsung packed up and closed the last Chinese factory in 2019 -- went to Vietnam instead.
Patrick McGee recently released Apple in China: The Capture of the World's Greatest Company: it better describes the anti-foreign political situation in China at the time and what it meant to the smartphone industry. And how Apple avoided Samsung's fate, but is now captured by it. See Chapter 26 "Despot" and on.
It handles DevOps tasks way faster than I would - setting up infra, writing migrations, config changes, etc. Project is still early stage so speed and quick iterations matter more than perfect process right now. Once there's real traffic and a team I'll tighten things up.
If I felt the need to optimise things like infra setup and config at an early stage of a project, I'd be worried that I'm investing effort into the wrong thing.
Having an LLM churn out infra setup for you seems decidedly worse than the `git push heroku:master` of old, where it was all handled for you. And, frankly, cheaper than however much money the LLM subscription costs in addition to the cloud.
But why have it execute the tasks directly? I use it to setup tasks in a just file, which I review and then execute myself.
Also, consider a prod vs dev shell function that loads your prod vs dev ENV variables and in prod sets your terminal colors to something like white on red.
We don't understand how humans think, and we don't yet understand completely how LLMs work. It may be that similar methods are being used, but they might also be different.
What is certain is that LLMs can perform as if they are doing what we call thinking, and for most intents and purposes this is more than enough.
I think the evidence is actually pretty strongly against them doing anything similar to "thinking". Certainly they are exhibiting some behaviour that we have traditionally only associated with thinking. But this comes along with lots of behaviour that is fundamentally opposite to thinking ("hallucination" being the major example).
It seems much more likely that they are doing some other behaviour that only sometimes resembles thinking, in the same way that when you press the middle autocomplete button on your phone keyboard it only sometimes resembles conversation.
You see, we are here observing a clash in the terminology. Hallucinations in humans is thinking, just not typical. So called "hallucinations" in LLM programs are just noise output, a garbage. This is why using anthropomorphic terms for programs is bad. Just like "thinking" or "reasoning".
I think the answer is somewhere in the middle, not as restrictive as parent, but also not as wide as AI companies want us to believe. My personal opinion is that hallucinations (random noise) are a fundamental building block of what makes human thinking and creativity possible, but we have additional modes of neuroprocessing layered on top of it, which filter and modify the underlying hallucinations in a way so they become directed at a purpose. We see the opposite if the filters fail, in some non-neurotypical individuals, due to a variety of causes. We also make use of tools to optimize that filter function further by externalizing it.
The flip side of this is that fundamentally, I don't see a reason why machines could not get the same filtering capabilities over time by adjusting their architecture.
I have never in my life met a person who hallucinates in the way ChatGPT etc do. If I did, I would probably assume they were deliberately lying, or very unwell.
Is this really common behaviour? I do not recognise it. Do people lie? Certainly yes. Do people misremember, or get details incorrect? Yes. But when was the last time you saw someone, say, fabricate an entire citation in a paper? People make transcription errors, they misremember dates, and they deliberately lie. But I don't think people accidentally invent entire facts.
To me, your entire claim here comes across as "hallucination". That is, I simply do not believe that you have not experienced people accidentally inventing entire facts, and so I don't believe you are genuinely unaware of people doing it.
To be clear, I'm not arguing you've made this claim in bad faith at all.
However, going back and examining my own writing, I have more than once found claims that I'm sure I believed at the time of making them, but that I in retrospect realise I had no actual backing for, and which were for that reason effectively pure fabrication.
An enduring memory of my school days was convincing the teacher that she was wrong about a basic fact of geography. I was convinced. I had also totally made up what I told her, and provided elaborate arguments in favour of my position.
To me this is innate human behaviour that I see on a regular basis. People accidentally invent entire "facts" all the time.
Indeed. The mere fact that we ended up with the anthropomorphic term "hallucination", rather than something purely mechanistic like "glitch", indicates that there's something about this AI pattern that feels familiar.
I'm obviously not claiming that "hallucination" is an appropriate term ("delusion" or "confabulation" are probably more apt), but there is something here that is clearly not just a bug, but rather a result of thinking being applied properly but to ungrounded premises. To my eyes, reading an AIs "hallucination" is not unlike reading the writings of a human on drugs, or with a mental condition like schizophrenia, or just of an analytic philosopher taking their made up axioms all the way to an alternate universe.
Can you please also hallucinate a plausible-sounding justification for this otherwise unsubstantiated statement?
Jokes aside, we do produce plausibile sounding stuff all the time well beyond the limit of what we actually know or can prove. I think there is a continuum between formulating statements about things we don't know for sure and we can't prove, guessing details here and there to fill gaps in our memory, misremembering things that we thought we knew, and making up entire facts that sound plausible but are completely invented. Yes, llms seem to have trouble introspecting what they actually know; but it sounds more like a missing skill rather than a fundamental difference in the way they reason.
I believe this is not a binary question, there is a spectrum. I think of LLMs as a sophisticated variation of a Chinese room. The LLMs are given statistical rules to apply to the given input and generate an output. The rules encode some of the patterns that we call thinking uses and so, some of their responses can be interpreted as thinking. But then, again, in certain conditions, the responses of mammals, unicellular organisms and even systems unrelated to carbon based life forms can be thought to be performing what we vaguely call thinking.
One problem is that we don't have a clear definition of thinking and my hunch is that we will never have a clear cut one as it falls in the same category of phenomena like alive/death states, altered states and weather systems. One hidden assumption that I often see implied in the usages of this word is that the word "thinking" implies some sort of "agency" which is another vague term normally ascribed to motile life forms.
All in all I think this debate ensues from trying to emulate something that we don't fundamentally understand.
Imagine in a world where medicine has not advanced and we lack any knowledge of human biology, we are trying to create artificial life forms by creating some heat resistant balloon and having it take in and push air. Someone would argue that the globe is alive because there is something in that taking in air and pushing it out that is like what humans do.
The Chinese Room is just a roundabout way of pleading human exceptionalism. To any particular human, all other humans are a Chinese Room, but that doesn't get addressed. Nor does it address what difference it makes if something is using rules as opposed to, what, exactly? It neither posits a reason why rules preclude understanding nor why understanding is not made of rules. All it does is say 'I am not experiencing it, and it is not human, therefore I dismiss it'. It is lazy and answers nothing.
> The Chinese Room is just a roundabout way of pleading human exceptionalism
Au contraire, LLMs have proven that Chinese Rooms that can casually fool humans do exist.
ELIZA could be considered a rudimentary Chinese Room, Markov chains a bit more advanced, but LLMs have proven that given enough resources, LLMs can be surprisingly convincing Chinese rooms.
I agree that our consciousness might be fully explained by a long string of deterministic electrochemical reactions, so we could be not that different; and until we can fully explain consciousness we can't close the possibility that a statistical calculation is conscious to some degree. It just doesn't seem likely IMO right now.
Food for thought: If I use the weights to blindly calculate the output tokens with pencil and paper, are they thinking, or is it a Chinese Room with a HUGE dictionary?
> ELIZA could be considered a rudimentary Chinese Room, Markov chains a bit more advanced, but LLMs have proven that given enough resources, LLMs can be surprisingly convincing Chinese rooms.
Eliza is not a Chinese room because we know how it works. The whole point of the Chinese Room is that you don't. It is a thought experiment to say 'since we don't know how this is producing output, we should consider that it is just following rules (unless it is human).
> Food for thought: If I use the weights to blindly calculate the output tokens with pencil and paper, are they thinking, or is it a Chinese Room with a HUGE dictionary?
Well, I never conceded that language models are thinking, all I did was say that the Chinese Room is a lazy way of concluding human exceptionalism.
But, I would have to conclude that if you were able to produce output which was coherent and appropriate, and exhibited all signs of what I understand a thinking system to do, then it is a possibility.
I don't know about you but we used to joke back in the day that the computer "is thinking" when taking long to process something.
Dictionaries usually provide some kind of useless circular definition. Thinking? The act of making thoughts. Thoughts? The result of thinking. I can't believe people used to pay for these things.
In any case it's something to do with taking input data, doing something with it, and generating new data related to it. That's more or less just recursive inference.
I think you are addressing the issue from a developer's perspective. I don't think TPUs are going to be sold to individual users anytime soon. What the article is pointing out is that Google is now able to squeeze significantly more performance per dollar than their peer competitors in the LLM space.
For example, OpenAI has announced trillion-dollar investments in data centers to continue scaling. They need to go through a middle-man (Nvidia), while Google does not, and will be able to use their investment much more efficiently to train and serve their own future models.
> Google is now able to squeeze significantly more performance per dollar than their peer competitors in the LLM space
Performance per dollar doesn't "win" anything though. Performance (as in speed) hardly cracks the top five concerns that most folks have when choosing a model provider, because fast, good models already exist at price points that are acceptable. That might mean slightly better margins for Google, but ultimately isn't going to make them "win"
It's not slightly better margins, we are talking about huge cost reductions on the main expense which is compute. In a context where companies are making trillion dollar investments, it matters a lot.
Also, performance and user choice are definitely impacted by compute. If they ever find a way to replace a job with LLMs, those who can throw more compute at it for a lower price point will win.
reply