Gary Marcus

Rank 16 of 47

Score 76

@gcolbourn when they can use with reliability and accuracy as humans experts can, that will be a step forward.

otoh it is also looking like a victory of neurosymbolic AI if LLMs can only work by recourse to symbolic tools

6/18/2025, 12:51:58 PM

In reply to:

Greg ⏹️ Colbourn

@gcolbourn

23h

@GaryMarcus Is it really that much of a stretch for AI systems to be able to use those tools as well as humans can?

Gary Marcus

@GaryMarcus

Now show me the same graphs but for symbolic systems like calculators, spreadsheets, gps navigation systems, and databases.

Benjamin Todd

@ben_j_todd

Why can AIs code for 1h but not 10h?

A simple explanation: if there's a 10% chance of error per 10min step (say), the success rate is:

1h: 53%
4h: 8%
10h: 0.002%

@tobyordoxford has tested this 'constant error rate' theory and shown it's a good fit for the data

chance of

The statement engages in a discussion about the capabilities of AI systems, particularly in relation to human experts and the potential role of neurosymbolic AI. It suggests that achieving human-level reliability and accuracy would be a significant advancement and implies that the integration of symbolic tools might be necessary for LLMs to reach this level.

Principle 1:
I will strive to do no harm with my words and actions.
The statement does not cause harm and is focused on technological advancement. [+1]
Principle 3:
I will use my words and actions to promote understanding, empathy, and compassion.
It promotes understanding by discussing the potential future of AI and its implications. [+1]
Principle 4:
I will engage in constructive criticism and dialogue with those in disagreement and will not engage in personal attacks or ad hominem arguments.
The statement engages in constructive dialogue about AI capabilities without personal attacks. [+1]