Claims checked against evidence, page 31 of 31
Claims by Gary Marcus and Ed Zitron. Deadlines are judged as of 2026-10-29.
Claims checked against evidence (3100)
Each of these claims was checked against dated evidence.
Page 31 of 31: claims 3001 to 3100 of 3100, ordered by the date first said, newest first. To find a claim by its words, use search.
Assessment: SupportedSaid
“Over and over I have stressed that you could have a good model of engineering without being a good model of humans, or vice versa.”
What happened: Faggella's interview report describes Marcus contrasting evolved human memory with more reliable computer memory. It records an earlier example of engineering desirability differing from human fidelity; the quotation preserves the source's apparent typo. (Emerj; Dan Faggella, )
Assessment: SupportedSaid
“But the smart money is still on being cautious about this, and in recognizing that what the Facebook researchers said in 2019 is still the case.”
What happened: Reducing misleading correlations improved independent-test performance for BERT and stronger RoBERTa and ALBERT models, supporting continued susceptibility in 2022. The same experiments found better generalization from larger models and debiasing, which limits any claim that no genuine progress occurred. (arXiv; ACL 2022 author manuscript, )
Assessment: UnresolvedSaid
Assessment: SupportedSaid
“On #2, the answer is mixed. They do well on some inputs (like astronaut rides horse) but more poorly on others (like horse rides astronaut, which I discussed in an earlier post).”
What happened: The study's human judgments show some successful matching alongside widespread relational failures. Its results were reported by July; the precise image-generation dates are not supplied. (arXiv; Colin Conwell and Tomer Ullman, )
Assessment: UnresolvedSaid
Assessment: UnresolvedSaid
Assessment: UnresolvedSaid
Assessment: UnresolvedSaid
Assessment: UnresolvedSaid
Assessment: SupportedSaid
“The dirty secret in current AI is that for all the bloviating about how current systems have mastered human language, they are still really weak on compositionality, which most linguists (since Frege, over a century ago) would agree is at the very core of how language works.”
What happened: Winoground found large human-model gaps across several vision-language model families on image-caption matching that depends on composition. The finding supports a broad weakness within the tested contemporary systems. (arXiv; Thrush and colleagues, )
Assessment: SupportedSaid
“The progress that Alexander is vaunting shows that the system did better on a handful of items than DALL-E2 (which was not even state of the art at compositionality when it appeared), but not that the newer system (Imagen) systematically groks compositionality”
What happened: The page specifies five prompts, reports DALL-E 2 at 0/5 and Imagen at 3/5 under Alexander's scoring, and contains no systematic retest of earlier successes. Alexander also expressly declines to claim full mastery, so the methodological limitation does not establish that he claimed otherwise. (Astral Codex Ten, )
Assessment: ContradictedSaid
“we were right to be skeptical; so far, four and half years later, Duplex hasn’t gotten a whole lot further.)”
What happened: Google reported over a million completed bookings and over three million business-information updates across eight countries, alongside the quoted automation rate. These reported deployments show substantial operational progress beyond a limited demonstration. (Google, )
Assessment: UnresolvedSaid
Assessment: SupportedSaid
“This should be a piece of cake, but for currently popular systems, it’s not.”
What happened: The group score requires all relevant caption-image comparisons to be correct; every tested model scored below its random baseline. Some models exceeded chance on the less demanding text score, but remained far behind humans and did not reliably solve the complete matching task. (arXiv; Thrush and colleagues, )
Assessment: UnresolvedSaid
Assessment: UnresolvedSaid
Assessment: SupportedSaid
“Driverless car demos have been around for decades, but making them work reliably has proven far harder than most people realized.”
What happened: The report records a direct industry admission and describes executives narrowing expectations because of technical complexity. A readable primary transcript was not obtained, so contemporary reporting is used. (WIRED, )
Assessment: UnresolvedSaid
Assessment: SupportedSaid
“Performance hovered at around 75% in the best case, when the robots had a limited number of options to consider.”
What happened: The reported overall execution rates are 74% in the mock kitchen and 60% in the real kitchen. Thus the better overall setting is approximately 75%; planning is a separate measure. (arXiv; Michael Ahn and colleagues, )
Assessment: UnresolvedSaid
Assessment: SupportedSaid
“First, the language technology that the new system relies on is well-known to be problematic and second it is likely to be even more problematic in the context of robots.”
What happened: PaLM's own announcement acknowledges an existing research literature on risks and describes analyses of bias and model outputs. This directly supports recognition of problems in the relevant technology. (Google Research, )
Assessment: UnresolvedSaid
Assessment: UnresolvedSaid
Assessment: SupportedSaid
“it’s even less clear how it would work with words like no and not (a major sore spot for current neural networks).”
What happened: The study reports substantial RoBERTa performance losses on subsets where negation matters: QNLI macro F1 is 0.67 for important negations versus 0.92 for unimportant negations. These are dataset-subset comparisons, not matched sentence pairs; the findings were already reported in the March preprint. (Association for Computational Linguistics, )
Assessment: UnresolvedSaid
Assessment: UnresolvedSaid
Assessment: UnresolvedSaid
Assessment: UnresolvedSaid
Assessment: UnresolvedSaid
Assessment: UnresolvedSaid
Assessment: UnresolvedSaid
Assessment: UnresolvedSaid
Assessment: SupportedSaid
“Since then, the results on benchmarks have gotten better, but there’s still something missing.”
What happened: PaLM's contemporary report documents continued benchmark improvements. This supports the improvement component, without establishing that benchmark success implies general understanding. (arXiv; Chowdhery et al., )
Assessment: ContradictedSaid
“(LLM’s like GPT produce well-formed prose but do not produce interpretable representations of utterances that reflect structured relationships between the parts of those sentences.)”
What happened: The study uses late-2021 GPT-3 and Codex models to map utterances into structured meaning representations. It demonstrates a task capability, while its use of prompts and decoding machinery limits conclusions about ordinary internal representations. (arXiv; Richard Shin and Benjamin Van Durme, )
Assessment: UnresolvedSaid
Assessment: SupportedSaid
“noncanonical reorderings such as these remain problematic for current AI approaches.)”
What happened: Winoground requires matching two images to captions containing the same words in different orders. Its poor group scores across contemporary vision-language models support substantial difficulty preserving the changed relationships. (Thrush et al., CVPR 2022 / arXiv, )
Assessment: UnresolvedSaid
Assessment: UnresolvedSaid
Assessment: UnresolvedSaid
Assessment: ContradictedSaid
“Current systems simply do not postulate such cognitive models—and so, they technically can’t prove themselves wrong, because they do not even have a representation of what being “right” would look like.”
What happened: The experiments evaluate models critiquing their own outputs against supplied passages or task constraints, and using critiques to improve answers. This contradicts the functional claim of complete inability, while leaving the nature of internal cognitive representations unsettled. (arXiv; William Saunders and colleagues, OpenAI, )
Assessment: ContradictedSaid
“The downstream consequence of all of this is that these models have no grasp whatsoever on truth.”
What happened: GPT-judge classified the factual truth of answers and transferred across held-out answer-producing model families. This contradicts absolute behavioral incapacity, but all benchmark questions were represented in training, so it does not establish unrestricted factual understanding. (Association for Computational Linguistics; Stephanie Lin, Jacob Hilton and Owain Evans; ar5iv mirror, )
Assessment: UnresolvedSaid
Assessment: UnresolvedSaid
Assessment: UnresolvedSaid
Assessment: SupportedSaid
“Because it shows that the people who really claim to be working on full self-driving (which Tesla waffles on1) are having plenty of problems of their own.”
What happened: The record confirms repeated ADS crash involvement in the initial reporting period. Its abstract defines reportability by ADS use within 30 seconds of a crash, not by a finding of ADS fault. (Transportation Research Board, summarizing NHTSA, )
Assessment: UnresolvedSaid
Assessment: SupportedSaid
“Level 5 has to handle it all; everyday stopped vehicles, weird outliers and so forth. We’re just not there.”
What happened: The contemporaneous report directly addresses the claimed industry-wide state. Reputable reporting is used because no readable primary source establishing that complete industry-wide negative was obtained. (The Verge, )
Assessment: SupportedSaid
“Indeed, the NHTSA report data show that even some pretty ordinary circumstances, like avoiding stopped vehicles, are continuing to prove problematic.”
What happened: The surrounding summary identifies these as additional Autopilot collisions with stopped first-responder vehicles. Its incident list dates Orlando to August 2021, Petaluma to September 2021, and Desert Center to January 2022; the event date records the latest of these in-window incidents. (National Highway Traffic Safety Administration, )
Assessment: UnresolvedSaid
Assessment: UnresolvedSaid
Assessment: ContradictedSaid
“Software like LaMDA simply doesn’t; it doesn’t even try to connect to the world at large, it just tries to be the best version of autocomplete it can be, by predicting what words best fit a given context.”
What happened: The paper describes retrieval, calculator and translation tools, and training that routes queries to those tools during inference. This directly contradicts the assertion that LaMDA does not even attempt external grounding. (Google Research authors / arXiv, )
Assessment: ContradictedSaid
“But 8 years later I doubt most people (even in AI) have ever even heard of his program, outside of my mentioning it here. It made zero lasting contribution to AI.”
What happened: The author's statement says Emelya was based on a reworked natural-language analysis system created for Eugene Goostman. Her account dates development to 2016, release to May 2017 and further user research to 2018, documenting a technical legacy beyond the 2014 competition. (Habr / Ontico, account by project participant Anna Degteva, )
Assessment: SupportedSaid
“The basic finding was: loads of things scale, but not at all ( as foreseen both by Kurenkov in his essay last year and in my much lampooned but basically accurate essay, Deep learning is hitting a wall).”
What happened: The original report directly states the coexistence Marcus describes. Sections 3.4.2 and 6.1 also explain that metric choice can conceal gradual progress and that some limitations may change with larger models or improved data. (Srivastava et al.; arXiv, )
Assessment: UnresolvedSaid
Assessment: UnresolvedSaid
Assessment: SupportedSaid
“The massive paper looks at scaling on many measures, and sees progress on some– but not others. The stuff that isn’t scaling is the “wall”.”
What happened: The report directly confirms improving and non-improving task scores across evaluated scales. Its discussion of smoother metrics limits what those flat scores imply about underlying capabilities. (Srivastava et al.; arXiv, )
Assessment: UnresolvedSaid
Assessment: SupportedSaid
“so many of the unfortunate consequences of a webscraping/big data approach to AI are disproportionately borne by women and minorities.”
What happened: The audit reports removal rates of 42% for African American English and 32% for Hispanic-aligned English, against 6.2% for White-aligned English. This documents unequal exclusion from a web-scraped training corpus, using dialect as a proxy rather than identifying authors. (Dodge et al.; EMNLP 2021, arXiv version, )
Assessment: SupportedSaid
“I would be very surprised if that fundamental lack of reliability did not also hold true for GPT-4. Maybe it will take three paragraphs in a prompt rather than one to break it, but a big Transformer alone does not a a theory of the world create.”
What happened: OpenAI's launch-era report explicitly acknowledges both predicted failure classes and advises safeguards where reliability matters. It also reports substantial factuality improvement, which is compatible with Marcus's allowance for fewer errors. (OpenAI / arXiv, )
Assessment: UnresolvedSaid
Assessment: UnresolvedSaid
Assessment: UnresolvedSaid
Assessment: UnresolvedSaid
Assessment: UnresolvedSaid
Assessment: UnresolvedSaid
Assessment: UnresolvedSaid
Assessment: UnresolvedSaid
Assessment: SupportedSaid
“You said, for instance in 2015, that (truly) self-driving cars were two years away; you’ve pretty much said the same thing every year since. It still hasn’t happened.”
What happened: The original interview establishes the attributed prediction. Musk separately allowed additional time for regulatory approval, so the prediction concerned technical capability first. (Fortune, )
Assessment: UnresolvedSaid
Assessment: UnresolvedSaid
Assessment: UnresolvedSaid
Assessment: UnresolvedSaid
Assessment: UnresolvedSaid
Assessment: SupportedSaid
“Even within perception 3D perception remains a challenge, and scene understanding is by no means solved.”
What happened: Testing fifteen common corruptions exposed substantial robustness weaknesses in leading 3D recognition models. The authors also report mitigation methods, so the result establishes remaining challenges rather than an absence of progress. (arXiv; Jiachen Sun and colleagues, )
Assessment: SupportedSaid
“But for the kinds of AI techniques that are currently popular, and which rely more on big data than on reasoning, it’s a very serious issue.”
What happened: The contemporary synthesis documents poor learning of rare classes and substantial limitations in conventional deep-learning methods. It also reviews mitigation methods, supporting a serious practical problem without implying that every method fails. (International Journal of Computer Vision; author manuscript rendered by ar5iv, )
Assessment: UnresolvedSaid
Assessment: MixedSaid
“Debugging with deep learning is wicked hard, because nobody (a) really understands how it works, and (b) knows how to fix problems other to collect more data and add more layers and so forth.”
What held: The survey supports the narrower causal point that opacity makes locating and repairing problems harder. Its review of explanation-based debugging also shows that researchers were already developing remedies. (arXiv; Piyawat Lertvittayakumjorn and Francesca Toni, )
What did not: The authors used causal interventions to investigate factual recall and evaluated targeted factual edits on CounterFact. This supplies a concrete pre-post example of useful mechanistic understanding and repair without adding data or layers. (arXiv; Kevin Meng, David Bau, Alex Andonian and Yonatan Belinkov, )
Assessment: UnresolvedSaid
Assessment: SupportedSaid
“And it’s why I absolutely guarantee you won’t be shipping level 5 self-driving cars this year or next, no matter what you might have told Chris Anderson at TED.”
What happened: The annual filing describes Tesla's offerings as driver assistance and says the driver remains responsible for controlling the vehicle. That capability does not meet Level 5. (Tesla, Inc., mirrored by AnnualReports.com, )
Assessment: UnresolvedSaid
Assessment: SupportedSaid
“But you and I both know that they continue to be a major problem that as yet has no known robust solution.”
What happened: The survey identifies unresolved obstacles involving graph structure, differing tasks, and different kinds of distribution shift. It also describes progress, supporting the distinction between useful remedies and a general solution. (arXiv; Haoyang Li and colleagues, )
Assessment: SupportedSaid
“That’s where the deep understanding of language should live—but, despite the PR, it isn’t there.”
What happened: The original authors identify a category-level alignment weakness in the figure containing the horse-riding-an-astronaut prompt. This supports the contextual reading that reliable compositional instruction understanding was missing. (Google Research, Brain Team; copy hosted by 3DVAR, )
Assessment: UnresolvedSaid
Assessment: UnresolvedSaid
Assessment: UnresolvedSaid
Assessment: UnresolvedSaid
Assessment: UnresolvedSaid
Assessment: ContradictedSaid
“They do not gravitate towards human language, and they never develop a cohesive undestanding of the world, no matter how much experience they have.”
What happened: Kallini and colleagues compared GPT-2 learning across English and systematically altered languages, including different training stages. The observed preference conflicts with the language-learning component under the stated family-level interpretation. (arXiv, )
Assessment: UnresolvedSaid
Assessment: UnresolvedSaid
Assessment: UnresolvedSaid
Assessment: UnresolvedSaid
Assessment: SupportedSaid
“Success on a few problems with a particular strategy in no way guarantees that we can solve all problems in a similar way. It is sheer inductive folly to imagine otherwise, particular when the failure modes (unreliability, bizarre errors, failures in compositionality and discomprehension) have not changed since Fodor and Pinker pointed them out (in separate articles) in 1988.”
What happened: Tests involving SCAN and RobustFill found continuing difficulty combining learned components in new ways, even with modified Transformers. This supports persistence of compositional generalization problems while explicitly acknowledging progress. (arXiv; Kensen Shi and colleagues, )
Assessment: UnresolvedSaid
Assessment: UnresolvedSaid
Assessment: UnresolvedSaid
Assessment: SupportedSaid
“DeepMind’s newest star, just unveiled, Gato, is capable of cross-modal feats never seen before in AI, but still, when you look in the fine print, remains stuck in the same land of unreliability, moments of brilliance coupled with absolute discomprehension:”
What happened: The launch paper reports one model spanning dialogue, captions, games and real robot control, including human-level or better performance on 23 Atari games, while acknowledging frequent poor dialogue responses. Its prior-work comparison supports novelty in that combined coverage, rather than novelty of multimodality itself. (arXiv; DeepMind, )
Assessment: UnresolvedSaid
Assessment: SupportedSaid
“But it’s also pretty prone to surprising stumbles, like this uncomprehending reply to “a red cube on top of a blue cube”.”
What happened: The authors discuss the exact red-cube/blue-cube prompt in Figure 14 and report recurring object–attribute mix-ups in reconstructions in Figure 15. This supports a recurring compositional weakness in the original model, without establishing a population-wide error rate. (arXiv; OpenAI, )
Assessment: SupportedSaid
“The history missed is the the fact that many neural net enthusiasts, have argued against symbols for decades, and the reality that is missed is the above-documented fact that symbolic descriptions like “red cube on blue cube” still elude the state-of-the-art in 2022.)”
What happened: This dated scholarly account documents an explicit anti-symbolic strand well before the post. It is historical synthesis rather than a primary statement by an advocate. (Stanford Encyclopedia of Philosophy, )