| Resolved true |
Prediction |
1 Jan 2025 |
Lack of reliability will continue to afflict generative AI in 2025.
Prediction · 1 Jan 2025 |
AI capabilities |
| Resolved true |
Prediction |
1 Jan 2025 |
Artificial general intelligence will not be achieved in 2025.
Prediction · 1 Jan 2025 |
AI timelines and forecasting |
| Resolved true |
Prediction |
1 Jan 2025 |
Hallucinations will continue to afflict generative AI in 2025.
Prediction · 1 Jan 2025 |
AI capabilities |
| Resolved true |
Prediction |
1 Jan 2025 |
Less than 10% of the workforce will be replaced by AI in 2025, probably less than 5%.
Prediction · 1 Jan 2025 |
AI, the economy and work |
| Unresolvable |
Prediction |
1 Jan 2025 |
Copyright lawsuits over generative AI will continue throughout 2025.
Prediction · 1 Jan 2025 |
AI policy and governance |
| Unresolvable |
Prediction |
1 Jan 2025 |
Valuations of major AI companies may start to fall in 2025.
Prediction · 1 Jan 2025 |
Economics and markets |
| Not researched |
Factual |
1 Jan 2025 |
Gary Marcus's predictions for 2024 on his Substack largely held up, with some caveats.
Factual · 1 Jan 2025 |
AI timelines and forecasting |
| Resolved true |
Prediction |
1 Jan 2025 |
The US will continue to have very little regulation protecting consumers from generative AI risks in 2025.
Prediction · 1 Jan 2025 |
AI policy and governance |
| Not researched |
Factual |
1 Jan 2025 |
All seven of Marcus's core 2024 predictions proved essentially correct, with a possible caveat about OpenAI's o3, announced in December 2024 but not released.
Factual · 1 Jan 2025 |
AI timelines and forecasting |
| Mostly true |
Prediction |
1 Jan 2025 |
A large-scale cyberattack in which generative AI plays an important causal role may well occur in 2025.
Prediction · 1 Jan 2025 |
Other global risks |
| Resolved true |
Prediction |
1 Jan 2025 |
AI agents will be heavily hyped throughout 2025 but remain far from reliable, except possibly in very narrow use cases.
Prediction · 1 Jan 2025 |
AI capabilities |
| Resolved true |
Prediction |
1 Jan 2025 |
AI power consumption will rise in 2025 and remain a major problem, while few generative AI companies will be transparent about their usage.
Prediction · 1 Jan 2025 |
AI compute, hardware and data |
| Mostly true |
Prediction |
1 Jan 2025 |
Few if any radiologists will be replaced by AI in 2025.
Prediction · 1 Jan 2025 |
AI, the economy and work |
| Resolved true |
Prediction |
1 Jan 2025 |
No single AI system will solve more than 4 of the AI 2027 Marcus-Brundage tasks by the end of 2025.
Prediction · 1 Jan 2025 |
AI capabilities |
| Resolved true |
Prediction |
1 Jan 2025 |
In 2025 humanoid robotics will see a lot of hype, but nobody will release a robot remotely as capable as Rosie the Robot.
Prediction · 1 Jan 2025 |
AI capabilities |
| Mostly true |
Prediction |
1 Jan 2025 |
Neurosymbolic AI will become much more prominent in 2025.
Prediction · 1 Jan 2025 |
AI capabilities |
| Not researched |
Factual |
1 Jan 2025 |
The weak spots Marcus raised in 2022 and in his 2001 book The Algebraic Mind (compositionality, factuality, reasoning) persist in AI systems.
Factual · 1 Jan 2025 |
AI capabilities |
| Unresolvable |
Prediction |
1 Jan 2025 |
In 2025 truly driverless cars will remain limited to a modest number of cities, mainly in the West and in good weather.
Prediction · 1 Jan 2025 |
Technology |
| Mostly true |
Prediction |
1 Jan 2025 |
OpenAI's Sora will continue to have trouble with physics in 2025.
Prediction · 1 Jan 2025 |
AI capabilities |
| Unresolvable |
Prediction |
1 Jan 2025 |
AI will presumably solve compositionality, factuality, reasoning and similar problems someday, though current techniques have thus far not been adequate.
Prediction · 1 Jan 2025 |
AI capabilities |
| Not checkable |
Not checkable |
2025 |
But in general, like we don’t have the same amount of sort of person hours on these alternatives.
Not checkable · 2025 |
|
| Not checkable |
Not checkable |
2025 |
For example, in biology, you still have to do empirical work.
Not checkable · 2025 |
|
| Not researched |
Factual |
2025 |
No neuro-symbolic approach has reached the needed level of productionization except in very narrow domains.
Factual · 2025 |
AI capabilities |
| Not checkable |
Not checkable |
2025 |
If you specify your constraints in prolog, but you specify the wrong constraints, you’re still in trouble.
Not checkable · 2025 |
|
| Not checkable |
Not checkable |
2025 |
Yeah, and, I mean, it raises the question about the possibility of moral progress.
Not checkable · 2025 |
|
| Not checkable |
Not checkable |
2025 |
Like, I don’t think there’s enough societal support for looking at alternatives.
Not checkable · 2025 |
|
| Not researched |
Factual |
2025 |
LLMs lack deep comprehension of what they discuss, and their reasoning is limited to things similar to what they have seen before.
Factual · 2025 |
AI capabilities |
| Not researched |
Factual |
2025 |
Some birds can perform variable binding and similar abstract operations.
Factual · 2025 |
Science and research methods |
| Not checkable |
Not checkable |
2025 |
That’s a choice. And I think it’s a wrong choice.
Not checkable · 2025 |
|
| Not checkable |
Not checkable |
2025 |
But I’ll also say I’ve seen a lot of demos that you know took longer to be realized.
Not checkable · 2025 |
|
| Not researched |
Factual |
2025 |
LLMs do not, even in principle, really represent world models of persistent entities and their properties.
Factual · 2025 |
AI capabilities |
| Not researched |
Factual |
2025 |
Bees calculating the solar azimuth function can generalize beyond data they have seen to lighting conditions they have never seen.
Factual · 2025 |
Science and research methods |
| Not checkable |
Not checkable |
2025 |
And someday, we will both be proven right.
Not checkable · 2025 |
|
| Not researched |
Factual |
2025 |
LLMs learn contextual relations among words but do not learn the underlying concepts.
Factual · 2025 |
AI capabilities |
| Not researched |
Factual |
2025 |
Neural networks only have a workaround to approximate variable binding rather than implementing it directly.
Factual · 2025 |
AI capabilities |
| Not checkable |
Not checkable |
2025 |
you wouldn’t build a GPS navigation system that way. You shouldn’t build a domestic robot that way.
Not checkable · 2025 |
|
| Not checkable |
Not checkable |
2025 |
Intelligence is multi dimensional. They’re smarter in some dimensions.
Not checkable · 2025 |
|
| Not checkable |
Not checkable |
2025 |
It has to start with neuro symbolic AI, which I think we agree about right?
Not checkable · 2025 |
|
| Not checkable |
Not checkable |
2025 |
I think that we could hard code things and make them fixed.
Not checkable · 2025 |
|
| Mostly true |
Factual |
2025 |
People put instructions in LLM system prompts but cannot get the systems to follow them.
Factual · 2025 |
AI safety and alignment |
| Own records only |
Factual |
2025 |
For 30 years people have told Gary Marcus that abstract reasoning and variable binding will emerge from neural nets.
Factual · 2025 |
AI capabilities |
| Not checkable |
Not checkable |
2025 |
I absolutely agree.
Not checkable · 2025 |
|
| Not checkable |
Not checkable |
2025 |
I think that people try to duck the second harder problem, and they say “well, we can never do it because humans don’t agree on values.”
Not checkable · 2025 |
|
| Not checkable |
Not checkable |
2025 |
I worry a lot about the security and maintainability of the code that’s being written.
Not checkable · 2025 |
|
| Not checkable |
Not checkable |
2025 |
These are two separate issues.
Not checkable · 2025 |
|
| Not checkable |
Not checkable |
2025 |
I think if you don’t have that as some part of your system, you’re in real trouble.
Not checkable · 2025 |
|
| Not researched |
Factual |
2025 |
The abstract idea of intelligent robots is not new.
Factual · 2025 |
Technology |
| Not checkable |
Not checkable |
2025 |
That’s the one we’re in now.
Not checkable · 2025 |
|
| Not checkable |
Not checkable |
2025 |
Lots of basic stuff that the LLM world has just not confronted.
Not checkable · 2025 |
|
| Not researched |
Factual |
2025 |
The further LLMs are taken from standard games, the more and the more quickly they break down.
Factual · 2025 |
AI capabilities |
| Not checkable |
Not checkable |
2025 |
fundamentally, they’re mimics, and they’re better than just the simple memorizing system.
Not checkable · 2025 |
|
| Unresolvable |
Factual |
2025 |
The shortcuts that LLMs use work about 80% of the time but not 100%.
Factual · 2025 |
AI capabilities |
| Unresolvable |
Factual |
2025 |
Marcus said he thinks many, or possibly all, animals can do some variable binding, and he cited Randy Gallistel's work on ant navigation as an example.
Factual · 2025 |
Science and research methods |
| Not checkable |
Not checkable |
2025 |
The system’s level of comprehension of the words that are put in those prompts is poor in some sense, and they’re not compliant with the…
Not checkable · 2025 |
|
| Not researched |
Factual |
2025 |
LLM systems still explain how to build biological weapons despite system prompts telling them not to.
Factual · 2025 |
AI safety and alignment |
| Not checkable |
Not checkable |
2025 |
But there’s also the case that complex software just usually doesn’t work.
Not checkable · 2025 |
|
| Not checkable |
Not checkable |
2025 |
We’re not that good as a society engineering complex software.
Not checkable · 2025 |
|
| Not checkable |
Not checkable |
2025 |
Investors are herd animals.
Not checkable · 2025 |
|
| Not checkable |
Not checkable |
2025 |
But I think that there’s a lot of sort of “human compatible,” to borrow Stuart Russell’s term, a lot of “human compatible” values that we…
Not checkable · 2025 |
|
| Mostly true |
Factual |
2025 |
LLMs, which are not AGI, are already eliminating some human work.
Factual · 2025 |
AI, the economy and work |
| Unresolvable |
Prediction |
2025 |
Marcus said the chance that someone, possibly the person he was speaking with or someone else, has a different architecture that works is maybe a little higher.
Prediction · 2025 |
AI timelines and forecasting |
| Unresolvable |
Prediction |
2025 |
There are some questions, such as the definition of life in abortion decisions, on which humans will never agree.
Prediction · 2025 |
Philosophy, ethics and society |
| Not researched |
Factual |
2025 |
LLMs do not reason well outside their training distribution.
Factual · 2025 |
AI capabilities |
| Own records only |
Factual |
2025 |
Gary Marcus has never personally experienced 99% correct output from AI in any of the things he has tried.
Factual · 2025 |
AI capabilities |
| Not checkable |
Not checkable |
2025 |
I’m not sure how much other animals can do that.
Not checkable · 2025 |
|
| Not checkable |
Not checkable |
2025 |
I think it should be a baseline value. You don’t kill people, at least not without a good reason.
Not checkable · 2025 |
|
| Unresolvable |
Prediction |
2025 |
When "it" arrives, people will look back at the early 2020s and ask why so much effort went in one direction when the answer lay elsewhere.
Prediction · 2025 |
AI timelines and forecasting |
| Not checkable |
Not checkable |
2025 |
That is a failure to adopt an explicit set of rules, to be constrained by them.
Not checkable · 2025 |
|
| Not researched |
Factual |
2025 |
An LLM does not really have the concept of three in a row in tic tac toe despite abundant literature on it.
Factual · 2025 |
AI capabilities |
| Unresolvable |
Prediction |
2025 |
Marcus said he thinks the chance that "it" (the development under discussion, which the quoted passage does not name) will come out of large language models is zero.
Prediction · 2025 |
AI timelines and forecasting |
| Not checkable |
Not checkable |
2025 |
These things have to be first class citizens in, I think, any AGI system.
Not checkable · 2025 |
|
| Not checkable |
Not checkable |
2025 |
That’s part of why I hedge my bets.
Not checkable · 2025 |
|
| Not checkable |
Not checkable |
2025 |
Right now. People have their own ideas about LLMs.
Not checkable · 2025 |
|
| Not checkable |
Not checkable |
2025 |
We really want systems that we can count on in the way that we can count on calculators.
Not checkable · 2025 |
|
| Not checkable |
Not checkable |
2025 |
That is a potential situation.
Not checkable · 2025 |
|
| Not researched |
Citation |
2025 |
Gary Marcus said he thinks the paper under discussion basically says the same thing he and his interlocutor are saying: that these systems can run algorithms but break down on larger problems.
Citation · 2025 |
AI capabilities |
| Not checkable |
Not checkable |
2025 |
Maybe it’s 2029. Maybe it’s 2039. Maybe it’s 2049.
Not checkable · 2025 |
|
| Not checkable |
Not checkable |
2025 |
I would say a minimal requirement for control, not a sufficient one, but a necessary one, is that we should be able to state, let’s say…
Not checkable · 2025 |
|
| Not checkable |
Not checkable |
2025 |
I don’t think that’s the situation we’re in now.
Not checkable · 2025 |
|
| Not checkable |
Not checkable |
2025 |
I think we could do more to prepare than we have.
Not checkable · 2025 |
|
| Not researched |
Factual |
2025 |
LLMs are opaque, which makes them difficult to debug.
Factual · 2025 |
AI capabilities |
| Unresolvable |
Prediction |
2025 |
Even if AGI is like a million smart people working 24/7, it will not magically solve all problems or remove the need for empirical data.
Prediction · 2025 |
AI timelines and forecasting |
| Not checkable |
Not checkable |
2025 |
We probably have similar dumb ideas, but one, the most obvious one, is we think it’s okay to eat meat, and probably we should stop thinking…
Not checkable · 2025 |
|
| Unresolvable |
Prediction |
2025 |
Many different AGIs will probably appear in many places once AGI is reached.
Prediction · 2025 |
AI timelines and forecasting |
| Not checkable |
Not checkable |
2025 |
But someday, I think people will see.
Not checkable · 2025 |
|
| Not yet due |
Prediction |
2025 |
It would be improbable for all of those outstanding AI problems to be solved within five years.
Prediction · 2025 |
AI timelines and forecasting |
| Not researched |
Factual |
2025 |
The best neuro-symbolic AI systems to date have been narrowly engineered.
Factual · 2025 |
AI capabilities |
| Mixed |
Factual |
2025 |
Current AI systems lack the knowledge graph needed to handle easy facts like those on the first page of Wikipedia.
Factual · 2025 |
AI capabilities |
| Not researched |
Factual |
2025 |
Current AI cannot be reliably given simple constraints, such as not explaining how to make biological weapons, and be expected to follow them.
Factual · 2025 |
AI safety and alignment |
| Not researched |
Factual |
2025 |
AI developers have chosen to pursue model-free and knowledge-representation-free systems.
Factual · 2025 |
AI capabilities |
| Not checkable |
Not checkable |
2025 |
I think for me, you need a world model of entities and their properties, stable over time, and that is tied closely together with knowledge…
Not checkable · 2025 |
|
| Not checkable |
Not checkable |
2025 |
Which is, I think if you don’t have explicit knowledge as part of your system, as first class citizens, you’re limited in control. And I…
Not checkable · 2025 |
|
| Not checkable |
Not checkable |
2025 |
And we want to get to one where they’re smart enough to follow the constraints, but not, you know, resisting what we tell them.
Not checkable · 2025 |
|
| Not checkable |
Not checkable |
2025 |
You’ve got some money to do it.
Not checkable · 2025 |
|
| Not researched |
Factual |
2025 |
The human ear lacks a good sense of what counts as correct in 19-tone music systems.
Factual · 2025 |
Other |
| Not checkable |
Not checkable |
2025 |
So to my mind, there are like hundreds of puzzles we haven’t solved.
Not checkable · 2025 |
|
| Not checkable |
Not checkable |
2025 |
We don’t need AGI to eliminate some human work.
Not checkable · 2025 |
|
| Not checkable |
Not checkable |
2025 |
I don’t really see that.
Not checkable · 2025 |
|
| Not checkable |
Not checkable |
2025 |
There’s work to be done, but there’s, you know, reasonable consensus in society to get started on it.
Not checkable · 2025 |
|
| Not checkable |
Not checkable |
2025 |
I think quantifier bindings, maybe.
Not checkable · 2025 |
|