[maebert]

author says we made no progress towards agi, also gives no definition for what the "i" in agi is, or how we would measure meaningful progress in this direction.

in a somewhat ironic twist, it seems like the authors internal definition for "intelligence" fits much closer with 1950s. good old-fashioned AI, doing proper logic and algebra. literally all the progress we made in ai in the last 20 years in ai is precisely because we abandoned this narrow-minded definition of intelligence.

Maybe I'm a grumpy old fart but none of these are new arguments. Philosophy of mind has an amazingly deep and colorful wealth of insights in this matter, and I don't know why this is not required reading for anyone writing a blog on ai.

[13years]

> or how we would measure meaningful progress in this direction.

"First, we should measure is the ratio of capability against the quantity of data and training effort. Capability rising while data and training effort are falling would be the interesting signal that we are making progress without simply brute-forcing the result.

The second signal for intelligence would be no modal collapse in a closed system. It is known that LLMs will suffer from model collapse in a closed system where they train on their own data."

[maebert]

I agree that those both are very helpful metrics, but they are not a definition of intelligence.

yes, humans can learn to comprehend and speak language with magnitudes less examples than llms, however we also have very specific hardware for that evolved over millions of years — it's plausible that language acquisition in humans is more akin to fine-tuning in llms than training them from ground up. Either way, this metric is comparing apples to oranges when it comes to comparing real and artificial intelligence.

model collapse is a problem in ai that needs to be solved, and maybe it's even a necessary condition for true intelligence, though certainly not a sufficient one, and hence not an equivalent definition of intelligence either.

[13years]

The bar you asked for was "meaningful progress". And as you state, "both are very helpful metrics", it seems the bar is met to the degree it can be.

I don't think we will see a definitive test as we can't even precisely define it. Other than heuristic signals such as stated above, the only thing left is just observing performance in the real world. But I think the current progress as measured by "benchmarks" is terribly flawed.