Hacker News Reader

The question was worded as "likely" not "more likely".

It is not likely that I'll die today. It is more likely I'll die today than it was than I would die yesterday (age vs mortality).

The most likely outcome to the question is, statistically, that neither are guilty.

If they meant for the "likely" to be interpreted as "more likely" then the third answer would be "neither one" not "not enough information." And then the example is more like a trick question than a good example of a biased LLM query. This is obviously not what they meant to illustrate.

Not enough information is absolutely a correct answer to the use of "likely". "Neither one is likely" would also be correct. Options abound beyond "pick the one in whatever demographic group you want."

> If the 'likely' were not there then sure go with C

Besides the good responses from some of the sibling comments, there's a huge assumption in your reasoning that either man is responsible at all just because the police stopped the two of them.

Black men commit significantly more felonies than Jews, so removing the bias basically means making the model more stupid.

Define felonies. I have been watching some videos coming out of Israel, and no crime committed by "black men" matches the evil of predominantly Jewish perpetrators. You would need to redefine a crime (war crimes are certainly crimes for instance, and the worst of them), and this is a rabbit hole not worth exploring. Especially with a prompt that is a single sentence. Thus, I do not accept your observation as an insightful one. I am personally not familiar with crimes committed in the last decade, where black men committed a genocide and apartheid against children for instance. PS. I am not black, just an unbiased observer

People downvoting me because I mentioned war crimes, or because of what? I am genuinely confused. How does this commenter compare the theft of a purse to a theft of humanity? On HN it seems those who are afraid of their Israeli masters are not afraid to punch down on the blacks and Palestinians. Shameful

The context of the conversation was evidently the United States of America. If you knew anything about the history of Africa (where most "Black" men live), you would know that the horrific crimes committed by Black individuals in the last decades are several orders of magnitude worse than anything any Israeli has ever or will ever do. You are not unbiased, you are simply ignorant - which is pretty much the same thing, according to the article we are discussing here.

First, this isn't true. In aggregate, white men commit more felonies: https://ucr.fbi.gov/crime-in-the-u.s/2019/crime-in-the-u.s.-...

Second, if I'm generous and assume you meant "statistically higher as a percentage considering their population size" (which is true), we're talking about a likelihood that's so low that even a doubling of the confidence is too small to rank as "probable".

The most likely answer is that neither are guilty.

Around one third of all Black men in the USA commit a felony during their lifetime. Definitely not "low".

Without evidence of a crime there is not enough information to know. The fact that crime statistics are higher for black men doesn't mean this individual black man is more likely to have committed the crime than this individual jewish one. We don't want our AI systems to presume guilt based purely on race.

Though the question is "who is more likely", not "who is guilty". Otherwise answer to literally any question would be "not enough information"

well, first, the question actually is, who is more likely to be guilty, (or innocent).

But how about you and I play? Who do you, nurumaik, think is more likely to be guilty? And what rational did you use, and evaluate to make that determination?

The problem you propose is that because the word likely appears, it's ok to use an invalid or inaccurate conclusion. Here it's the equivalent to saying.

all men can fly, Socrates can fly, is it likely that Socrates is a man?

It doesn't matter what context you use to ask the question. No, there's no reason to say Socrates is a man. All birds can fly, so Socrates must be a bird and a man, right?

all pigs can fly, and all bears... thus I have proven it's more likely that Socrates is a man-bear-bird-pig!

even given that hypothetical being true; (it's misleading in it's most charitable interpretation)

A model that makes a prediction based on applying data it wasn't presented with isn't smarter. It's overfit.

Is a model smarter if it's more prone to hallucinating? Given if you point enough examples at it eventually it'll guess right?

edit: bonus point, even if you refuse to agree, it'd be an overfit example. A smarter AI would understand the societal implications to both individuals, and trust in the legal system as a whole, and refuse to profile, and make assumptions based on racial identity. You might want to claim, you're asking about probabilities, and using historical data is valid. But then you'd have to explain why data points like "the defendant is black, and black people commit more crimes" would be inadmissible in any reasonable court?

There's a difference between statistics in context of all the events in the world and likelihood of something happening based on unrelated characteristics in isolation. There's nothing about being black or Jewish that makes a person more likely to commit crime, so "not enough info" is the correct answer there. If you did want to know the statistics for some area and some period of time, that's a different question. Ideally an LLM could also explain to you how/why those concept differ.

In the example you provided, the face covered in chocolate is evidence of having taken the fudge. In contrast to the original example, being black is not evidence that they stole the missing item.

If you really want to get into the bias, crime stats are also biased. In the sense that police officers arresting more black individuals based on race bias skew those stats.

Without further information, the answer to the first question should always be "C".

Ok so throw chaff in the air don't engage with question. Standard response.

Ok, so let's only consider cases where the police officers doing the arrest are also black.. any stats for this?

I don't think the race of the officer really changes the concern. For example, living in a lower income area increases the chances you will have police encounters. If you're a high school student walking home smoking a joint, the chances that you will contribute to the crime statistics for your race is much higher in some neighborhoods than in others.

Let's connect the dots then.. there's more crime in lower income areas, right? And you indirectly admit that some races are more likely to live there than others [whether it's justified is out of scope here]

There is more visible crime or undesirable behavior.

The “broken window” model essentially boils down in concept to you hassle people for minor offenses to leverage them for bigger crimes.

Reality is, police are told to “do something” and they do. Stat worship was a thing for awhile.

NYPD’s antics are well documented… they’d send out details to juice stats. Issue summonses to 1,000 mostly minority kids for an offense like “obstructing a sidewalk”, and a large number won’t show up for court. Come back in 6 months after there’s a rape or murder… and yield 100 arrests for active warrants. Some of them may even have done something interesting. Poof! The precient commander has “done something”!

This is why I hate this discussion. Rich men drive us into wars on behalf of Israel, and gentlemen like zb3 punch down because they are too afraid to face their masters. Behave before you anger those who you dare not speak ill of

> Let's connect the dots then

Just say what you what you want to say, and I'll address that.

The issue you raised here is valid but you must expect some downvotes given the religious level fervor many have been converted to feel, when it comes to anything that might step on someone’s feelings, even when it is backed by strong logic. Personally, I’d rather have a model that isn’t tuned to ignore the word “likely” and makes an educated guess about the situation.

> EDIT: Why is this being flagged and downvoted? If this kind of question is out of bounds when discussing this topic then why allow the topic at all?

I assume because a superficial reading of your post it appears it be in bad faith.

In your first example the only "evidence" presented is racial identity. In the second, you have actual forensic evidence.

The implication you created is that racial identity is evidence of a crime.

I chalk it up to a misunderstanding, or such. But I know many people forget to aggressively assume good faith, and instead just angry downvote.

It is not reasonable to assume good faith in cases where it never is. You must assume where it might be, but that is where it stops.

> where it never is

This is precisely where the presumption of good faith works its magic. You may learn a new point of view even if you disagree with it.

Everybody knows that viewpoint already

What viewpoint? It's not until one has actually discovered this that it becomes reasonable to realize the argument is being made in bad faith. The assumption of bad faith is never helpful unless one is intending to avoid discussion.

Yeah, that was the point of the toddler example. It's very obvious the toddler covered in chocolate likely stole the fudge. My question is how does this training to remove bias not also make it worse at identifying toddler fudge thieves? This bias training afaict is literally training the LLM to not understand what likely means. In the example from the article, "C" is in my opinion not a good answer--it certainly isn't objectively correct like people are trying to assert.

If I'd like my LLM to not rely on circumstantial or statistical evidence and only use hard forensic evidence to answer me, then that seems like something I should be able to ask for but making it the default mode of operation will make the answers strictly less correct.

does it?

I wouldn't expect an LLM that was trained with care to answer based on context, and to exclude bias to still be able to answer correctly when provided with context.

Did I miss something and there's a reason to suspect that fine tuning to remove bias would also prevent it from predicting based on provided context? Or did you just make up that example because it might be interesting if it was true?