I'm a little tired of people complaining that DeepSeek won't answer questions about Tiananmen Square. Where was this energy when OpenAI refused to answer questions about US politics? Try asking Gemini about Donald Trump's felonies and see what it says.
(I am not equating Chinese bad things with Western bad things. But I am tired of the intentional blindness that we all sometimes have.)
@davidnjoku every LLM is inherently biased based on both the views of the team behind it and the materials it’s trained on.
How we’re this far down the “AI” conversation with so many people choosing to disregard that is somewhat alarming.
Of course next up is some LLM aggregator that uses its own model to query all the others then patch together a response taking into account all replies.
@wiredfire I'd love the honesty if people said, "I want an LLM that has the same biases as me." Instead we pretend that ours are good and theirs is bad.
A couple of years ago, Dall-E and Bard refused to generate images of Black doctors. They seemed to think that doctors had to be white. There were barely any complaints - until Google overcorrected and started generating images of Black founding fathers. And that made all the front pages. Hmm, strange.
@davidnjoku reminds me of the old “puzzler”:
A man and his son are in a car accident. They’re taken to hospital. The doctor steps back and says “I cannot treat this boy, he is my son”. How can this be?
Solution: women can be doctors..!
Alarming how many folk used to really be perplexed when that was posed.
We have such deep rooted biases and ignoring them is a Bad Idea™. Far better to be aware so we can consciously spot them.
@davidnjoku @wiredfire "I want an LLM that has the same biases as me" seems like it would turn very quickly into alternative fact laundering
@davidnjoku And DeepSeek is Open Source, so you can put Tiananmen Square back into your instance of it.
@jonathanpeterson Fair point.
@davidnjoku yes but china bad and authoritarian and USA good and free i am very smart
The difference is that people in the West can use either AI, but people in China can only use one.
@number6 I'm not sure that's true. I imagine they have, it will soon have, many AI chatbots in China - and not one of them will contain info about Tiananmen Square, just like not one of ours contains info about Trump's crimes.
I'm not saying both countries are exactly the same, but I am saying we're not as great as we like to think.
The difference is that you absolutely will not be able to get info about Tiananmen there through any means in China.
Maybe AI doesn't want to talk about Trump, but all of the standard search engines will happily serve up sites talking about Trump's convictions.
I don't know why people want to soft-speak on this. China is an actual authoritarian, totalitarian government that actively suppresses free speech.
Free speech is a fundamental human right, not a 1st world luxury.
Yeah, perhaps people are just getting more stupid, l most people like us will look at several sources for info.
@zleap You give me too much credit. I'm lazy; I know that I should rely on diverse sources of information, but it's too much effort so I don't. ,
@davidnjoku I asked my refrigerator to tell me about Tiananmen Square. Its only response was a muffled clanking noise that could have been the ice maker.
Seriously this is a stupid metric. As you say, lots of LLMs have lots of limitations. Some of the things are sus to various non-equivalent degrees, sure, but like, come on.