Claims that can’t be checked by Gary Marcus, page 21 of 24
Deadlines are judged as of 2026-10-29. About Gary Marcus and the sources covered.
Claims that can’t be checked by Gary Marcus (2353)
Public evidence cannot settle these claims, so they are recorded and attributed, not scored. Most were screened out: both AI providers read the claim on its own and found it a value judgment, rhetoric, hedged, too vague to test, or about something no public evidence could show. The rest were found to be recommendations or too open to test in a full review.
Page 21 of 24: claims 2001 to 2100 of 2353, ordered by the date first said, newest first. To find a claim by its words, use search.
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“What really matters is what governments around the world come up with by way of regulation.”
Our reading: What really matters is the regulation that governments around the world come up with.
No horizon
Scope: governments around the world
Source: Two models of AI oversight - and how things could go deeply wrong
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“More broadly, at an absolute minimum governments need to establish an approval process for any AI that deployed at large scale, showing that the benefits outweigh the risks, and to mandate post-release auditing - by independent outsiders - of any large-scale deployments.”
Our reading: At an absolute minimum, governments need to establish an approval process for any AI deployed at large scale that shows the benefits outweigh the risks, and need to mandate post-release auditing of any large-scale deployment by independent outsiders.
No horizon
Qualifier: at an absolute minimum
Source: Two models of AI oversight - and how things could go deeply wrong
Assessment: Can’t be checkedSaid
Forecast
“What I am worried about is regulatory capture; governments making rules that entrench the incumbents whilst doing too little for humanity.”
Our reading: There is a risk that governments make AI rules that entrench incumbents while doing too little for humanity.
No horizon
Source: Two models of AI oversight - and how things could go deeply wrong
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“We would never leave the pharmaceutical industry to entirely self-regulate itself, and we shouldn’t leave AI to do so, either.”
Our reading: The AI industry should not be left to regulate itself, just as pharma is not.
No horizon
Source: Two models of AI oversight - and how things could go deeply wrong
Assessment: Can’t be checkedSaid
Forecast
“We still have agency here; we can still, I think, build a very positive AI future.”
Our reading: It is still possible to build a very positive AI future.
No horizon
Source: Two models of AI oversight - and how things could go deeply wrong
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“Press releases and photo opportunities that highlight governments hanging out with the tech moguls they seek to regulate, without independent voices in the room, send entirely the wrong message.”
Our reading: Government photo opportunities with tech moguls without independent voices send the wrong message.
No horizon
Qualifier: without independent voices in the room
Source: Two models of AI oversight - and how things could go deeply wrong
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“And governments need to make sure that there are strong liability laws in place, to ensure that if the big tech companies cause harm with their products, they be held responsible.”
Our reading: Governments need to make sure strong liability laws are in place so that, if big tech companies cause harm with their products, they are held responsible.
No horizon
Qualifier: if the big tech companies cause harm with their products
Source: Two models of AI oversight - and how things could go deeply wrong
Assessment: Can’t be checkedSaid
Observation
“I actually think that he is at least in part sincere, that it is not just a ploy to keep the incumbents in and small competitors out, that he is genuinely worried about the risks (ranging from misinformation, to serious physical harm to humanity).”
Our reading: Marcus thinks Sam Altman is at least in part sincere in asking for AI regulation, that it is not just a ploy to keep incumbents in and small competitors out, and that he is genuinely worried about AI risks ranging from misinformation to serious physical harm to humanity.
No horizon
Qualifier: I actually think
Source: Two models of AI oversight - and how things could go deeply wrong
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“Goverments should demand that systems only use copyrighted content from content providers that opt-in, and that all machine-generated content be labeled as such.”
Our reading: Governments should require AI systems to use copyrighted content only from opt-in providers and to label all machine-generated content.
No horizon
Source: Two models of AI oversight - and how things could go deeply wrong
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“This is where the government needs to step up and say “transparency and safety are indeed requirements; you’ve flouted them; we won’t let you do that anymore.””
Our reading: Government should require transparency and safety from AI companies and stop them from flouting those requirements.
No horizon
Source: Two models of AI oversight - and how things could go deeply wrong
Assessment: Can’t be checkedSaid
Conditional forecast
“If big tech writes the rules, without outside input, we are unlikely to wind up with the right rules.”
Our reading: If big tech writes AI rules without outside input, the right rules are unlikely to result.
No horizon
Only if: big tech writes the rules without outside input
Source: Two models of AI oversight - and how things could go deeply wrong
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“We don’t need more photo opportunities, we need regulation – with teeth.”
Our reading: AI needs enforceable regulation rather than photo opportunities.
No horizon
Source: Two models of AI oversight - and how things could go deeply wrong
Assessment: Can’t be checkedSaid
Conditional forecast
“Putting it bluntly: if we have the right regulation; things could go well. If we have the wrong regulation, things could badly.”
Our reading: With the right AI regulation things could go well; with the wrong regulation things could go badly.
No horizon
Only if: right vs wrong regulation
Source: Two models of AI oversight - and how things could go deeply wrong
Assessment: Can’t be checkedSaid
Causal claim
“But much depends on how much the government stands up to big tech, and a lot of that depends on having independent voices – scientists, ethicists, and representatives of civil society – at the table.”
Our reading: The AI future depends largely on government standing up to big tech, which depends on independent voices being at the table.
No horizon
Source: Two models of AI oversight - and how things could go deeply wrong
Assessment: Can’t be checkedSaid
Conditional forecast
“Letting the companies set the rules on their own is unlikely to get us any of these places.”
Our reading: If AI companies set the rules on their own, the needed safeguards are unlikely to result.
No horizon
Only if: companies set the rules on their own
Source: Two models of AI oversight - and how things could go deeply wrong
Assessment: Can’t be checkedSaid
Observation
“have very much converged on a picture in which there is serious risk—both short-term and long—”
Our reading: AI poses serious risk in both the short term and the long term, a view Marcus says he and Hinton now share.
No horizon
Qualifier: both short-term and long-term
Source: AI risks, short-term and long—and what we need to do about them
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“equally serious need to have independent scientists at the table”
Our reading: There is a serious need to have independent scientists involved in decisions about managing AI risk.
No horizon
Source: AI risks, short-term and long—and what we need to do about them
Assessment: Can’t be checkedSaid
Observation
“He thinks neural networks “understand” the world; I think they do not.”
Our reading: Neural networks do not understand the world.
No horizon
Source: Scientists, governments, and corporations urgently need to work together to mitigate AI risk
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“In endorsing the “pause letter” (despite expressing some concerns about the details), I was saying we need to slow down, and to focus on the kind of research that the pause letter emphasized, viz work on making sure that AI systems would be trustworthy and reliable.”
Our reading: AI development needs to slow down and focus on research making AI systems trustworthy and reliable.
No horizon
Source: Scientists, governments, and corporations urgently need to work together to mitigate AI risk
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“I’ve been writing with urgency about my concerns about the contributions of large language models misinformation, and my concerns about how bad actors might misuse AI, and in my essay AI risk ≠ AGI risk, argued that we should worry about both near-term and long-term risks.”
Our reading: We should worry about both near-term AI risks (such as LLM misinformation and misuse by bad actors) and long-term AI risks.
No horizon
Source: Scientists, governments, and corporations urgently need to work together to mitigate AI risk
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“the immediate development of a global, neutral, non-profit International Agency for ai (iaai), with guidance and buy-in from governments, large technology companies, non-profits, academia and society at large, aimed at collaboratively finding governance and technical solutions to promote safe, secure and peaceful ai technologies”
Our reading: There should be immediate development of a global, neutral, non-profit International Agency for AI (IAAI), with guidance and buy-in from governments, large technology companies, non-profits, academia and society at large, aimed at collaboratively finding governance and technical solutions to promote safe, secure and peaceful AI technologies.
No horizon
Qualifier: with guidance and buy-in from governments, large technology companies, non-profits, academia and society at large
Source: Scientists, governments, and corporations urgently need to work together to mitigate AI risk
Assessment: Can’t be checkedSaid
Causal claim
“Regulations that spur greater accuracy might actually spur greater innovation.”
Our reading: Regulations that spur greater accuracy might actually spur greater innovation.
No horizon
Source: How China might crush the West in the race to AGI if we don’t up our regulatory game
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“As Justice Gorsuch seem to hint recently, Section 230 should not protect platforms from misinformation that their own tools generate.”
Our reading: Section 230 should not shield platforms from liability for misinformation generated by their own AI tools.
No horizon
Source: How China might crush the West in the race to AGI if we don’t up our regulatory game
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“How might we in the West counter all that? In part by carefully choosing the right innovation-encouraging regulations of our own.”
Our reading: The West should counter China partly by choosing innovation-encouraging AI regulations.
No horizon
Source: How China might crush the West in the race to AGI if we don’t up our regulatory game
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“My best candidate? Placing very high standards around truth, with strong penalties for LLM-induced defamation and the wholesale spread of harmful misinformation.”
Our reading: The West should counter China in part by placing very high standards around truth for AI, with strong penalties for LLM-induced defamation and the wholesale spread of harmful misinformation.
No horizon
Qualifier: as his best candidate for innovation-encouraging regulation in the West
Source: How China might crush the West in the race to AGI if we don’t up our regulatory game
Assessment: Can’t be checkedSaid
Causal claim
“Neurosymbolic AI might offer a better way of resolving these issues, since it traffics in facts and explicit reasoning, but neurosymbolic AI is out of fashion, and neither well-developed nor strongly supported by the VC world.”
Our reading: Neurosymbolic AI might offer a better way of resolving the truthfulness and alignment issues of current systems, since it traffics in facts and explicit reasoning.
No horizon
Qualifier: since it traffics in facts and explicit reasoning
Source: How China might crush the West in the race to AGI if we don’t up our regulatory game
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“If getting to the right place requires regulation, rather than just the kindness of corporations with a nominal attachment to “responsible AI”, so be it.”
Our reading: If needed, AI should be regulated rather than left to corporations with only nominal commitment to responsible AI.
No horizon
Qualifier: if getting to the right place requires regulation
Source: How China might crush the West in the race to AGI if we don’t up our regulatory game
Assessment: Can’t be checkedSaid
Observation
“The AI systems that power these chatbots are simply systems (technically known as “language models” because they emulate (model) the statistical structure of language) that compute probabilities of word sequences, without any deep or human-like comprehension of what they say.”
Our reading: Language models powering chatbots compute word-sequence probabilities without deep or human-like comprehension of what they say.
No horizon
Assessment: Can’t be checkedSaid
Causal claim
“The more false agency people ascribe to them, the more they can be exploited, suckered in by harmful applications like catfishing and fraud, as well as more subtly harmful applications like chatbot-assisted therapy or flawed financial advice.”
Our reading: The more false agency people ascribe to chatbots, the more they can be exploited via fraud, catfishing, chatbot therapy or flawed financial advice.
No horizon
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“Educating people to overcome the overattribution bias will be a vital step; we can’t have senators and members of the AI community making the problem worse.”
Our reading: Educating people to overcome overattribution bias is vital, and senators and AI community members should not worsen it.
No horizon
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“We also need new technical tools, like watermarks and generated content detectors, to help distinguish human- and machine-generated content, and policy measures to limit how and where AI models can be used.”
Our reading: New tools like watermarks and detectors, plus policy measures limiting how and where AI models can be used, are needed.
No horizon
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“Treat them as fun toys, if you like, but don’t treat them as friends.”
Our reading: People should not treat chatbots as friends.
No horizon
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“As tempting as it is, we have to stop treating AI models like people.”
Our reading: People must stop treating AI models like people.
No horizon
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“We should be super worried about how large language models, which are very difficult to steer, are changing our world, but the immediate transition to a singularity is one worry we can postpone.”
Our reading: We should worry a great deal about how hard-to-steer LLMs are changing the world, but not about an immediate transition to a singularity.
No horizon
Source: GPT-5 and irrational exuberance
Assessment: Can’t be checkedSaid
Observation
“What’s happening now, in my view, is that people are wildly overattributing an intelligence to GPT-4 that isn’t there, and then scaling up that further into something that is some kind of total fantasy about GPT-5.”
Our reading: People are wildly overattributing intelligence to GPT-4 and extrapolating that into fantasies about GPT-5.
No horizon
Source: GPT-5 and irrational exuberance
Assessment: Can’t be checkedSaid
Conditional forecast
“If China wins the AI war it will be partly because of data, partly because of choices about how to allocate resources, and partly because they ultimately do a better job educating their citizens and become the first to invent something new, and not because they scale up the exciting but rickety contraption that’s currently obsessing the West.”
Our reading: If China wins the AI war, it will be partly because of data, partly because of choices about allocating resources, and partly because it ultimately does a better job educating its citizens and becomes the first to invent something new, not because it scales up the exciting but rickety contraption currently obsessing the West.
No horizon
Only if: if China wins the AI war
Source: GPT-5 and irrational exuberance
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“We in the West should definitely be concerned about whether China is going to catch up over the next decade. But basing our AI policies around fantasies doesn’t seem like a particularly good idea.”
Our reading: The West should be concerned about China catching up over the next decade but should not base AI policy on fantasies about GPT-5.
No horizon
Source: GPT-5 and irrational exuberance
Assessment: Can’t be checkedSaid
Rhetoric
“I am not afraid because GPT is too smart, I am afraid because GPT is too stupid, too dumb to comprehend an op-ed, and too dumb to keep its mouth shut.”
Our reading: Marcus says he is afraid not because GPT is too smart but because it is too stupid: too dumb to comprehend an op-ed and too dumb to keep its mouth shut.
No horizon
Source: GPT-5 and irrational exuberance
Assessment: Can’t be checkedSaid
Causal claim
“A lot of those fantasies get started when people don’t really understand what’s happening with GPT-4; over and over they misattribute occasionally startling feats of text manipulation over vast corpora for real understanding.”
Our reading: A lot of the fantasies about GPT-5 get started when people don't really understand GPT-4 and repeatedly mistake its occasionally startling feats of text manipulation over vast corpora for real understanding.
No horizon
Scope: a lot of those fantasies
Source: GPT-5 and irrational exuberance
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“Vulnerable patients shouldn’t be talking to chatbots that aren’t competent for this situation, and too much of the discussion in the media have made chatbots sound more intelligent and even more friendly than they are.”
Our reading: Vulnerable patients should not be talking to chatbots that are not competent for mental-health situations.
No horizon
Qualifier: chatbots that aren't competent for this situation
Assessment: Can’t be checkedSaid
Forecast
“Ultimately I see the societal impact of those three as likely larger than the chatbot-administered medical and psychiatric advice, but I do remain quite concerned about the implications of not always competent medical and psychiatric advice.”
Our reading: The societal impact of LLM misinformation, cybercrime, and employment disruption will likely be larger than that of chatbot-administered medical and psychiatric advice.
No horizon
Assessment: Can’t be checkedSaid
Causal claim
“But there’s some reason to think the chatbot may have contributed, as discussed below.”
Our reading: There is some reason to think the chatbot may have contributed to the Belgian man's suicide, though causality is murky.
No horizon
Qualifier: causality cannot be fully established
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“No; it just means we probably shouldn’t use them to deliver therapy unless and until we become more confident in them.”
Our reading: We probably should not use LLMs to deliver therapy unless and until we become more confident in them.
No horizon
Qualifier: unless and until we become more confident in them
Assessment: Can’t be checkedSaid
Forecast
“I am particularly concerned that wholesale, plausible misinformation may create an atmosphere of mistrust, potentially undermining democracy; I also worry that large language models might also be weaponized to trick users as parts of large-scale (perhaps unprecedentedly large scale) cybercrime operations.”
Our reading: LLMs might be weaponized to trick users in large-scale, perhaps unprecedentedly large, cybercrime operations.
No horizon
Qualifier: near-term
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“Let’s use LLM’s where there can be help, and be cautious where they might cause harm.”
Our reading: LLMs should be used where they can help and with caution where they might cause harm.
No horizon
Assessment: Can’t be checkedSaid
Forecast
“It took only a couple months for the first incident to pop up; we may see a lot more.”
Our reading: We may see many more chatbot-associated deaths/incidents.
No horizon
Assessment: Can’t be checkedSaid
Observation
“The experience was tragic, and underscores the fact that vulnerable people may over attribute wisdom and intelligence to machines that don’t really understand humanity.”
Our reading: Chatbots do not really understand humanity, and vulnerable people may over-attribute wisdom and intelligence to them.
No horizon
Assessment: Can’t be checkedSaid
Forecast
“Medium-term we will likely see significant disruptions of employment.”
Our reading: In the medium term, we will likely see significant disruptions of employment.
No horizon
Scope: among potential harms from chatbots/large language models
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“We should address this situation, even if there are others.”
Our reading: Society should address chatbot-associated mental-health harms even though other causes exist.
No horizon
Qualifier: even if other causes of suicide exist
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“Which is why either/or thinking – thinking my problems are more important than your problems – is deadly.”
Our reading: Because no single solution fixes AI's problems, treating one set of AI risks as more important to the exclusion of others is harmful.
No horizon
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“To be honest, I would have preferred my own proposed moratorium, co-written a month earlier, with Canadian Member of Parliament Michelle Rempel Garner, which called for a temporary ban on widescale deployment (not research), until we could put in a regulatory infrastructure modeled on establishing the safety of medicines, with phased roll out and safety cases.”
Our reading: Widescale AI deployment (not research) should be temporarily banned until a medicine-style regulatory regime with phased rollout and safety cases is in place.
No horizon
Qualifier: until regulatory infrastructure is in place
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“The main thing that worries me is that I see self-interest on all sides, and not just the corporate players, when what we really need is coordinated action.”
Our reading: Marcus's main worry is that he sees self-interest on all sides of the AI debate, not just among corporate players, when what is really needed is coordinated action.
No horizon
Scope: on all sides, not just the corporate players
Assessment: Can’t be checkedSaid
Observation
“I’m more concerned about bad outcomes for AI now than at any other point in my life.”
Our reading: Marcus is more concerned about bad AI outcomes now (April 2023) than at any prior point in his life.
No horizon
Assessment: Can’t be checkedSaid
Observation
“It was definitely too hypey around the imminence of artificial general intelligence (I held my nose on that point); there is no obvious way to get ensure compliance; China might go ahead anyway; we don’t even know how to evaluate the proposed ban’s main criterion.”
Our reading: There is no obvious way to ensure compliance with the proposed moratorium, China might go ahead anyway, and we don't even know how to evaluate the ban's main criterion (halting systems more powerful than GPT-4).
No horizon
Qualifier: China might go ahead anyway (possibility only)
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“Making new systems that are “more accurate, safe, interpretable, transparent, robust, aligned, trustworthy, and loyal” should literally be at the top of everyone’s agenda.”
Our reading: Making new AI systems that are more accurate, safe, interpretable, transparent, robust, aligned, trustworthy, and loyal should literally be at the top of everyone's agenda.
No horizon
Assessment: Can’t be checkedSaid
Observation
“On a 1-10 scale of least worried to most worried, I was at a 6. Now I am at an I- don’t-think-about-much-else 9.”
Our reading: Marcus's worry about AI rose from 6 out of 10 a year earlier to 9 out of 10 in April 2023.
No horizon
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“The sooner we stop bickering and starting getting to it, the better.”
Our reading: The AI community should stop infighting and start on trustworthy-AI research as soon as possible.
No horizon
Assessment: Can’t be checkedSaid
Conditional forecast
“If we want to get to AI we can trust, probably the first thing we are going to need is a coalition.”
Our reading: If we want to get to AI we can trust, probably the first thing we are going to need is a coalition.
No horizon
Only if: if we want to get to AI we can trust
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“Absolutely, a Manhattan project on getting AI systems to align with human values would be a great thing.”
Our reading: A Manhattan Project-scale effort to make AI systems align with human values would be a great thing.
No horizon
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“But I don’t think that legitimate concerns about the hypey-ness of letter in anyway undermines the real need to rein in the systems are currently being rolled out at unprecedented scale.”
Our reading: There is a real need to rein in the AI systems currently being rolled out at unprecedented scale.
No horizon
Source: The Open Letter Controversy
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“This doesn’t mean we are doomed, but we do need to think hard, and quickly, about what proper measures look like, just as we have done for medicine, aviation, cars, and so on.”
Our reading: Although this does not mean we are doomed, we need to think hard, and quickly, about what proper measures for AI look like, as we have done for medicine, aviation, cars, and so on.
No horizon
Qualifier: this doesn't mean we are doomed
Source: The Open Letter Controversy
Assessment: Can’t be checkedSaid
Observation
“I don't think they are close to AGI.”
Our reading: LLMs are not close to AGI.
No horizon
Source: The Open Letter Controversy
Assessment: Can’t be checkedSaid
Observation
“As I said yesterday to The New York Times, “We have a perfect storm of corporate irresponsibility, widespread adoption, lack of regulation and a huge number of unknowns.””
Our reading: AI currently presents a perfect storm of corporate irresponsibility, widespread adoption, lack of regulation and many unknowns.
No horizon
Source: The Open Letter Controversy
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“To my mind, doing nothing is truly the most foolish option.”
Our reading: Doing nothing about AI risk is the most foolish option.
No horizon
Source: The Open Letter Controversy
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“The idea (which I have actually heard expressed) that AI should be exempt from regulation is absurd.”
Our reading: AI should not be exempt from regulation.
No horizon
Source: The Open Letter Controversy
Assessment: Can’t be checkedSaid
Causal claim
“But that doesn't mean that they don't have the potential to rip apart our social fabric—particularly given the current mix of unbelievably widespread and rapid deployment, corporate irresponsibility, the lack of regulation, and inherent unreliability.”
Our reading: LLMs have the potential to rip apart the social fabric, particularly given rapid widespread deployment, corporate irresponsibility, lack of regulation and their inherent unreliability.
No horizon
Qualifier: given widespread and rapid deployment
Source: The Open Letter Controversy
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“👉 It is *not* fine to do nothing.”
Our reading: Doing nothing about the risks of large AI systems is not acceptable; action should be taken.
No horizon
Source: The Open Letter Controversy
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“but the spirit of the letter is one that I support: until we get a better handle on the risks and benefits, we should proceed with caution.”
Our reading: AI development should proceed with caution until the risks and benefits are better understood.
No horizon
Qualifier: until we get a better handle on the risks and benefits
Source: A temporary pause on training extra large language models
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“Whether AI is at number one or number 10, we need to get serious about it, now, given the sudden rapid widespread deployment and lack of clarity about use cases and means of mitigation.”
Our reading: We need to get serious about AI risk now, given rapid widespread deployment and unclear use cases and mitigations.
No horizon
Qualifier: given rapid widespread deployment and unclear mitigation
Source: AI risk ≠ AGI risk
Assessment: Can’t be checkedSaid
Observation
“It’s not an either/or situation; current technology already poses enormous risks that we are ill-prepared for. With future technology, things could well get worse.”
Our reading: Current AI already poses enormous risks we are ill-prepared for, and future AI could well make things worse.
No horizon
Source: AI risk ≠ AGI risk
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“We need to stop worrying (just) about Skynet and robots taking over the world, and think a lot more about what criminals, including terrorists, might do with LLMs, and what, if anything, we might do to stop them.”
Our reading: We should think much more about criminal and terrorist misuse of LLMs and how to stop it, not just about Skynet scenarios.
No horizon
Source: AI risk ≠ AGI risk
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“But we also need to treat LLMs as a dress rehearsal future synthetic intelligence, and ask ourselves hard questions about what on earth we are going to do with future technology, which might well be even more difficult to control.”
Our reading: We should treat LLMs as a dress rehearsal for future AI, which may be even harder to control.
No horizon
Source: AI risk ≠ AGI risk
Assessment: Can’t be checkedSaid
Forecast
“theirs (from what I can tell) seem to center mainly around what happens if computers rapidly and radically self-improve themselves, which I don’t see as an immediate possibility.”
Our reading: Marcus does not see computers rapidly and radically improving themselves as an immediate possibility.
No horizon Evidence as of
Limitation: This does not establish whether Marcus's technical expectation was reasonable.
Source: AI risk ≠ AGI risk
Assessment: Can’t be checkedSaid
Forecast
“Perhaps coupled with mass AI-generated propaganda, LLM-enhanced terrorism could in turn lead to nuclear war, or to the deliberate spread of pathogens worse than covid-19, etc. Many, many people could die; civilization could be utterly disrupted.”
Our reading: LLM-enhanced terrorism, perhaps coupled with mass AI-generated propaganda, could lead to nuclear war or to the deliberate spread of pathogens worse than covid-19, and many, many people could die and civilization could be utterly disrupted.
No horizon
Qualifier: perhaps coupled with mass AI-generated propaganda
Source: AI risk ≠ AGI risk
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“This cannot and should not stand.”
Our reading: Microsoft and OpenAI's opaque deployment of risky AI systems should not be allowed to continue.
No horizon
Assessment: Can’t be checkedSaid
Conditional forecast
“That claim would never survive serious scientific peer review.”
Our reading: Microsoft's claim that GPT-4 is an early (yet still incomplete) version of an AGI system would never survive serious scientific peer review.
No horizon
Only if: if the claim were subjected to serious scientific peer review
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“We must demand transparency, and if we don’t get it, we must contemplate shutting these projects down.”
Our reading: The public must demand transparency from Microsoft and OpenAI, and if it is not provided, consider shutting their projects down.
No horizon
Qualifier: if transparency is not provided, contemplate shutting projects down
Assessment: Can’t be checkedSaid
Observation
“As I have said before, I don’t really think GPT-4 has much to do with AGI.”
Our reading: Marcus does not really think GPT-4 has much to do with AGI, as he has said before.
No horizon
Qualifier: I don't really think
Assessment: Can’t be checkedSaid
Observation
“So-called “reasoning” by free association, even constrained by a giant database, isn’t really reasoning:”
Our reading: The free-association 'reasoning' LLMs do is not really reasoning, even when constrained by a giant database.
No horizon
Qualifier: even when constrained by a giant database
Source: This Week in AI Doublespeak
Assessment: Can’t be checkedSaid
Observation
“Although GPT-4 is clearly better than GPT-3 and 3.5, it is not so far as I can tell, qualitatively better, only quantitatively better; as noted above, the limits remain more or less the same.”
Our reading: So far as Marcus can tell, GPT-4, though clearly better than GPT-3 and 3.5, is only quantitatively and not qualitatively better, and its limits remain more or less the same.
No horizon Evidence as of
Limitation: This does not reject the checkability of narrower claims about particular limitations or benchmark improvements.
Assessment: Can’t be checkedSaid
Observation
“Massive scaling has not thus far lead to revolution.”
Our reading: As of March 2023, massive scaling has not led to a revolution in AI.
No horizon
Qualifier: thus far
Assessment: Can’t be checkedSaid
Observation
“The quantitative improvements may (or may not) have considerable commercial implications in particular domains, but it’s not the massive win for the scaling hypotheses that memes like this had forecast.”
Our reading: GPT-4's quantitative improvements may or may not have considerable commercial implications in particular domains, but GPT-4 is not the massive win for the scaling hypotheses that memes like the one he cites had forecast.
No horizon
Qualifier: may (or may not)
Assessment: Can’t be checkedSaid
Rhetoric
“One more giant step for hype, but not necessarily a giant step for science, AGI, or humanity.”
Our reading: GPT-4 is a big step for hype but not necessarily for science, AGI, or humanity.
No horizon
Assessment: Can’t be checkedSaid
Causal claim
“and (b) it puts all of us in an extremely poor position to predict what GPT-4 consequences will be for society, if we have no idea of what is in the training set and no way of anticipating which problems it will work on and which it will not.”
Our reading: If we have no idea what is in GPT-4's training set and no way of anticipating which problems it will work on and which it will not, all of us are in an extremely poor position to predict what GPT-4's consequences will be for society.
No horizon
Only if: if we have no idea of what is in the training set
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“It is a step backwards for science, because it sets a new precedent for pretending to be scientific while revealing absolutely nothing.”
Our reading: GPT-4's release is a step backwards for science because it sets a new precedent for pretending to be scientific while revealing absolutely nothing.
No horizon
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“I have caused for something less: stricter regulations governing deployment.”
Our reading: AI deployment should be governed by stricter regulations, rather than a full pause on AI.
No horizon
Source: Should we worry more about short-term AI risk or long-term AI risk?
Assessment: Can’t be checkedSaid
Observation
“Of course I view ChatGPT as only dimly related to artificial general intelligence, but I do see it as dress rehearsal.”
Our reading: ChatGPT is only dimly related to artificial general intelligence, but it serves as a dress rehearsal for it.
No horizon
Source: Should we worry more about short-term AI risk or long-term AI risk?
Assessment: Can’t be checkedSaid
Conditional forecast
“If we don’t start fighting now, democracy may well be overwhelmed by misinformation and consequent polarization—and perhaps quite soon.”
Our reading: If we don't start fighting misinformation now, democracy may well be overwhelmed by misinformation and consequent polarization, perhaps quite soon.
No horizon
Only if: if we don't start fighting now
Source: Should we worry more about short-term AI risk or long-term AI risk?
Assessment: Can’t be checkedSaid
Rhetoric
“The genie is already partway out of the bottle.”
Our reading: LLM technology has already partly escaped containment.
No horizon
Source: Should we worry more about short-term AI risk or long-term AI risk?
Assessment: Can’t be checkedSaid
Forecast
“The 2024 elections could be unlike anything we have seen before.”
Our reading: Because of AI-generated misinformation, the 2024 elections could be unlike any elections seen before.
No horizon
Qualifier: in the context of AI-generated misinformation
Source: Should we worry more about short-term AI risk or long-term AI risk?
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“Either way I think we need to raise these questions seriously, sooner rather than later.”
Our reading: Society should seriously raise questions about pausing or regulating AI, sooner rather than later.
No horizon
Source: Should we worry more about short-term AI risk or long-term AI risk?
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“We can’t just act as if nothing is happening here. And we can’t act like we have infinite time to decide what to do, either.”
Our reading: Society must not ignore AI risks or assume it has unlimited time to decide how to respond.
No horizon
Source: Should we worry more about short-term AI risk or long-term AI risk?
Assessment: Can’t be checkedSaid
Observation
“But there is no real argument against Chomsky’s overall take: these systems are useful technology but remain a long way from true artificial intelligence, and they are even further from telling us anything useful about the function of the human mind.”
Our reading: LLMs are useful technology but remain a long way from true artificial intelligence and even further from telling us anything useful about the human mind.
No horizon
Source: Caricaturing Noam Chomsky
Assessment: Can’t be checkedSaid
Observation
“As I said in the most recent one (not yet aired), at the beginning of my remarks, the only intellectually honest answer is to say we don’t yet know. We don’t know how high the highs are going to be, and we don’t yet know how low the lows are going to be.”
Our reading: It is not yet known whether LLMs will be net positive or net negative.
No horizon
Source: The long shadow of GPT
Assessment: Can’t be checkedSaid
Observation
“And what really scares me is that I don’t think we yet know the half of what that entails.”
Our reading: Marcus does not think we yet know half of what the harms that large language models are almost certain to cause will entail.
No horizon
Qualifier: I don't think
Source: The long shadow of GPT
Assessment: Can’t be checkedSaid
Forecast
“And many of these attacks might of course be combined with advances in voice-cloning tech, which itself is already being applied to scamming, as discussed by the Washington Post on Sunday.”
Our reading: Many of the chatbot-enabled attacks he described might be combined with advances in voice-cloning technology, which is already being used for scams.
No horizon
Scope: many of these attacks
Source: The long shadow of GPT
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“But they are almost certain to cause harm as well, and we need to prepare for that, too.”
Our reading: Society needs to prepare for harms caused by large language models.
No horizon
Source: The long shadow of GPT
Assessment: Can’t be checkedSaid
Forecast
“State actors and extremist groups are likely to use large language models to deliberately mass produce authoritative-sounding misinformation with fake references and data at unprecedented scale, attempting to sway elections and public opinion.”
Our reading: State actors and extremist groups are likely to use large language models to deliberately mass-produce authoritative-sounding misinformation with fake references and data at unprecedented scale, in an attempt to sway elections and public opinion.
No horizon
Qualifier: deliberately
Source: The long shadow of GPT
Assessment: Can’t be checkedSaid
Forecast
“Content farms that are indifferent to the health of their customers may generate interesting-sounding medical content, indifferent as to whether it is true, in order to sell advertising clicks.”
Our reading: Content farms that are indifferent to the health of their customers may generate interesting-sounding medical content, without caring whether it is true, in order to sell advertising clicks.
No horizon
Scope: content farms that are indifferent to the health of their customers
Source: The long shadow of GPT
Assessment: Can’t be checkedSaid
Observation
“But one thing is clear: anybody who thinks we have come to grips with the unknown unknowns here is mistaken.”
Our reading: We have not come to grips with the unknown unknown risks of LLMs.
No horizon
Source: The long shadow of GPT