Claims that can’t be checked by Gary Marcus, page 25 of 25

Claims that can’t be checked by Gary Marcus (2,489)

Opinions, rhetoric, hedged or vague claims, and things no public evidence could show. Listed, not scored.

Claims 2,401 to 2,489 of 2,489, newest first. Page 25 of 25.

1. Assessment: Can’t be checkedSaid 2022-08-18

   “Without a much greater degree of interpretability and verifiability, its hard to see how we can get to that level of safety.”

2. Assessment: Can’t be checkedSaid 2022-08-18

   “Misunderstandings will arise; without really firm checks on these systems, some of those misunderstandings could lead to mayhem.”

3. Assessment: Can’t be checkedSaid 2022-08-18

   “It’s not just that large language models can counsel suicide, or sign off on genocide, or that they can be toxic and that they are incredibly (over)sensitive to the details of their training set - it’s that if you put them inside a robot, and they misunderstand you, or fail to fully appreciate the implications of your request, they could cause major damage.”

4. Assessment: Can’t be checkedSaid 2022-08-18

   “And the more actions a robot might consider, the worse performance is likely to be.”

5. Assessment: Can’t be checkedSaid 2022-08-18

   “Putting aside robots for the moment, we already know that so-called large language models are like bulls in a china shop: awesome, powerful, and reckless.”

6. Assessment: Can’t be checkedSaid 2022-08-18

   “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.”

7. Assessment: Can’t be checkedSaid 2022-08-18

   “In a general-purpose, humanoid domestic robot, 75% is nowhere near good enough.”

8. Assessment: Can’t be checkedSaid 2022-08-18

   “The PaLM-SayCam experiment makes it pretty clear that the list of 21 problems is due for an update.”

9. Assessment: Can’t be checkedSaid 2022-08-18

   “As I have argued throughout the essays in this Substack (eg. Three ideas from linguistics that everyone in AI should know and The New Science of Alt Intelligence) , that’s because large language models are superficial statistical mimics, rather than systems that traffic in rich cognitive models of the world around them.”

10. Assessment: Can’t be checkedSaid 2022-08-18

    “The reality is that the field has no compelling idea at the current time how to solve the many “alignment” problems that bedevil large language models.”

11. Assessment: Can’t be checkedSaid 2022-08-18

    “But that precisely what Google’s new system is, a stitching-together of superficial and incorrigible language understanders with powerful and potentially dangerous humanoid robots.”

12. Assessment: Can’t be checkedSaid 2022-08-18

    “I am not surprised that this can (kinda sorta) be done, but I am not sure it should be done.”

13. Assessment: Can’t be checkedSaid 2022-07-20

    “And it is worth noting that even if they did work, there would be, of course, no guarantee that they would work in a fashion that was similar to humans; AlphaGo plays a mean game of Go but that doesn’t mean it gives insight into how human players learn Go.)”

14. Assessment: Can’t be checkedSaid 2022-07-20

    “If we want to make progress, we need to start by recognizing the enormity of the problem.”

15. Assessment: Can’t be checkedSaid 2022-06-22

    “Ellipsis, comparative and pronouns are routine, everyday components of language, readily acquired by children. DALLE-mini does not seem to grasp them.”

16. Assessment: Can’t be checkedSaid 2022-06-22

    “But we see no reason to think that current large language models encode stable representations of the world at all. All they are doing, literally, is predicting next words in sentences.”

17. Assessment: Can’t be checkedSaid 2022-06-22

    “Moving beyond prediction into reference, particularly in a system that is inherently uninterpretable, is an immense leap that (a) cannot be taken for granted and (b) should be a focus of research.”

18. Assessment: Can’t be checkedSaid 2022-06-22

    “We would suggest that all the attention that has been placed on predicting sequences of words has come at the expense of developing a theory of how such processes might culminate in cognitive models of the world.”

19. Assessment: Can’t be checkedSaid 2022-06-22

    “Rather, it means identify the second of the balls that are blue, regardless of which position; systems like GPT-3 likely have trouble with this sort of thing, as Ernie Davis just reported to us in an email.”

20. Assessment: Can’t be checkedSaid 2022-06-22

    “Considerable efforts are being made towards making “cross-modal” models like Gato that take in visual input as well strings of words; we think this is well worth investigating but unlikely in itself to suffice.”

21. Assessment: Can’t be checkedSaid 2022-06-22

    “For that, we may yet need another revolution.”

22. Assessment: Can’t be checkedSaid 2022-06-22

    “It may produce both a blue cube and a red cube, but as likely as not, the red cube winds up on top of the blue cube.”

23. Assessment: Can’t be checkedSaid 2022-06-22

    “Current AI systems don’t really show up for any of that.”

24. Assessment: Can’t be checkedSaid 2022-06-22

    “Yet even more wild is the widespread belief that language models lacking the major foundational components of human language (e.g., bounded compositionality mixed with unbounded generativity) will have any hope of meaningfully interfacing with audio/visual models in a way that accurately reflects human conceptual structure.”

25. Assessment: Can’t be checkedSaid 2022-06-22

    “A good language understanding (or language production) system almost certainly needs to have some large-scale learning as a vital component.”

26. Assessment: Can’t be checkedSaid 2022-06-22

    “But if all a system is doing is predicting the next word in what someone says, its understanding is inevitably superficial.”

27. Assessment: Can’t be checkedSaid 2022-06-22

    “In our view, large language models are basically sidestepping this entire set of issues, and to their peril.”

28. Assessment: Can’t be checkedSaid 2022-06-22

    “Ultimately, the right approach to large-scale learning may not rest on machinery that is finely geared to predicting sequences of words, but rather on developing a new approach to machine learning that is fundamentally geared towards learning how hierarchically structured sets of words map onto meanings that are used in the service of updating cognitive models.”

29. Assessment: Can’t be checkedSaid 2022-06-17

    “A couple weeks ago I got so fed up with so many smart people spending so much time experimenting with GPT-3 and DALL-E and so forth (when I, personally, think that there are deeper problems to be solved), that I wrote a poem.”

30. Assessment: Can’t be checkedSaid 2022-06-15

    “For example, we lack data on whether drivers use such systems as directed, and need to enforce vigorous mechanisms to make sure drivers don’t tune out.”

31. Assessment: Can’t be checkedSaid 2022-06-15

    “Clearly, the number of incidents would go up, as the number of vehicles went up, and would also go up significantly as the systems were more regularly challenged with more difficult driving conditions (bad weather, more erratic pedestrians, etc), and likely jump radically if we took humans out of the loop altogether.”

32. Assessment: Can’t be checkedSaid 2022-06-15

    “Is that good, or bad? I don’t know it, but it’s enough to be concerning. Why? Because .75 incidents a day comes from a very limited trial of computer-assisted driving relative to the ambition of full-self-driving.”

33. Assessment: Can’t be checkedSaid 2022-06-15

    “My informal read (still lacking a tremendous amount of information that I as a scientist would really like to see) is that neither Tesla nor Waymo is anywhere close to something we could use every day, all the time, wherever one wanted to go.”

34. Assessment: Can’t be checkedSaid 2022-06-15

    “The best argument that Tesla might not survive if its massive bet on self-driving fails comes from Elon Musk himself.”

35. Assessment: Can’t be checkedSaid 2022-06-15

    “It’s fantastic that NHTSA is now pushing manufacturers to release incident data, and but we need soooo much more data (eg about miles driven) if we are to look at the money question—how safe is all this, really—in a truly scientific way.”

36. Assessment: Can’t be checkedSaid 2022-06-15

    “But our experiments in building them need to be strictly regulated, especially when nonconsenting human beings become unwilling participants. And our requirements on data transparency need to be vastly stricter.”

37. Assessment: Can’t be checkedSaid 2022-06-15

    “Does that mean that society should give up on building driverless cars? Absolutely not.”

38. Assessment: Can’t be checkedSaid 2022-06-15

    “This morning’s fresh data dump from NHTSA should be a bit of a wakeup call to those who have gone long on Tesla, if they are thinking that a solution to Full Self Driving is in sight. The data aren’t devastating, but they aren’t so promising either, and they raise of lot of questions.”

39. Assessment: Can’t be checkedSaid 2022-06-12

    “In truth, literally everything that the system says is bullshit.”

40. Assessment: Can’t be checkedSaid 2022-06-12

    “Fooling people into thinking a program is intelligent is just not the same as building programs that actually are intelligent.”

41. Assessment: Can’t be checkedSaid 2022-06-12

    “Neither LaMDA nor any of its cousins (GPT-3) are remotely intelligent.”

42. Assessment: Can’t be checkedSaid 2022-06-12

    “All they do is match patterns, draw from massive statistical databases of human language. The patterns might be cool, but language these systems utter doesn’t actually mean anything at all.”

43. Assessment: Can’t be checkedSaid 2022-06-12

    “But there is no absolutely no reason whatever for us to waste time wondering whether anything anyone in 2022 knows how to build is sentient. It is not.”

44. Assessment: Can’t be checkedSaid 2022-06-12

    “None of the software in either of those systems has survived in modern efforts at “artificial general intelligence”, and I am not sure that LaMDA and its cousins will play any important role in the future of AI, either.”

45. Assessment: Can’t be checkedSaid 2022-06-12

    “What these systems do, no more and no less, is to put together sequences of words, but without any coherent understanding of the world behind them, like foreign language Scrabble players who use English words as point-scoring tools, without any clue about what that mean.”

46. Assessment: Can’t be checkedSaid 2022-06-12

    “To be sentient is to be aware of yourself in the world; LaMDA simply isn’t.”

47. Assessment: Can’t be checkedSaid 2022-06-11

    “In my view, no way; I think the actual odds are less than 20%.”

48. Assessment: Can’t be checkedSaid 2022-06-11

    “While there’s no denying that the two of us have often clashed, on this point—the point about scale alone not likely to be enough, and about the need for some new discovery (i.e., paradigm shifts)—we are actually in complete agreement.”

49. Assessment: Can’t be checkedSaid 2022-06-11

    “GPT-2 is both a triumph for empiricism, and, in light of the massive resources of data and computation that have been poured into them, a clear sign that it is time to consider investing in different approaches.”

50. Assessment: Can’t be checkedSaid 2022-06-11

    “I say that we have given the scaling hypothesis a really good look (with a larger budget than all but a handful of projects in history), and as such the failures of large scale systems are evidence (not proof) that we ought to seriously consider alternatives, e.g., here:”

51. Assessment: Can’t be checkedSaid 2022-06-11

    “That said, I, of course, could be right about foreseeing the need for a paradigm shift, but wrong about what that paradigm shift turns out to be.”

52. Assessment: Can’t be checkedSaid 2022-06-11

    “That’s why we need new approaches, exactly as I have been arguing.”

53. Assessment: Can’t be checkedSaid 2022-06-11

    “Personally I’d put the probability that we need to shift at 90%, well above the 60% that Alexander suggests, and I would put the probability that we will need to embrace symbol-manipulation as part of the mix at 80%, more than double Alexander’s 34%.”

54. Assessment: Can’t be checkedSaid 2022-06-11

    “Personally I’d put the probability that we need to shift at 90%, well above the 60% that Alexander suggests, and I would put the probability that we will need to embrace symbol-manipulation as part of the mix at 80%, more than double Alexander’s 34%.”

55. Assessment: Can’t be checkedSaid 2022-06-07

    “Scott’s more marked success on six questions may have been a fluke.”

56. Assessment: Can’t be checkedSaid 2022-06-07

    “The core problem that follows from that has always been that these systems aren’t stable or trustworthy; a crumbling house follows from a rotten foundation.”

57. Assessment: Can’t be checkedSaid 2022-06-07

    “I think all that’s scandalous.”

58. Assessment: Can’t be checkedSaid 2022-06-07

    “But I see no reason whatsoever to think that the underlying problem — a lack of cognitive models of the world —have been remedied.”

59. Assessment: Can’t be checkedSaid 2022-06-07

    “Either way, though, there is still no real physical, social, or psychological understanding, just richer parroting of larger corpora (and some reinforcement learning to keep the system more on check).”

60. Assessment: Can’t be checkedSaid 2022-06-07

    “Maybe we will find fewer absolute howlers per attempt than we did with GPT-3, but it still won’t be so proficient as to have developed a genuine causal model of the world.”

61. Assessment: Can’t be checkedSaid 2022-06-07

    “And yes, there has been improvement eg from GPT-2 to GPT-3—but zero qualitative change.”

62. Assessment: Can’t be checkedSaid 2022-05-31

    “Four years later, many people continue to hope that deep learning will be a panacea; that still seems pretty unrealistic to me; I still think we need a whole lot more techniques.”

63. Assessment: Can’t be checkedSaid 2022-05-31

    “Seven years is a long time, but the field is going to need to invest in other ideas if we are going to get to AGI before the end of the decade. Or else outliers alone might be enough to keep us from getting there.”

64. Assessment: Can’t be checkedSaid 2022-05-31

    “Unexpected circumstances have been and continue to be the bane of contemporary AI techniques, and probably will be until there is a real revolution.”

65. Assessment: Can’t be checkedSaid 2022-05-31

    “So far as I know your cars are still mostly just doing a kind of glorified hash-table-like lookup (viz deep learning), and hence still having a lot of problems with unexpected situations.”

66. Assessment: Can’t be checkedSaid 2022-05-31

    “Being realistic, 7 years might well not be enough to (a) invent those tools (if they are don’t already exist) and (b) take them from the lab into production.”

67. Assessment: Can’t be checkedSaid 2022-05-31

    “Things like programming verification might eventually help, but again, in deep learning we don’t yet have the tools for writing verifiable code. If you are to win the bet, we are probably going to have to solve that problem too, and soon.”

68. Assessment: Can’t be checkedSaid 2022-05-28

    “Against a long history of overclaiming in AI, such strong statements demand an awful lot of evidence. Because Google won’t allow skeptics to try the system, we can only make guesses. But from the few clues that have filtered out, there is enormous reason for skepticism.”

69. Assessment: Can’t be checkedSaid 2022-05-28

    “it casts doubt on “an astronaut riding a horse”, and suggests success there is not because the network knows how to extract meanings from individual words”

70. Assessment: Can’t be checkedSaid 2022-05-28

    “I am pretty sure this kind of failure would be easy to replicate.”

71. Assessment: Can’t be checkedSaid 2022-05-28

    “The answer to the second question—does Imagen have a deep understanding of language—is almost certainly “no”: if a system can draw a given picture with incantation X, but not with perfectly sensible alternative incantation Y, it’s pretty hard to say that the system understands Y.”

72. Assessment: Can’t be checkedSaid 2022-05-28

    “DALL-E 2 is, I dare say, not as smart as it seems.”

73. Assessment: Can’t be checkedSaid 2022-05-28

    “Judging by all the premature declarations I have seen from the machine learning community of late, I am not sure AI has yet fully learned it.”

74. Assessment: Can’t be checkedSaid 2022-05-21

    “A new breed of AI systems, with a firmer innate basis, closer to the Kantian model, but out of step with current Zeitgeist, might eventually tell us more.”

75. Assessment: Can’t be checkedSaid 2022-05-21

    “Human minds try to relate their language to what they know about the world; large language models don’t. Which means the latter just aren’t telling us much about the former.”

76. Assessment: Can’t be checkedSaid 2022-05-21

    “Large language models can, parasitically, glom onto human data, but they could glom onto any other data set just as well (e.g. programming languages); as such, they tell us far too little about how humans have come to be what they are.”

77. Assessment: Can’t be checkedSaid 2022-05-21

    “Systems like GPT-3 don’t tell us why human languages have the special character that they do. As such, there is little explanatory value.”

78. Assessment: Can’t be checkedSaid 2022-05-21

    “Humans know these things; if AI’s are to understand language, they should refract language through this sort basic understanding of common sense, rather than needlessly starting as blank slates and acquiring statistics without concepts.”

79. Assessment: Can’t be checkedSaid 2022-05-21

    “They are an embarrassment to Locke’s conjecture, and a strong argument for exploring their very antithesis, which would be Kantian systems that start with representations of space and time and causality, and at least some kernels of universal grammar.”

80. Assessment: Can’t be checkedSaid 2022-05-14

    “I don’t think any pure deep learning/big data system can match it.”

81. Assessment: Can’t be checkedSaid 2022-05-14

    “We are in early days in AI.”

82. Assessment: Can’t be checkedSaid 2022-05-14

    “But anyone who is candid will recognize that these kinds of errors reveal that something is, for now, deeply amiss.”

83. Assessment: Can’t be checkedSaid 2022-05-14

    “Let us all encourage a field that is open-minded enough to work in multiple directions, without prematurely dismissing ideas that happen to be not yet fully developed.”

84. Assessment: Can’t be checkedSaid 2022-05-14

    “It may just be that the best path to artificial (general) intelligence isn’t through Alt Intelligence, after all.”

85. Assessment: Can’t be checkedSaid 2022-05-14

    “A large fraction of current AI researchers, or at least those currently in power, don’t (so far as I can tell) give a damn about any of this.”

86. Assessment: Can’t be checkedSaid 2022-05-14

    “AI should certainly not be a slavish replica of human intelligence (which after all is flawed in its own ways, saddled with lousy memory and cognitive bias). But it should look to human (and animal cognition) for clues.”

87. Assessment: Can’t be checkedSaid 2022-05-14

    “It may well be better than anything else we currently have, but the fact that it still doesn’t really work, even after all the immense investments that have been made in it, should give us pause.”

88. Assessment: Can’t be checkedSaid 2022-05-14

    “It is sheer hubris to believe otherwise.”

89. Assessment: Can’t be checkedSaid 2022-05-14

    “To some (such as myself), the abiding problem of discomprehension might—despite the immense progress—signal the need for a fundamental rethink, such as the one that Davis and I offered in our book Rebooting AI.”