Gary Marcus 2029 AI predications

Ṁ300Ṁ566

2028

20%

The LLM race will be basically a tie between China and U.S.

67%

Pure LLMs will still hallucinate (alternative architectures may do better).

50%

Pure LLMs will still make stupid errors (alternative architectures may do better).

42%

Profit margins for LLMs will be slim.

76%

There will still be plenty of jobs in which humans experts far outclass AI.

84%

Some current (2025) occupations, though, will no longer exist.

56%

Driverless taxi rides will be common, but still in limited cities, available in less than 50% of the world’s major (100k+) cities.

87%

Domain-specific models will still outperform general-purpose chatbots in many domains (board games, video games, protein folding, logistics, navigation, etc).

30%

Humanoid home robots over 5’ tall will still be demos, not in widespread release.

For his 2024 predictions, his own resolutions are somewhat debatable. So I will try to resolve this by "general consensus" of some sort. As that is not super well-defined, I will not be betting.

Market context

Technology

Technical AI Timelines

OpenAI

AI Impacts

Get

1,000

to start trading!

2 Comments

14 Holders

51 Trades

Sort by:

Nice market, if you're curious, check out my criteria that I used for /realDonaldTrump/gary-marcus-2026-ai-predictions and LMK what you think.

@realDonaldTrump While your criteria are more well-defined, I am not sure they are better at reaching "the correct objective resolution of the question".
- I am critical of letting people score their own predictions, have seen some questionable takes in the past (not necessarily from him, but in general).
- The fact that "LLMs hallucinate and make errors" are things we are betting on here means I would also be hesitant to hand over resolution to an LLM.
- Polls for resolution are also kinda questionable because traders have an interest in how things resolve.
I think for most of these questions, the answer will be very clear and I will be able to resolve them without any discussion. For any that are murky, I will try to foster discussion about how to resolve them once we get closer I guess. I know that fuzzy resolution criteria are bad, but also that's also why we are on Manifold right? We can strive for real correctness at the cost of fuzzy criteria because it's just internet points.