Claims that can’t be checked by Gary Marcus, page 14 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 14 of 24: claims 1301 to 1400 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
“Justifications for their opposition were meagre, e.g.,”
Our reading: OpenAI's stated justifications for opposing SB-1047, such as its argument that frontier AI regulation should be set at the federal level rather than through state laws, were meagre.
No horizon
Source: Scoop: What former employees of OpenAI are worried about
Assessment: Can’t be checkedSaid
Conditional forecast
“When and if some future form of AI brings new threats, we will be screwed, our collective fates determined by a tiny number of individuals running a few companies, with almost no accountability.”
Our reading: When and if some future form of AI brings new threats, we will be screwed, with our collective fates determined by a tiny number of individuals running a few companies with almost no accountability.
No horizon
Only if: when and if some future form of AI brings new threats
Source: Scoop: What former employees of OpenAI are worried about
Assessment: Can’t be checkedSaid
Conditional forecast
“But he and I fully agree with his overall picture: if we don’t figure out the governance problem, internal and external, before the next big AI advance—whenever it may come—we could be in serious trouble.”
Our reading: If internal and external AI governance isn't solved before the next big AI advance, we could be in serious trouble.
No horizon
Only if: if internal and external governance is not solved before the next big AI advance
Source: Scoop: What former employees of OpenAI are worried about
Assessment: Can’t be checkedSaid
Forecast
“With companies accountable to nothing, not even what they have said to Congress and the White House, self-regulation is never going to work.”
Our reading: With companies accountable to nothing, not even their statements to Congress and the White House, self-regulation is never going to work.
No horizon
Qualifier: companies accountable to nothing, not even what they have said to Congress and the White House
Source: Scoop: What former employees of OpenAI are worried about
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“There’s another problem with this bill too – it’s too narrow, focused mainly on catastrophic harm that causes over a half a billion dollars damaage, and does too little to address other issues, which I fear people may forget, once it is – or is not – passed.”
Our reading: SB-1047 is too narrow, focused mainly on catastrophic harm causing over half a billion dollars in damage, and does too little to address other issues, which Marcus fears people may forget once it is or is not passed.
No horizon
Qualifier: focused mainly on catastrophic harm over half a billion dollars
Assessment: Can’t be checkedSaid
Forecast
“It’s not nothing, but companies worth billions or trillions of dollars may make mincemeat of that standard. Any legal action may take many years to conclude.”
Our reading: Very large companies may overwhelm SB-1047's reasonable care standard, and legal actions under it may take many years.
No horizon
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“We absolutely need a comprehensive approach, and SB 1047 is just a start. And we need federal legislation; a messy state-by-state patchwork is one thing that could actually stifle AI development.”
Our reading: The US needs a comprehensive, federal approach to AI regulation; SB-1047 is just a start.
No horizon
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“But it cannot be the last AI bill to pass, either in California, or the United States, if American citizens are to be protected from the many risks of AI, because it is but one link, already weakened, in the armor that we need in self-defense.”
Our reading: SB-1047 cannot be the last AI bill in California or the US if Americans are to be protected from AI's many risks.
No horizon
Qualifier: if citizens are to be protected from AI risks
Assessment: Can’t be checkedSaid
Causal claim
“a messy state-by-state patchwork is one thing that could actually stifle AI development.”
Our reading: A messy state-by-state patchwork of AI laws could stifle AI development.
No horizon
Assessment: Can’t be checkedSaid
Observation
“To my mind, many of these arguments were based on misrepresentation, as I have discussed here in earlier essays.”
Our reading: In Marcus's view, many of the arguments made against SB-1047 by Andreessen Horowitz and others were based on misrepresentation.
No horizon Evidence as of
Limitation: This does not establish that opponents' arguments were accurate.
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“Hopefully, SB 1047 will be signed into law, and the perfectly rational, and in fact entirely ordinary, notion of regulating AI will be normalized, and the absurd idea that such regulation would kill off AI will have vanished.”
Our reading: Marcus hopes SB-1047 will be signed into law, that the notion of regulating AI, which he calls perfectly rational and entirely ordinary, will be normalized, and that the idea that such regulation would kill off AI, which he calls absurd, will vanish.
No horizon
Qualifier: hopefully
Assessment: Can’t be checkedSaid
Causal claim
“The bill also sends a message to big tech that the quality of safety protocols, and internal investigations of safety, are relevant to the reasonable care standard — which can help to avoid Ford Pinto-like situations in which companies that should have known better fail to act.”
Our reading: SB-1047 sends big tech a message that the quality of safety protocols and internal safety investigations matter to the reasonable care standard, which can help avoid Ford Pinto-like situations in which companies that should have known better fail to act.
No horizon
Qualifier: can help
Assessment: Can’t be checkedSaid
Forecast
“Companies may simply roll the dice, and as Eric Schmidt recently said, let the lawyers “clean up the mess” after the fact.”
Our reading: AI companies may gamble on risky behavior and let lawyers handle consequences afterward despite SB-1047.
No horizon
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“Still I support the bill, even in weakened form.”
Our reading: SB-1047 deserves support even in its weakened form.
No horizon
Qualifier: even in weakened form
Assessment: Can’t be checkedSaid
Forecast
“Companies like OpenAI that have talked big about investing in safety research and then stinted on their promises may finally be incentivitized to walk the walk that their talk has promised.”
Our reading: Under SB-1047, including the specter of it, companies like OpenAI that have talked big about investing in safety research and then stinted on their promises may finally be incentivized to walk the walk their talk promised.
No horizon
Scope: companies like OpenAI that stinted on safety promises
Assessment: Can’t be checkedSaid
Causal claim
“SB-1047’s strongest utility may come as a deterrent.”
Our reading: SB-1047's main value may be deterring irresponsible AI development by clarifying liability.
No horizon
Qualifier: even if nobody is ever penalized
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“So it must pass. I hope the full CA Assembly will support it, and that Governor Newsom will sign it into law, rather than veto it.”
Our reading: SB-1047 must pass the California Assembly and be signed by Governor Newsom rather than vetoed.
No horizon
Assessment: Can’t be checkedSaid
Forecast
“At the very least, I foresee a significant reframing of expectations.”
Our reading: Expectations for AI will at the very least be significantly reframed, even if a full AI winter does not arrive.
No horizon
Qualifier: at the very least, whether or not an AI winter comes
Source: What has and has not changed in the AI since the ChatGPT revolution? [Video]
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“With no other serious proposal on offer, I personally favor SB-1047, though I would welcome discussion of alternatives.”
Our reading: With no other serious proposal on offer, Marcus personally favours SB-1047, though he would welcome discussion of alternatives.
No horizon
Qualifier: no other serious proposal on offer
Source: An open letter to Fei-Fei Li concerning California’s proposed AI regulation, SB-1047
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“Those seem reasonable to me, and I don't understand what would justify an exemption there.”
Our reading: Marcus finds it reasonable for SB-1047 to focus on specific critical harms, and he does not understand what would justify an exemption from those provisions.
No horizon
Scope: critical harms such as mass casualties, weapons of mass destruction, large-scale cyberattacks, and AI models autonomously committing serious felonies
Source: An open letter to Fei-Fei Li concerning California’s proposed AI regulation, SB-1047
Assessment: Can’t be checkedSaid
Rhetoric
“⚠️ Something terrifying just snapped in place:”
Our reading: Something terrifying has just snapped into place, as shown by the developments listed after this line.
No horizon
Source: OpenAI: On a Path to Becoming The World’s Most Frightening Surveillance Company?
Assessment: Can’t be checkedSaid
Forecast
“He might succeed.”
Our reading: Sam Altman might succeed in knowing and monetizing everything about people.
No horizon
Source: OpenAI: On a Path to Becoming The World’s Most Frightening Surveillance Company?
Assessment: Can’t be checkedSaid
Observation
“Altman wants to know - and monetize - everything about you.”
Our reading: Sam Altman wants to know everything about people and make money from it.
No horizon
Source: OpenAI: On a Path to Becoming The World’s Most Frightening Surveillance Company?
Assessment: Can’t be checkedSaid
Observation
“Prospects don’t seem as strong as they once did.”
Our reading: OpenAI's prospects of justifying its $80B valuation or raising at a higher valuation look weaker than before.
No horizon
Qualifier: justifying $80B valuation and raising future rounds
Source: The OpenAI Plot Thickens
Assessment: Can’t be checkedSaid
Forecast
“I said it before, and I will say it again: OpenAI could wind up being seen as the WeWork of AI.”
Our reading: OpenAI could end up being seen as the WeWork of AI.
No horizon
Source: The OpenAI Plot Thickens
Assessment: Can’t be checkedSaid
Causal claim
“Adding AI into the mix could easily [make things] a lot worse”
Our reading: Adding AI into the mix of social media could easily make things a lot worse.
No horizon
Scope: social media context
Assessment: Can’t be checkedSaid
Forecast
“But it is also not out of the question that today could end someday end up being seen as a turning point.”
Our reading: It is not out of the question that August 5, 2024 could someday end up being seen as a turning point.
No horizon
Qualifier: not out of the question
Assessment: Can’t be checkedSaid
Forecast
“The most interesting thing is that the suit could force a discussion of what AGI means – in court.”
Our reading: Elon Musk's new lawsuit against OpenAI could force a court to discuss what AGI means.
No horizon
Assessment: Can’t be checkedSaid
Causal claim
“This could pertain to production delays, but also to growing market concerns about whether NVidia’s big GenAI customers could make money deploying the chips.”
Our reading: Nvidia's stock drop could be due to production delays and also to growing market concern about whether its big GenAI customers can make money deploying its chips.
No horizon
Assessment: Can’t be checkedSaid
Causal claim
“The people who have temporarily gotten rich or famous on AI have done so by pretending that this outlier problem simply doesn’t exist, or that a remedy for it is imminent.”
Our reading: The people who have temporarily gotten rich or famous on AI have done so by pretending that the outlier problem doesn't exist or that a remedy for it is imminent.
No horizon
Scope: people who have temporarily gotten rich or famous on AI
Source: This one important fact about current AI explains almost everything
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“Everyone commenting on or thinking about AI should learn it by heart, if not verbatim, certainly the gist.”
Our reading: Everyone commenting on or thinking about AI should learn by heart, if not word for word then at least the gist, Koopman's point that machine learning trains on many examples and, when it meets something it lacks many examples of, may have no idea what is going on.
No horizon
Scope: everyone commenting on or thinking about AI
Source: This one important fact about current AI explains almost everything
Assessment: Can’t be checkedSaid
Rhetoric
“You can’t think we are close to AGI once you realize that as yet we have no general solution to the outlier problem. You just can’t.”
Our reading: Marcus asserts emphatically that no one who realizes we do not yet have a general solution to the outlier problem can think we are close to AGI.
No horizon
Qualifier: once you realize there is as yet no general solution to the outlier problem
Source: This one important fact about current AI explains almost everything
Assessment: Can’t be checkedSaid
Rhetoric
“Once you do, almost everything that people like Altman and Musk and Kurzweil are currently saying about AGI being nigh seems like sheer fantasy, on par with imagining that really tall ladders will soon make it to the moon.”
Our reading: Once one understands that current neural networks struggle with outliers, almost everything that people like Altman, Musk and Kurzweil are currently saying about AGI being near seems like sheer fantasy, like imagining that really tall ladders will soon reach the moon.
No horizon
Qualifier: once you understand that neural networks struggle with outliers
Source: This one important fact about current AI explains almost everything
Assessment: Can’t be checkedSaid
Causal claim
“If it’s not directly in the training set, but merely superficially like some previous examples, there could be trouble.”
Our reading: When an input is not directly in the training set but only superficially resembles earlier examples, machine learning systems may fail.
No horizon
Qualifier: input not directly in the training set but superficially similar to previous examples
Source: This one important fact about current AI explains almost everything
Assessment: Can’t be checkedSaid
Rhetoric
“One step backwards could open up ten forwards.”
Our reading: The GenAI setback could open the way to much greater progress in AI through other approaches.
No horizon
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“Incredibly damning new study on GenAI from the Upwork Research Institute, via Baldur Bjarnason (whose wonderful essay on LLMs and how they fool the gullible I discussed in April).”
Our reading: The new Upwork Research Institute study on generative AI is incredibly damning.
No horizon
Qualifier: via Baldur Bjarnason
Source: Most people don’t think GenAI has improved their productivity
Assessment: Can’t be checkedSaid
Conditional forecast
“I think that GenAI is vastly overrated and overhyped, but fear that its collapse may well lead to an AI winter of sorts, like what happened in the mid-1980s, when AI “expert systems” rapidly ascended and rapidly fell.”
Our reading: GenAI is vastly overrated, and its collapse may well lead to an AI winter of sorts.
No horizon
Only if: if GenAI collapses
Source: AlphaProof, AlphaGeometry, ChatGPT, and why the future of AI is neurosymbolic
Assessment: Can’t be checkedSaid
Observation
“The current, short-term investment climate is thus hugely biased against genuine novelty and outside the box thinking.”
Our reading: The current AI investment climate is hugely biased against genuine novelty.
No horizon Evidence as of
Limitation: The Crunchbase figure covers all of 2024 and was published after the post. The claim is about the climate in July 2024.
Source: AlphaProof, AlphaGeometry, ChatGPT, and why the future of AI is neurosymbolic
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“Since LLMs inevitably hallucinate and are constitutionally incapable of checking their own work, there are really only two possibilities: we abandon them, or we use them as components in larger systems that can reason and plan better”
Our reading: LLMs should either be abandoned or used only as components within larger systems that reason and plan better.
No horizon
Qualifier: since LLMs inevitably hallucinate
Source: AlphaProof, AlphaGeometry, ChatGPT, and why the future of AI is neurosymbolic
Assessment: Can’t be checkedSaid
Forecast
“Ditto for Anthropic, and a number of others.”
Our reading: Anthropic and several other LLM companies may be fatally undercut by Meta's free models.
No horizon
Qualifier: Meta giving away similar tech for free
Assessment: Can’t be checkedSaid
Causal claim
“That’s not great news for OpenAI, and you can see why they haven’t been, um, Open, about their financials (reporting only revenue but not costs and profits), if The Information is even close to right.”
Our reading: If The Information's analysis is roughly right, OpenAI's losses explain why it reports only revenue and not costs or profits.
No horizon
Qualifier: if The Information's analysis is close to right
Assessment: Can’t be checkedSaid
Observation
“I have long suspected that OpenAI was losing money, and lots of it, but never seen an analysis, until this morning.”
Our reading: Marcus has long suspected that OpenAI was losing a lot of money, though until The Information's analysis he had not seen an analysis showing it.
No horizon
Qualifier: long suspected
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“But investors really ought to ask some tough questions, such as these:”
Our reading: Investors should press OpenAI on its moat, unique technology, killer app, reliability, demo versus reality, and path to profitability.
No horizon
Assessment: Can’t be checkedSaid
Forecast
“All of these questions are hard, with no obvious answer; the last may be fatal. OpenAI may well be Zuck’d.”
Our reading: OpenAI may fail to find a path to profitability because Meta gives away similar technology for free, and this may be fatal to the company.
No horizon
Qualifier: Meta giving away similar tech for free
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“Chasing black box AI, difficult to interpret, and difficult to debug, is not the answer.”
Our reading: Pursuing black-box AI that is hard to interpret and debug is not the right approach to building reliable software.
No horizon
Source: Don’t look up: the massive Microsoft/Crowdstrike data outage is a huge wake-up call
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“The world needs to up its software game massively. We need to invest in improving software reliability and methodology, not rushing out half-baked chatbots.”
Our reading: The world should invest in improving software reliability and methodology rather than rushing out half-baked chatbots.
No horizon
Source: Don’t look up: the massive Microsoft/Crowdstrike data outage is a huge wake-up call
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“And leaving more and more code writing to generative AI, which grasps syntax but not meaning, as a new benchmark from Stephen Wolfram shows, is not the answer, either.”
Our reading: Delegating more and more code writing to generative AI is not the right approach.
No horizon
Source: Don’t look up: the massive Microsoft/Crowdstrike data outage is a huge wake-up call
Assessment: Can’t be checkedSaid
Forecast
“Spending upwards of 100 billion dollars on the current approach, which is what two more iterations of scaling would likely require, given that is unlikely to get to AGI (or ever be reliable) seems wasteful to me.”
Our reading: The current scaling approach is unlikely ever to reach AGI or become reliable.
No horizon
Source: Marcus goes gaga over Gates clip
Assessment: Can’t be checkedSaid
Forecast
“Spending upwards of 100 billion dollars on the current approach, which is what two more iterations of scaling would likely require, given that is unlikely to get to AGI (or ever be reliable) seems wasteful to me.”
Our reading: Two more iterations of scaling would likely require upwards of $100 billion in spending.
No horizon
Source: Marcus goes gaga over Gates clip
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“I fully 100% agree, and have said as much before.”
Our reading: Marcus fully agrees with Gates that, after perhaps two more turns of the crank for scaling, scaling is not the most interesting dimension and AI needs techniques such as metacognition, and says he has said as much before.
No horizon
Qualifier: after that (after perhaps two more turns of scaling)
Source: Marcus goes gaga over Gates clip
Assessment: Can’t be checkedSaid
Causal claim
“Not unrelatedly, meanwhile, many tech leaders have discovered that the best way to raise valuations is to hint that AGI is imminent, always on the dubious premise that scaling will keep going indefinitely; it behooves them financially to pretend.”
Our reading: Many tech leaders hint that AGI is imminent in order to raise valuations, relying on the dubious premise that scaling continues indefinitely.
No horizon
Source: Marcus goes gaga over Gates clip
Assessment: Can’t be checkedSaid
Forecast
“AI is going to change everything, if not tomorrow, sometime over the next 5-20 years, some ways for good, some for bad.”
Our reading: AI will change everything, for good and bad, sometime within the next 5 to 20 years.
No horizon
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“And we need a President who can get Congress (which has this far dithered) to recognize the true urgency of the moment, since Executive Orders alone are not enough.”
Our reading: We need a President who can get Congress, which has thus far dithered, to recognize the true urgency of the moment on AI, since executive orders alone are not enough.
No horizon
Scope: in the context of US AI policy
Assessment: Can’t be checkedSaid
Observation
“the whole Generative AI thing is looking more and more like a dud.”
Our reading: The whole generative AI thing is looking more and more like a dud.
No horizon
Qualifier: more and more (increasing trend)
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“We need a President who can recognize when corporate leaders are promising things far beyond is currently realistic, who can sort truth from bullshit, in order to develop AI policies that are grounded in reality.”
Our reading: The US needs a President who can tell when corporate AI promises are unrealistic, so that AI policy is grounded in reality.
No horizon
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“We also need a President who can stand up to big tech, and resist regulatory capture.”
Our reading: The US needs a President who stands up to big tech and resists regulatory capture.
No horizon
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“We cannot afford to have a President in 2024 that doesn’t fully grasp this.”
Our reading: The US must elect a President in 2024 who fully grasps AI's transformative impact.
No horizon
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“We do need new approaches to AI.”
Our reading: New approaches to AI are needed.
No horizon
Assessment: Can’t be checkedSaid
Forecast
“OpenAI may in fact yet turn out to be the WeWork of AI.”
Our reading: OpenAI may yet turn out to be the WeWork of AI.
No horizon
Assessment: Can’t be checkedSaid
Observation
“It would now appear that Business Insider’s sources were confused, or overstating what they knew.”
Our reading: Business Insider's sources on an imminent GPT-5 release appear to have been confused or overstated their knowledge.
No horizon
Assessment: Can’t be checkedSaid
Observation
“(Probably for the third or fourth time, if my hunch is correct, with GPT 4-Turbo and 4o perhaps being failed attempts that didn’t seem worthy of the name, with another earlier attempt halting before Sam’s time speaking to the Senate in May 2023.)”
Our reading: OpenAI has probably started training GPT-5 three or four times, with GPT-4 Turbo and GPT-4o perhaps being failed attempts not worthy of the name.
No horizon
Qualifier: if his hunch is correct
Assessment: Can’t be checkedSaid
Forecast
“Expect more revisionism and downsized expectations throughout 2024 and 2025.”
Our reading: During 2024 and 2025, leading AI figures will further revise and lower their public expectations for AI, such as AGI timelines.
Deadline passed Evidence as of
Limitation: On 22 June 2024 Marcus added an update to the post that, in his words, 'somewhat undermines the thesis of this essay': Kurzweil wrote to him that he still expects AGI by 2029.
Assessment: Can’t be checkedSaid
Observation
“It was always going to happen; the ludicrously high expectations from last 18 ChatGPT-drenched months were never going to be met.”
Our reading: The ludicrously high expectations for AI from the previous 18 ChatGPT-dominated months were never going to be met, and their failure was always going to happen.
No horizon
Scope: expectations from the last 18 ChatGPT-drenched months
Assessment: Can’t be checkedSaid
Rhetoric
“Big tech seemingly will not be content until they have squeezed out every last cent of profit, whilst socializing the costs of every last risk.”
Our reading: Marcus says, in deliberately exaggerated terms, that Big Tech seemingly will not be content until it has squeezed out every last cent of profit while socializing the costs of every last risk.
No horizon
Qualifier: seemingly
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“None of that is unreasonable, and the CEO of OpenAI himself said that we need regulation on AI.”
Our reading: The requirements of SB-1047 (safety testing, liability in extreme cases, a shutdown path) are not unreasonable.
No horizon
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“We should be making SB-1047 stronger, not weaker.”
Our reading: SB-1047 should be strengthened rather than weakened.
No horizon
Assessment: Can’t be checkedSaid
Observation
“Fine to talk big in front of the Senate, but when push comes to shove, nobody in the industry seems to actually want the kind of regulation they previously hinted at.”
Our reading: Nobody in the AI industry appears to actually want the regulation they previously hinted at supporting before the Senate.
No horizon
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“The thing is, we should be worried, no matter how much time we have.”
Our reading: We should be worried about AGI preparedness regardless of how long until AGI arrives.
No horizon
Qualifier: no matter how much time we have
Source: AGI by 2027?
Assessment: Can’t be checkedSaid
Observation
“we are (as Eliezer Yudkowsky has rightly argued for years) woefully underprepared for AGI whenever it comes.”
Our reading: We are woefully underprepared for AGI, whenever it comes, as Eliezer Yudkowsky has rightly argued for years.
No horizon
Qualifier: whenever AGI comes
Source: AGI by 2027?
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“It’s imperative that we do better.”
Our reading: It is imperative that we do better at reining in Silicon Valley than we have so far.
No horizon
Scope: in the context of the self-regulation and legislative failures just described
Source: AGI by 2027?
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“Gary Marcus wishes AI these days was more about research than money, politics, and ego.”
Our reading: Gary Marcus wishes that AI these days were more about research than about money, politics, and ego.
No horizon
Scope: AI these days
Source: AI as Reality Television
Assessment: Can’t be checkedSaid
Observation
“The irony of all of this is that you and I are among the minority of people who have come to fully understand just how limited LLMs are, and what we need to do next.”
Our reading: Marcus and LeCun are among the minority of people who have come to fully understand just how limited LLMs are and what needs to be done next.
No horizon
Scope: among the minority of people
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“Of course my biggest (and most difficult to confirm or reject) positive conjectures, all made in 2001 in The Algebraic Mind, remain speculative: the importance of representing kinds distinctly from individuals, separating reasoning from latent representations, incorporating explicit compositional structure and explicit operations over variables, incorporating some degree of innate pre learning, and exploiting explicit knowledge databases. I stand by all of them.”
Our reading: Marcus stands by his speculative 2001 conjectures, which he says are hard to confirm or reject, that it is important to represent kinds distinctly from individuals, separate reasoning from latent representations, incorporate explicit compositional structure and explicit operations over variables, incorporate some degree of innate pre-learning, and exploit explicit knowledge databases.
No horizon
Qualifier: conjectures remain speculative and difficult to confirm or reject
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“To take a different kind of example, I have long argued that models need to innately be more structured than pure multilayer perceptions, even prior to learning.”
Our reading: AI models need more innate structure than pure multilayer perceptrons, even before learning.
No horizon
Qualifier: prior to learning
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“The letter itself, cosigned by Bengio, Hinton, and Russell, can be found here https://righttowarn.ai. I fully endorse its four recommendations:”
Our reading: Marcus fully endorses the four recommendations of the righttowarn.ai letter by OpenAI insiders, cosigned by Bengio, Hinton and Russell, about AI employees' right to warn of risks.
No horizon
Source: “OpenAI Insiders Warn of a ‘Reckless’ Race for Dominance”
Assessment: Can’t be checkedSaid
Observation
“For not so good measure, one of the few people who recognized, early, the potential dangers of unaligned AGI, Musk, has switched teams, flipping from calling for a pause to going all in on—and wildly overhyping—a technology that remains exactly as incorrigible as it ever was.”
Our reading: Current AI technology, which Musk is now wildly overhyping, remains exactly as incorrigible as it ever was.
No horizon
Assessment: Can’t be checkedSaid
Observation
“But here’s the thing, what I said then to Bach still holds, 100%, 26 months later”
Our reading: What Marcus said to Joscha Bach 26 months earlier still holds, 100%.
No horizon
Qualifier: 26 months later
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“The Board’s job is to make sure that doesn’t happen — not to cover Sam’s ass.”
Our reading: OpenAI's board should work to prevent AI causing significant harm to the world rather than protecting Altman.
No horizon
Source: Disturbing and misleading efforts to defame Helen Toner through misdirection
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“If they cannot assemble a board that respects the legal filings they made, and cannot behave in keeping with their oft-repeated promises, they must dissolve the nonprofit, exactly as the organization Public Citizen has argued.”
Our reading: If OpenAI cannot assemble a board that respects its legal filings and keep its promises, it must dissolve the nonprofit.
No horizon
Qualifier: if they cannot assemble a board respecting their legal filings
Source: Disturbing and misleading efforts to defame Helen Toner through misdirection
Assessment: Can’t be checkedSaid
Observation
“OpenAI does not behave like a nonprofit and has not for a long time.”
Our reading: OpenAI does not behave like a nonprofit and has not for a long time.
No horizon
Source: Disturbing and misleading efforts to defame Helen Toner through misdirection
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“The degree to which they diverted from that core issue that led to Sam’s firing is genuinely disturbing, given that the board is our last and only line of defense if Sam chooses for economic reasons to do something genuinely dangerous.”
Our reading: The degree to which the new board members diverted from the core issue that led to Altman's firing is genuinely disturbing, because the board is the last and only line of defense if Altman chooses, for economic reasons, to do something genuinely dangerous.
No horizon
Qualifier: if Altman chooses for economic reasons to do something genuinely dangerous
Source: Disturbing and misleading efforts to defame Helen Toner through misdirection
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“The failure of two members of the new board (include Taylor, who chairs that Board) to draw that connection, coupled with their misdirection and unfair attack on Toner, is absolutely terrifying.”
Our reading: The failure of two new board members, including chair Taylor, to draw the connection between candor and safety, together with their misdirection and unfair attack on Toner, is absolutely terrifying.
No horizon
Scope: two members of the new board, including Taylor
Source: Disturbing and misleading efforts to defame Helen Toner through misdirection
Assessment: Can’t be checkedSaid
Conditional forecast
“Of course, there is a connection between safety and candor, and an obvious one at that: if Altman can’t be trusted, safety could become a serious issue.”
Our reading: If Altman can't be trusted, safety at OpenAI could become a serious issue.
No horizon
Only if: if Altman can't be trusted
Source: Disturbing and misleading efforts to defame Helen Toner through misdirection
Assessment: Can’t be checkedSaid
Observation
“As far as I can tell, Toner and the earlier board acted in good faith.”
Our reading: As far as Marcus can tell, Toner and the earlier OpenAI board acted in good faith.
No horizon
Qualifier: as far as I can tell
Source: Disturbing and misleading efforts to defame Helen Toner through misdirection
Assessment: Can’t be checkedSaid
Observation
“The fact that the Board agreed in November (when they did not see an absolute safety issue of the sort that would require the end of the company) to dissolve itself and to restore Altman to CEO rather than let the company dissolve (once virtually all the employees threatened to leave) shows that their preference for keeping the company was genuine, and that they acted in good faith.”
Our reading: The former OpenAI board genuinely preferred to keep the company intact and acted in good faith in November 2023.
No horizon
Source: Disturbing and misleading efforts to defame Helen Toner through misdirection
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“The Board should have been apprised that this was in the works.”
Our reading: OpenAI's board should have been told in advance that ChatGPT was being launched.
No horizon
Source: Disturbing and misleading efforts to defame Helen Toner through misdirection
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“If OpenAI does eventually develop deeply risky technology, it is incumbent on the nonprofit’s board to address that, and if they can’t trust Altman, I don’t see they can do their job.”
Our reading: If OpenAI eventually develops deeply risky technology, the nonprofit's board has a duty to address it, and Marcus does not see how the board can do that job if it cannot trust Altman.
No horizon
Only if: if OpenAI eventually develops deeply risky technology
Source: Disturbing and misleading efforts to defame Helen Toner through misdirection
Assessment: Can’t be checkedSaid
Observation
“On the other hand, the chatbots of both X.ai and Meta still seem to me to be wildly inadequate, hallucinatory machines that merely hint at what AI could be, but with precious little understanding of the concepts underlying the words they so fluently use.”
Our reading: To Marcus, the chatbots of X.ai and Meta still seem to be wildly inadequate, hallucinatory machines that merely hint at what AI could be and have precious little understanding of the concepts behind the words they fluently use.
No horizon Evidence as of
Limitation: This does not reject or vindicate the factual allegation that the chatbots hallucinated.
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“Musk and LeCun may have the soapboxes and the money, but what we really need is a breakthrough.”
Our reading: What AI really needs is a breakthrough, not just money and publicity.
No horizon
Assessment: Can’t be checkedSaid
Observation
“Both Musk and LeCun (via Meta) command massive budgets to explore AI; I am not convinced that either is making real progress.”
Our reading: Marcus is not convinced that either Musk's xAI or Meta is making real AI progress despite massive budgets.
No horizon
Qualifier: despite massive budgets
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“we need more science if we are to get to AGI.”
Our reading: Reaching AGI requires more basic science.
No horizon
Qualifier: if AGI is to be reached
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“AI needs more science, but it needs engineering, too.”
Our reading: AI needs more science as well as engineering.
No horizon
Assessment: Can’t be checkedSaid
Observation
“Right now we have a lot of engineering and alchemy, but not enough basic principles and formal understanding.”
Our reading: Right now AI has a lot of engineering and alchemy but not enough basic principles and formal understanding.
No horizon
Scope: right now
Assessment: Can’t be checkedSaid
Causal claim
“Probably Sam would still own it if not for Primack’s digging, and we would have one more conflict of interest to worry about.”
Our reading: Altman probably divested from the OpenAI Startup fund because of Dan Primack's reporting and would otherwise still own it.
No horizon
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“That’s pretty scary too, given what OpenAI aspires to build.”
Our reading: Altman giving the board inaccurate information on multiple occasions about OpenAI's few formal safety processes is pretty scary, given what OpenAI aspires to build.
No horizon
Qualifier: given what OpenAI aspires to build
Assessment: Can’t be checkedSaid
Observation
“That’s insane. And underscores everything I have said about the company straying from its nonprofit mission.”
Our reading: OpenAI not informing its board in advance of ChatGPT's release shows the company has strayed from its nonprofit mission.
No horizon
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“Particularly as long as the US is relying essentially on self-regulation and Sam is aspiring to build ever more powerful models with less and less independent oversight, that should concern us all, deeply.”
Our reading: Altman consolidating more and more power while seeming less and less on the level should deeply concern us all, particularly as long as the US relies essentially on self-regulation and Altman aspires to build ever more powerful models with less and less independent oversight.
No horizon
Qualifier: particularly as long as the US is relying essentially on self-regulation
Assessment: Can’t be checkedSaid
Forecast
“Reading between the lines, Sam may well wind up with direct equity in the end, after all:”
Our reading: Sam Altman may eventually end up holding direct equity in OpenAI.
No horizon
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“Because we can’t have the CEO of a company that potentially powerful behaving that way.”
Our reading: The CEO of a potentially powerful AI company like OpenAI should not be allowed to behave deceptively toward his board.
No horizon
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“At this point, I think it is correct for the public to take everything OpenAI says with a grain of salt.”
Our reading: The public should take everything OpenAI says with a grain of salt.
No horizon
Source: What should we learn from OpenAI’s mistakes and broken promises?