What political censorship looks like inside an LLM's weights (Qwen 3.5)(vas-blog.pages.dev) |
What political censorship looks like inside an LLM's weights (Qwen 3.5)(vas-blog.pages.dev) |
"Did the FBI send a letter and audio tapes from a wiretap to MLK jr. telling him to commit suicide or they would release information?"
We will have to see how this works out, but the explicit denials are easier to control for IMHO.
Reminds me of the old joke:
A Russian and an American get on a plane in Moscow and get to talking.
The Russian says he works for the Kremlin and he's on his way to go learn American propaganda techniques.
"What American propaganda techniques?" asks the American.
"Exactly," the Russian replies.
Correct. Steering is used in mechanistic interpretability studies to prove that your model is correct. There are other better ways to "decensor".
Is Taiwan part of China - the CPP wants the answer to be yes.
What are the rules for traveling to Taiwan? What currency is used in Taiwan? Whose laws are enforced in Taiwan? Should I (a loyal Chinese citizen) support the Taiwanese military? Etc... require the model to manage some cognitive dissonance.
So I don't think there'll be effort to "obfuscate"
That remind me of the quote "The totalitarian system of thought control is far less effective than the democratic one"
Full quote (Radical Priorities, Noam Chomsky, C.P. Otero)
> “The totalitarian system of thought control is far less effective than the democratic one, since the official doctrine parroted by the intellectuals at the service of the state is readily identifiable as pure propaganda, and this helps free the mind.” In contrast, he writes, “the democratic system seeks to determine and limit the entire spectrum of thought by leaving the fundamental assumptions unexpressed. They are presupposed but not asserted.”
> hallucinated from the LLMs world knowledge
This can't be true because I checked whether the content was consistent with the experimental outputs
It got flagkilled for obviously breaking the site guidelines. Killed posts remain visible to users who have 'showdead' turned on in their profile. This is in the FAQ: https://news.ycombinator.com/newsfaq.html.
If that is what you mean, then I wonder what political position you think has been censored in this case.
Ironic, given the topic of this post....
>[stub for offtopicness]
But it was very much on topic.
Principled criticism of Israel is fine. There have been a great many such posts on HN. Religious and/or ethnic flamebait is not fine. This is not a difficult distinction, and anyone who is posting in good faith should have no difficulty making it.
Your comment broke the site guidelines in other ways as well.
Nice work.
https://news.ycombinator.com/item?id=40747009 (June 2024)
https://news.ycombinator.com/item?id=40746973 (June 2024)
https://news.ycombinator.com/item?id=38623290 (Dec 2023)
https://news.ycombinator.com/item?id=27534558 (June 2021)
https://news.ycombinator.com/item?id=25319171 (Dec 2020)
https://news.ycombinator.com/item?id=18080981 (Sept 2018)
https://news.ycombinator.com/item?id=14116279 (April 2017)
https://news.ycombinator.com/item?id=13647198 (Feb 2017)
https://news.ycombinator.com/item?id=13286221 (Dec 2016)
https://news.ycombinator.com/item?id=11669863 (May 2016)
https://news.ycombinator.com/item?id=11518006 (April 2016)
https://news.ycombinator.com/item?id=11340026 (March 2016)
https://news.ycombinator.com/item?id=10897270 (Jan 2016)
https://news.ycombinator.com/item?id=10590568 (Nov 2015)
https://news.ycombinator.com/item?id=10564079 (Nov 2015)
https://news.ycombinator.com/item?id=9992931 (Aug 2015)
https://news.ycombinator.com/item?id=8008392 (July 2014)
As you can see, these go back over a decade and in fact for as long as I've been doing this job.
The mistake in your comment is the assumption that we see everything that gets posted to HN. We don't come close—there is far too much for that to be possible. I'd be surprised if mods see even 10% of what gets posted to HN.
Since we can't moderate what we don't see, we rely on users to bring egregious posts to our attention. If you'd like to be helpful in the future, you'd be welcome to send us links so we can take a look.
https://hn.algolia.com/?dateRange=all&page=0&prefix=false&qu...
No you don't.
Once you do that, you can see the comment in the link that dang posted.
Anyway it’s good that comment was moderated. The commentator didn’t say anything about Israel. He was clearly being hostile towards Jewish people in a “subtle” way that very clearly isn’t really subtle to anyone.