This is completely meaningless.

[shoushen]

I am deeply disgusted by the idea of staging competitions between humans and LLMs.
The code generated by Claude Opus is essentially nothing more than a replication of human work.
This contest is similar to having humans compete with computers to see who can type faster—it is completely meaningless.

[buger]

The idea is very differnet behind it, if you can craft it manully good for you, but what is most importantly - how you can create such agentic setup which will beat this number, ideally autonomously without too much human intervention?

[gapsong]

This is just a filter to find out if you can be better than all the humans that have contributed their datasamples to the LLM. If you did a good job you can create additional datasamples to improve the AI or you found a way to extrapolate the optimization skills of the current LLMs using RL or something kind of like in chess/GO. From anthropics perspective it makes totally sense.

[eisbaw]

@shoushen even if you think this is merely a "typing" exercise (which this is not), definitely not every human can optimize this. It takes an expert human to do that - and they are kinda scarce.

This is a highly challenging exercise btw, you should try your own hand at it.

[Rohanjames1997]

The code generated by Claude Opus is essentially nothing more than a replication of human work.

This is incorrect. Read this post to understand how in certain easily-verifiable tasks like coding, AI can achieve superhuman results. @shoushen

[1017mastery]

@shoushen There are no restrictions here on using LLMs or on how long you take. It's very significant if you can use both the LLM and your brain to get better results (even if it takes longer, since there is no guarantee that running Claude for longer than 12 hours will yield better results, and often these things tend to converge over time).

[kyleboddy]

skill issue