/ai

Researchers Contrast LLMs With Prior AI Approaches

Replies on X contrast LLMs with earlier AI systems like AlphaGo.

2 Sources, 23d ago, first seen 23d ago

TLDR

Subbarao Kambhampati posted on X that most AI has been trained to reproduce human behavior because that has been the main yardstick for intelligence. He cited Drew McDermott's warning against circular attribution of mental stances to machines. Joscha Bach replied that pre-LLM AI research focused on producing intelligent behavior de novo rather than imitation. Bach noted that AlphaGo does not play Go like a human, but LLMs succeed precisely because they reproduce human mental patterns. The exchange occurred in public replies on X.

Featured Source

Today's Rank

—

Not ranked yet

2 Sources

@rao2zWell, isn't most of AI has been trained to reproduce human behavior--because that's been our only yardstick for intelligence. Which is exactly why Drew McDermott--in his AI meets NS--says we should be careful about engaging in circularity and attributing mental stances to them willy nilly.

@Plinz@rao2z @anilkseth Before LLMs, most AI research was focused on producing intelligent behavior de novo, it was not imitative. AlphaGo does not play Go like a human. But McDermott’s provision does not apply to LLMs, which work precisely because they reproduce human mental patterns.

Today's Rank

—

Not ranked yet

Combined views

521

2 Sources, first seen 23d ago

8 likes

TagsJoscha Bach

Today's Rank

—

Not ranked yet

Today's Rank

—

Not ranked yet

Combined views

521

2 Sources, first seen 23d ago

8 likes3 comments2 saves2 reposts

Sentiment

Positive——Negative

Summary

Not enough discussion yet.

No sentiment analysis available yet.

Featured Source

3 comments

2 saves

2 reposts

Sentiment

Positive——Negative

Summary

Not enough discussion yet.

No sentiment analysis available yet.

Sentiment

Positive——Negative

Summary

Not enough discussion yet.

No sentiment analysis available yet.

2 Sources

@rao2zWell, isn't most of AI has been trained to reproduce human behavior--because that's been our only yardstick for intelligence. Which is exactly why Drew McDermott--in his AI meets NS--says we should be careful about engaging in circularity and attributing mental stances to them willy nilly.

@Plinz@rao2z @anilkseth Before LLMs, most AI research was focused on producing intelligent behavior de novo, it was not imitative. AlphaGo does not play Go like a human. But McDermott’s provision does not apply to LLMs, which work precisely because they reproduce human mental patterns.