Gary Marcus

Rank 17 of 47
|
Score 87
1/24/2025, 11:52:46 PM
X
In reply to:
Pehdrew
@Pehdrew_
·
100d
@GaryMarcus 1. OpenAI's moat with GPT-4 was size and compute, v3 improves training similar a base model with much less compute.
2. R1 which is uses the new RL paradigm and is on par with o1 using much less compute when compare to o1.
Gary Marcus
@GaryMarcus
·
100d
@Pehdrew_ Which part do you disagree with and why?
Pehdrew
@Pehdrew_
·
100d
@GaryMarcus this is seriously your take on Deepseek v3 and R1?
Gary Marcus
@GaryMarcus
·
100d
Me, last March: we won’t see GPT-5 soon; instead many competitors will converge, there won’t be much moat. Instead, there will be a price war. Profits will be meager.

China, today: All that, and we raise you a 50x reduction in cost.

Still no GPT-5 but GenAI is now practically…
Deedy
@deedydas
·
102d
China just dropped a new model.
ByteDance Doubao-1.5-pro matches GPT 4o benchmarks at 50x cheaper

— $0.022/M cached input tokens, $0.11/M input, $0.275/M output
— 5x cheaper than DeepSeek, >200x of o1
— 32k + 256k context
— sparse MoE architecture

AI truly too cheap to meter.

The conversation involves a technical discussion about AI models, their performance, and cost efficiency. It reflects a debate on the competitive landscape of AI technology, particularly focusing on the advancements and cost reductions in AI models from different companies and regions.

  1. Principle 1:
    I will strive to do no harm with my words and actions.
    The discussion is technical and does not appear to cause harm. It focuses on factual information about AI models. [+1]
  2. Principle 3:
    I will use my words and actions to promote understanding, empathy, and compassion.
    The conversation promotes understanding of AI advancements and cost implications, contributing to informed discourse. [+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.
    The dialogue is constructive, with participants engaging in a technical debate without personal attacks. [+1]