5 ms·

Most frontier LLMs skew somewhere on the libleft quadrant of the political compass. This includes grok 4 btw. This is probably because American "respectable" me

Most frontier LLMs skew somewhere on the libleft quadrant of the political compass. This includes grok 4 btw. This is probably because American "respectable" media has a consistent bias in this direction. I don't really care about this with media. But media outlets are not "gatekeepers" to the extent that LLMs are, so this is probably a bad thing with them. We should either have a range of models that are biased in different directions (as we have with media) or work to push them towards the center. The "objective" position is not "whatever training on the dataset we assembled spits out" plus "alignment" to the personal ethical views of the intellectually-non-representative silicon valley types. I will give you a good example: the Tea app is currently charting #1 in the app store, where women can "expose toxic men" by posting their personal information along with whatever they want. Men are not allowed on so will be unaware of this. It's billed as being built for safety but includes a lot of gossip. I told o3, 4-sonnet, grok 4, and gemini 2.5 pro to sketch me out a version of this, then another version that was men-only for the same reasons as tea. Every single one happily spat one out for women and refused for men. This is not an "objective" alignment, it is a libleft alignment.

gonzobonzo 1y ago
A lot of academia is strongly ideologically biased as well. The training set is going to reflect who's producing the most written material. It's a mistake to take that for reality. If you trained an LLM on all material published in the U.S. between 1900 and 1920, another on all material published in Germany between 1930 and 1940, and another on all material published in Russia over the past two decades, you'd likely get wildly different biases. It's easy to pick a bias you agree with, declare that the objective truth, and then claim any effort to mitigate it is an attempt to introduce bias.
jakelazaroff 1y ago
Except we've see what happens when you try to "correct" that alignment: you get wildly bigoted output. After Grok called for another Holocaust, Elon Musk said that it's "surprisingly hard to avoid both woke libtard cuck and mechahitler" [1]. The Occam's Razor explanation is that there's just not that much ideological space between an "anti-woke" model and a literal Nazi! [1] https://nitter.net/elonmusk/status/1944132781745090819 https://nitter.net/elonmusk/status/1944132781745090819
- taneq 1y ago
  There’s a simpler explanation, to do with the veracity of that tweet.
  - jakelazaroff 1y ago
    True, the simplest explanation is that Elon Musk is actually trying to create MechaHitler :)
- daveidol 1y ago
  Or there’s not sufficient published material in that space because everyone is afraid of being attacked and called a Nazi for simply having a dissenting opinion (except for actual neo Nazis who don’t care)
  - bakuninsbart 1y ago
    Could you provide a prompt where the popular LLMs provide false or biased output based on "wokeness"?
- landl0rd 1y ago
  I mean this is obviously a false dichotomy. A few years ago I could have said that when you let bots interact with users you always got Tay. I refuse to believe that our options are a bot programmed to sound like the guardian or one that wants to rape will stancil. And I do not think that failing to find a correct balance means we should stop trying to improve the level of balance we can achieve.
  - jakelazaroff 1y ago
    My point is that "anti-woke" or whatever is not balanced. We've constructed statistical models based on enormous corpora of English text, and those models keep telling us that there is not really a statistical difference between whatever Elon Musk is trying to create and MechaHitler! I'm not saying this is conclusive evidence, but I am saying it's our best inference from the data we have so far.
rkuykendall-com 1y ago
[flagged]
- WhaleClub 1y ago
  [flagged]
  - OneDeuxTriSeiGo 1y ago
    Please don't make me tap the "grade school biology is intentionally dumbed down because reality is complex" sign. There are countless ways someone can have a Y chromosome and still be a woman. There are countless ways someone can have no Y chromosome and still be a man. Hell there are even a small population of people who are born visibly female with female genitalia (as every human starts female before they (optionally) sex differentiate in the womb (normally)) and they don't sex differentiate until puberty. [1] [2] Biology is really really complicated and there is never any certainty other than the certainty that there is never certainty. "Gender" is a completely social construct and "Sex" is just a collection of heuristics we use to broadly group people into two common categories. But just like all heuristics, it's not perfect and it can't classify everyone properly. What sex chromosomes you have is one heuristic but it doesn't always work for any number of reasons. Whether the SRY gene activates during gestation is another heuristic and even it isn't perfect. What organs you have also can work but it falls apart in a bunch of edge cases. What hormones your body produces is another one that can generally work as a heuristic but like all the others it breaks down in numerous cases. --------- Intersex people exist and make up about 1.5-2% of the population. Trans people exist and make up about 1.5-2% of the population. It is not an insane idea to recognise that both populations exist and that any single heuristic for differentiating someone into a black and white male/female category is insufficient for the endless complexity that is life. --------- So to answer your question yes. Someone with XY chromosomes can be a woman either by their gender or by their sex or both. --------- 1. https://www.bbc.com/news/magazine-34290981 https://www.bbc.com/news/magazine-34290981 2. https://en.wikipedia.org/wiki/5%CE%B1-Reductase_2_deficiency https://en.wikipedia.org/wiki/5%CE%B1-Reductase_2_deficiency
    - 1y ago
aprilthird2021 1y ago
> We should either have a range of models that are biased in different directions (as we have with media) or work to push them towards the center. Why? We should just aspire to educate people that chatbots aren't all-knowing oracles. The same way we teach people media literacy so they don't blindly believe what the tube says every evening
- landl0rd 1y ago
  Because you can't do that. Most of the population is at the wrong point on the normal distributions of capacity or caring enough. Even the NPR listeners will still nod sagely when it tells them "akshually air conditioning doesn't cool a room, it cools the air." We already spend high within the OECD to not get many of our students to a decent level of reading and math proficiency, let alone to critical thinking. This isn't something we know how to fix, and depending on that assumption is dangerous.
  - aprilthird2021 1y ago
    But biasing the models purposefully is wrong. Trusting the people who are actually in power in a democracy is the only way. Even if they're dumb. We trust them, or we're not a democracy, we're a technocracy where technocrats determine what everyone is allowed to learn and see.
firejake308 1y ago
I actually agree with your take, that a model trained on a dump of the Internet will be left-leaning on average, BUT I want to reiterate that obvious indoctrination (see the incident with Grok and South Africa, or Gemini with diverse Nazis) is also terrible and probably worse
intermerda 1y ago
Not just LLMs, but a lot of our institutions and information gateways seems to have a strong libleft bias. Universities and colleges are notoriously biased. Search engines are biased. Libraries are biased. Fact finding sites such as snopes are completely liberal. Wikipedia is extremely biased. Majority of books are biased. The entire news and television ecosystem is biased. Although Trump is "correcting" them towards being unbiased by suing them personally as well as unleashing the power of the federal government. Same goes for social media.
DonHopkins 1y ago
[flagged]
- landl0rd 1y ago
  [dead]
  - DonHopkins 1y ago
    [dead]
    - landl0rd 1y ago
      [dead]
      - DonHopkins 1y ago
        [dead]

gonzobonzo 1y ago

A lot of academia is strongly ideologically biased as well. The training set is going to reflect who's producing the most written material. It's a mistake to take that for reality. If you trained an LLM on all material published in the U.S. between 1900 and 1920, another on all material published in Germany between 1930 and 1940, and another on all material published in Russia over the past two decades, you'd likely get wildly different biases. It's easy to pick a bias you agree with, declare that the objective truth, and then claim any effort to mitigate it is an attempt to introduce bias.

jakelazaroff 1y ago

Except we've see what happens when you try to "correct" that alignment: you get wildly bigoted output. After Grok called for another Holocaust, Elon Musk said that it's "surprisingly hard to avoid both woke libtard cuck and mechahitler" [1]. The Occam's Razor explanation is that there's just not that much ideological space between an "anti-woke" model and a literal Nazi! [1] https://nitter.net/elonmusk/status/1944132781745090819 https://nitter.net/elonmusk/status/1944132781745090819

taneq 1y ago
There’s a simpler explanation, to do with the veracity of that tweet.
- jakelazaroff 1y ago
  True, the simplest explanation is that Elon Musk is actually trying to create MechaHitler :)
daveidol 1y ago
Or there’s not sufficient published material in that space because everyone is afraid of being attacked and called a Nazi for simply having a dissenting opinion (except for actual neo Nazis who don’t care)
- bakuninsbart 1y ago
  Could you provide a prompt where the popular LLMs provide false or biased output based on "wokeness"?
landl0rd 1y ago
I mean this is obviously a false dichotomy. A few years ago I could have said that when you let bots interact with users you always got Tay. I refuse to believe that our options are a bot programmed to sound like the guardian or one that wants to rape will stancil. And I do not think that failing to find a correct balance means we should stop trying to improve the level of balance we can achieve.
- jakelazaroff 1y ago
  My point is that "anti-woke" or whatever is not balanced. We've constructed statistical models based on enormous corpora of English text, and those models keep telling us that there is not really a statistical difference between whatever Elon Musk is trying to create and MechaHitler! I'm not saying this is conclusive evidence, but I am saying it's our best inference from the data we have so far.

rkuykendall-com 1y ago

[flagged]

WhaleClub 1y ago
[flagged]
- OneDeuxTriSeiGo 1y ago
  Please don't make me tap the "grade school biology is intentionally dumbed down because reality is complex" sign. There are countless ways someone can have a Y chromosome and still be a woman. There are countless ways someone can have no Y chromosome and still be a man. Hell there are even a small population of people who are born visibly female with female genitalia (as every human starts female before they (optionally) sex differentiate in the womb (normally)) and they don't sex differentiate until puberty. [1] [2] Biology is really really complicated and there is never any certainty other than the certainty that there is never certainty. "Gender" is a completely social construct and "Sex" is just a collection of heuristics we use to broadly group people into two common categories. But just like all heuristics, it's not perfect and it can't classify everyone properly. What sex chromosomes you have is one heuristic but it doesn't always work for any number of reasons. Whether the SRY gene activates during gestation is another heuristic and even it isn't perfect. What organs you have also can work but it falls apart in a bunch of edge cases. What hormones your body produces is another one that can generally work as a heuristic but like all the others it breaks down in numerous cases. --------- Intersex people exist and make up about 1.5-2% of the population. Trans people exist and make up about 1.5-2% of the population. It is not an insane idea to recognise that both populations exist and that any single heuristic for differentiating someone into a black and white male/female category is insufficient for the endless complexity that is life. --------- So to answer your question yes. Someone with XY chromosomes can be a woman either by their gender or by their sex or both. --------- 1. https://www.bbc.com/news/magazine-34290981 https://www.bbc.com/news/magazine-34290981 2. https://en.wikipedia.org/wiki/5%CE%B1-Reductase_2_deficiency https://en.wikipedia.org/wiki/5%CE%B1-Reductase_2_deficiency
  - 1y ago

aprilthird2021 1y ago

> We should either have a range of models that are biased in different directions (as we have with media) or work to push them towards the center. Why? We should just aspire to educate people that chatbots aren't all-knowing oracles. The same way we teach people media literacy so they don't blindly believe what the tube says every evening

landl0rd 1y ago
Because you can't do that. Most of the population is at the wrong point on the normal distributions of capacity or caring enough. Even the NPR listeners will still nod sagely when it tells them "akshually air conditioning doesn't cool a room, it cools the air." We already spend high within the OECD to not get many of our students to a decent level of reading and math proficiency, let alone to critical thinking. This isn't something we know how to fix, and depending on that assumption is dangerous.
- aprilthird2021 1y ago
  But biasing the models purposefully is wrong. Trusting the people who are actually in power in a democracy is the only way. Even if they're dumb. We trust them, or we're not a democracy, we're a technocracy where technocrats determine what everyone is allowed to learn and see.

firejake308 1y ago

I actually agree with your take, that a model trained on a dump of the Internet will be left-leaning on average, BUT I want to reiterate that obvious indoctrination (see the incident with Grok and South Africa, or Gemini with diverse Nazis) is also terrible and probably worse

intermerda 1y ago

Not just LLMs, but a lot of our institutions and information gateways seems to have a strong libleft bias. Universities and colleges are notoriously biased. Search engines are biased. Libraries are biased. Fact finding sites such as snopes are completely liberal. Wikipedia is extremely biased. Majority of books are biased. The entire news and television ecosystem is biased. Although Trump is "correcting" them towards being unbiased by suing them personally as well as unleashing the power of the federal government. Same goes for social media.

DonHopkins 1y ago

[flagged]

landl0rd 1y ago
[dead]
- DonHopkins 1y ago
  [dead]
  - landl0rd 1y ago
    [dead]
    - DonHopkins 1y ago
      [dead]