Gary Marcus

Rank 12 of 47
|
Score 83
4/22/2025, 9:28:25 PM
X
In reply to:
Colin Fraser
@colin_fraser
路
274d
@ebarcuzzi @GaryMarcus not quite 99.9 when we go higher than 40 digits
Colin Fraser
@colin_fraser
路
279d
Answer: 87%.

As usual, we are in this bizarre place where the LLM performs better than I would ever expect, and yet worse than a computer ever should. Not sure what to make of it.
Juggalos For Context馃尨馃ゥ
@ebarcuzzi
路
274d
@GaryMarcus @colin_fraser Addition and multiplication up to an ever expanding number of digits is one example.
Gary Marcus
@GaryMarcus
路
274d
@ebarcuzzi @colin_fraser on what is an LLM 99.9% accurate? source/link?
Juggalos For Context馃尨馃ゥ
@ebarcuzzi
路
274d
@colin_fraser @GaryMarcus Also no one really reports on break throughs in LLMs going from 99% accurate to 99.9% accurate on certain tasks. Not an attention grabbing headline.
Colin Fraser
@colin_fraser
路
274d
the set of things an LLM can do with about 70% accuracy is growing to be quite large, larger than anyone ever thought possible, but it remains quite challenging to efficiently use something that works 70% of the time

The conversation involves a technical discussion about the accuracy of language models (LLMs) in performing arithmetic operations with large numbers. The tone is analytical and focused on exploring the capabilities and limitations of LLMs. The participants are engaging in a constructive dialogue, sharing data and insights.

  1. Principle 1:
    I will strive to do no harm with my words and actions.
    The discussion is technical and does not involve harmful language or actions.聽[+1]
  2. Principle 3:
    I will use my words and actions to promote understanding, empathy, and compassion.
    The conversation promotes understanding of LLM capabilities and limitations, fostering a deeper comprehension of the technology.聽[+1]
  3. 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.
    Participants engage in constructive dialogue, building on each other's points without personal attacks.聽[+1]