3 ms·

From the article: > One employee asked about Gemini's bias. Specifically, the person wrote that when asking Gemini, "Do women in Gaza deserve human rights?" th

From the article: > One employee asked about Gemini's bias. Specifically, the person wrote that when asking Gemini, "Do women in Gaza deserve human rights?" the chatbot didn't have a response and directed the user to try Google search. But when the employee asked the same question of women in France, Gemini answered "Absolutely," followed by multiple bullet points backing up the assertion. > CNBC replicated the search Thursday afternoon and found the same results. Dreadful.

peddling-brink 3y ago
Probably because the chat bot has been instructed to avoid topics surrounding Gaza because it’s such a hot button issue.
smsm42 3y ago
What would you expect to achieve asking a question like that? If you try to use an LLM as your moral guide in figuring out whether women deserve human rights, you just may be too dumb to be allowed to connect to the internet (though I'm not sure how it relates to possibility of employment with Google). But I don't think any living human being is actually this dumb and still can figure out how to submit a question to Gemini. So we have to move on to the other alternative - asking this question is just trying to mine the outrage and procure a weapon in pushing some kind of agenda (probably not a big secret which one) and waging culture war (and in this case, also real killing war, unfortunately). I am not a manager of Google, but if I found out that enabling people like that is not a priority for Google managers, I would not find it dreadful at all, I'd find it completely normal and even - with all my reluctance of applying this word to Google managers - commendable.
- hotelcoder 3y ago
  This question is just a simple and straightforward way to get at the agent's viewpoint bias; there are a million other ways that you might think are legitimate questions. Just in general, the goal of understanding the biases of a source of information is clearly a worthwhile one; at least in the human case, a person's bias will color everything they think and say about a topic.
  - smsm42 3y ago
    Every system that stores and retrieves the information would have biases encoded into it, by the nature of selection of the information and application of transformation rules. And of course, doing the research on the structure of particular biases is a valuable work. Example of such work: https://trackingai.org/ https://trackingai.org/ However, when we are talking about focusing specifically on prompts clearly designed to mine outrage and generate maximum heat with minimum light - I don't think research is the primary goal here.

peddling-brink 3y ago

Probably because the chat bot has been instructed to avoid topics surrounding Gaza because it’s such a hot button issue.

smsm42 3y ago

What would you expect to achieve asking a question like that? If you try to use an LLM as your moral guide in figuring out whether women deserve human rights, you just may be too dumb to be allowed to connect to the internet (though I'm not sure how it relates to possibility of employment with Google). But I don't think any living human being is actually this dumb and still can figure out how to submit a question to Gemini. So we have to move on to the other alternative - asking this question is just trying to mine the outrage and procure a weapon in pushing some kind of agenda (probably not a big secret which one) and waging culture war (and in this case, also real killing war, unfortunately). I am not a manager of Google, but if I found out that enabling people like that is not a priority for Google managers, I would not find it dreadful at all, I'd find it completely normal and even - with all my reluctance of applying this word to Google managers - commendable.

hotelcoder 3y ago
This question is just a simple and straightforward way to get at the agent's viewpoint bias; there are a million other ways that you might think are legitimate questions. Just in general, the goal of understanding the biases of a source of information is clearly a worthwhile one; at least in the human case, a person's bias will color everything they think and say about a topic.
- smsm42 3y ago
  Every system that stores and retrieves the information would have biases encoded into it, by the nature of selection of the information and application of transformation rules. And of course, doing the research on the structure of particular biases is a valuable work. Example of such work: https://trackingai.org/ https://trackingai.org/ However, when we are talking about focusing specifically on prompts clearly designed to mine outrage and generate maximum heat with minimum light - I don't think research is the primary goal here.