its not useful for anything a human can do with actual oversight. it can be used as a minor minor search tool but i disparage it being used for active governance.
The article mentioned helping with reading public comments/feedback
Let’s say there are 10k comments, but in reality, there are only 100 unique problems/suggestions from the citizens due to duplicated feedback
It’s unrealistic for them to read all of them, so let’s assume a system they may currently use is randomly pick 200 comments to read to get a general sense of the public
However, let’s say a few of those 100 unique issues in the 10k comments are very important, but only have a small number of submissions due to whatever reason, and most of the other issues of the 100 are highly duplicated
So out of the 200 randomly picked comments, they miss some important but very infrequently submitted feedback
Now dump all that 10k into an LLM to categorize the duplicated submissions into a summary/a few comments. Count the number of comments in each compressed issue with the LLM. They still manually read 200 randomly picked comments and do the same categorization but manually to ensure that the LLM summarization doesn’t misinterpret the comments and it does the bucketing of issues correctly (check that the distribution of problems in the 200 is roughly the same as the one from the LLM). This is good enough to validate the LLM output
Now after the bucketing of duplicated comments, it’s possible for them to read all 100 unique comments with human eyes, since the leftover comments that couldn’t be compressed due to low number of duplications can all be read in their original form since they aren’t wasting time on the highly duplicated comments
filtering things through a chatbot is not at all helpful, we’ve seen how useless they are at identifying actual problems and issues, and it risks actually ignoring the unique concerns of the common people. It just receives feedback and spits it out in a slightly reliable manner. Although it could be just used for those things, and while it could be useful in some cases, erasing autonomous elements within a democratic system is not indicative toward a better society. Advocating for using a capitalist LLM to start replacing government functions is a catering to neoliberal and liberal elements.
They should be focusing more on cracking down on capitalist elements instead of seeing the US implementing bullshit and then say “we need to do that oh my goodness”. There’s definitely a liberal ideological bloc in china that absolutely worships the USA, and is deeply seated within the communist party. They should be focusing on implementing a democratically planned economy with computerized elements to solve the economic calculation problem.
That’s why there’s still the manual validation step with the randomly sampled set
Thats of course with the first implementation, but no matter it will increase its role within the government and start ‘streamlining’ relying on well known unreliable chat LLMs.
also, they are doing multiple things at once. not everybody is going to only focus on anti corruption
im not saying anti-corruption, im saying restoring a socialist economy and socialist government.
but no matter it will increase its role within the government and start ‘streamlining’ relying on well known unreliable chat LLMs.
That’s a big assumption that the government officials will necessarily become morons due to their decision to use Deep seek in ways that we don’t even know
That’s a big assumption that the government officials will necessarily become morons due to their decision to use Deep seek in ways that we don’t even know
its a big assumption to say that there wont be significant errors from incorporation of a glorified chatbot into governance will be helpful
Don’t be a Luddite, LLMs are incredibly useful for certain tasks. Don’t let capitalist stupidity and greed influence your opinions
its not useful for anything a human can do with actual oversight. it can be used as a minor minor search tool but i disparage it being used for active governance.
The article mentioned helping with reading public comments/feedback
Let’s say there are 10k comments, but in reality, there are only 100 unique problems/suggestions from the citizens due to duplicated feedback
It’s unrealistic for them to read all of them, so let’s assume a system they may currently use is randomly pick 200 comments to read to get a general sense of the public
However, let’s say a few of those 100 unique issues in the 10k comments are very important, but only have a small number of submissions due to whatever reason, and most of the other issues of the 100 are highly duplicated
So out of the 200 randomly picked comments, they miss some important but very infrequently submitted feedback
Now dump all that 10k into an LLM to categorize the duplicated submissions into a summary/a few comments. Count the number of comments in each compressed issue with the LLM. They still manually read 200 randomly picked comments and do the same categorization but manually to ensure that the LLM summarization doesn’t misinterpret the comments and it does the bucketing of issues correctly (check that the distribution of problems in the 200 is roughly the same as the one from the LLM). This is good enough to validate the LLM output
Now after the bucketing of duplicated comments, it’s possible for them to read all 100 unique comments with human eyes, since the leftover comments that couldn’t be compressed due to low number of duplications can all be read in their original form since they aren’t wasting time on the highly duplicated comments
filtering things through a chatbot is not at all helpful, we’ve seen how useless they are at identifying actual problems and issues, and it risks actually ignoring the unique concerns of the common people. It just receives feedback and spits it out in a slightly reliable manner. Although it could be just used for those things, and while it could be useful in some cases, erasing autonomous elements within a democratic system is not indicative toward a better society. Advocating for using a capitalist LLM to start replacing government functions is a catering to neoliberal and liberal elements.
They should be focusing more on cracking down on capitalist elements instead of seeing the US implementing bullshit and then say “we need to do that oh my goodness”. There’s definitely a liberal ideological bloc in china that absolutely worships the USA, and is deeply seated within the communist party. They should be focusing on implementing a democratically planned economy with computerized elements to solve the economic calculation problem.
That’s why there’s still the manual validation step with the randomly sampled set
also, they are doing multiple things at once. not everybody is going to only focus on anti corruption
Thats of course with the first implementation, but no matter it will increase its role within the government and start ‘streamlining’ relying on well known unreliable chat LLMs.
im not saying anti-corruption, im saying restoring a socialist economy and socialist government.
That’s a big assumption that the government officials will necessarily become morons due to their decision to use Deep seek in ways that we don’t even know
its a big assumption to say that there wont be significant errors from incorporation of a glorified chatbot into governance will be helpful
I’m a lud, fuck LLMs