ML Newsnew | past | best | rss | submit about
How Far Are We from AGI (arxiv.org)
14 points by belter on May 17, 2024 | hide | past | pdf | 7 comments on HN

In plain words: A wide-ranging review sorts what general intelligence would need into three layers—inside the model, at its interfaces, and across systems—then maps levels, tests, and steps to get there. It finds language models alone fall short, so we need better safety checks and system-level skills.

Abstract · How Far Are We From AGI: Are LLMs All We Need?

The evolution of artificial intelligence (AI) has profoundly impacted human society, driving significant advancements in multiple sectors. AGI, distinguished by its ability to execute diverse real-world tasks with efficiency and effectiveness comparable to human intelligence, reflects a paramount milestone in AI evolution. While existing studies have reviewed specific advancements in AI and proposed potential paths to AGI, such as large language models (LLMs), they fall short of providing a thorough exploration of AGI's definitions, objectives, and developmental trajectories. Unlike previous survey papers, this work goes beyond summarizing LLMs by addressing key questions about our progress toward AGI and outlining the strategies essential for its realization through comprehensive analysis, in-depth discussions, and novel insights. We start by articulating the requisite capability frameworks for AGI, integrating the internal, interface, and system dimensions. As the realization of AGI requires more advanced capabilities and adherence to stringent constraints, we further discuss necessary AGI alignment technologies to harmonize these factors. Notably, we emphasize the importance of approaching AGI responsibly by first defining the key levels of AGI progression, followed by the evaluation framework that situates the status quo, and finally giving our roadmap of how to reach the pinnacle of AGI. Moreover, to give tangible insights into the ubiquitous impact of the integration of AI, we outline existing challenges and potential pathways toward AGI in multiple domains. In sum, serving as a pioneering exploration into the current state and future trajectory of AGI, this paper aims to foster a collective comprehension and catalyze broader public discussions among researchers and practitioners on AGI.

Tao Feng, Chuanyang Jin, Jingyu Liu, Kunlun Zhu, Haoqin Tu, Zirui Cheng, Guanyu Lin, Jiaxuan You
arXiv:2405.10313 · cs.AI, cs.CL, cs.CY, cs.LG · submitted May 16, 2024 · updated Nov 24, 2024
abstract · pdf · html

add comment on HN
Also discussed: May 2024 (5 points, 4 comments)
jbandela1 on May 17, 2024 [–]
I would say we don’t really know what AGI is.

As humans, we are basically working with an N of 1 and trying to figure out what is the essential part.

At one time, being able to do math might have seemed to be it. Maybe playing chess? Maybe having a conversation (Eliza)? Mane generating art? Maybe generating summaries? They are impressive but not quite it.

We are like the shopper who only has a vague idea of what they are looking for but can tell after some inspection when it is not quite it.

I think developments in AI have been helpful in helping us learn more about human intelligence as well.

reply
jokoon on May 17, 2024 [–]
This is like being in 1200 and asking "how far we are from flying like birds".

One very important lesson I learned while talking with a physics professor, is that often, breakthrough in science are made either by chance, or because of things in other fields.

I'll repeat that same complaint: study cognitive science, neuroscience, and psychology, and have more multi-domain/epistemology stuff about a definition of intelligence, and ask broader questions, especially in the realm of evolution. STOP testing ML models on GPUs.

reply
klipklop on May 17, 2024 | parent [–]
Things seem to be speeding up though. Look at the short period of time we went from flying a basic plane to landing on the moon. It was not hundreds of years between those two events. Perhaps as human population grows the "by chance" part increases/improves? More humans means more scientific die rolls.

But yeah who knows if we will stumble upon the solution to AGI next week or in 1000 years.

reply
benmanns on May 17, 2024 [–]
If anyone is interested, there are markets on Manifold trying to predict this: https://manifold.markets/ai

To be honest, even as a participant and enthusiast, I’m not sure how useful or accurate it is. I can push the prediction by 13 years with $1 worth of Manifold currency.

reply
chx on May 17, 2024 [–]
First sentence:

> The evolution of artificial intelligence (AI) has profoundly impacted human society, driving significant advancements in multiple sectors.

OK this is not serious science.

reply
qsort on May 17, 2024 | parent [–]
Remove the word "serious". It's an op-ed, we can take it for what it is. The opinion of a scientist isn't science, but might be interesting.
reply
chx on May 17, 2024 | parent [–]
Let me clarify: I am not going to waste my time reading the rambling of someone drunk on the AI kool aid.

> profoundly impacted human society,

So far there are fears, well founded fears but any impact is yet to be measured and "profound" has certainly not happened yet.

> driving significant advancements in multiple sectors.

ROTLFMAO. There's nothing more to be said. What is happening is asshole corporate owners destroying their own companies by firing competent workers and replacing them with a bullshit printer. Not quite sure how that's an "advancement".

reply
About | RSS | RSS (all) | HN arXiv