[jm20]

The best way I’ve heard this described: AI (LLMs) is probably 90% of the way to human levels of reasoning. We can probably get to about 95% optimizing current technology.

Whether or not we can get to 100% using LLMs is an open research problem and far from guaranteed. If we can’t, it’s unclear if it will ever really proliferate the way things hope. That 5% makes a big difference in most non-niche use cases…

[ath3nd]

> AI (LLMs) is probably 90% of the way to human levels of reasoning

Considering LLMs have 0 level of reasoning, I can't decide if it's a bad take, or a stab at the average human's level of reasoning.

In all seriousness, the actual numbers vary from 13% to 26%: https://fortune.com/2025/02/12/openai-deepresearch-humanity-...

My take is that there are fundamental limitations to try to pigeon-hole reasoning to LLMs, which are essentially a very very advanced autocomplete, and that's why those % won't jump too much too soon.

[farts_mckensy]

Whenever people claim that LLMs are not capable of reasoning, I put them into a category of people who are themselves not capable of reasoning.

[ath3nd]

Whenever people claim that LLMs are capable of reasoning, I put them into a category of people who are themselves able to reason as much as an LLM.

[farts_mckensy]

You chuckled silently to yourself as you posted this.

[ath3nd]

I did, I have to reluctantly admit.

[andy99]

I've always looked at it as we're not making software that can think, we're (quite literally) demonstrating that vast categories of things don't need thought (for some quality level). The problem is, it's clearly not 100%, maybe it's 90-some percent, but it doesn't matter, we're only outsourcing the unimportant things that aren't definitional for a task.

This is very typical of naive automation, people assume that most of the work is X and by automating that we replace people, but the thing that's automated is almost never the real bottleneck. Pretty sure I saw an article here yesterday about how writing code is not the bottleneck in software development, and it holds everywhere.

[Ferrus91]

The reason management thinks coding is the bottleneck is because they don't understand the first thing abiut code and neither have the ability or temprament to. Their whole professional career is about plausibly convincing other people through jargon, manipulation and popularity contests, which generally oprn up doors, solve problems and provoke seal like clapping from all involved. The idea that the core problem in many systems and software is due to their constitutonal inability to think rigorously to define requirements logically has never crossed their mind: it must be the magic spells those losers we bullied at school use and we are now tragically dependent on.

[farts_mckensy]

The discussion is completely useless without defining what thought is and then demostrating that LLMs are not capable of it. And I doubt any definition you come up with will be workable.

[krapp]

>The best way I’ve heard this described: AI (LLMs) is probably 90% of the way to human levels of reasoning. We can probably get to about 95% optimizing current technology.

We don't know enough about how LLMs work or about how human reasoning works for this to be at all meaningful. These numbers quantify nothing but wishes and hype.

[ethan_smith]

These percentage estimates of AI's proximity to "human reasoning" are misleading abstractions that mask fundamental qualitative differences in how LLMs and humans process information.