|
|
|
|
|
|
by stubybubs
850 days ago
|
|
|
Everyone is up in arms about the Gemini image generator, forgetting that it is not a search engine. It is generative AI, creating something new and random, every time. It is generally an input to a creative endeavor, not basic research. Unfortunately if you really want to know what a Viking looked like you're going to have to do some actual research. It's wild to me that people expect something that can generate an image of a penguin roller-skating on a loaf of bread shredding on a guitar to generate a historically accurate image of a Viking or whatever. These two problems are very different. The text interface, also not a search engine. And it is definitely not a *truth engine*, which is what people like Musk expect Grok to be. It's not capable of that level of reason or understanding. Other commenters have brought up that asking it any type of question like is X worse than Hitler, whether that X is Biden or Trump or anybody else, it's going to respond "it's hard to say." It's being tuned to not pass value judgements at all, because that is not what LLMs are for. It's never going to be your truth engine. If you think these things are thinking, or that they have values, you've been duped. As a side note, I think all the big companies are making the same mistake of trying to get a generalized model for all queries, or with the layer that directs them to a specific tuned model. I want sharp detail and lack of hallucinations when it generates code. I want it loose when I'm generating creative text. But in no case to do I want it to talk to me about values. |
|
|
It might be creating things that are new but it does not create something random otherwise it would be useless.
> It's being tuned to not pass value judgements at all
This is a value judgement. Also it clearly has bias in where it chooses to pass value judgement and where it chooses not to.
The people that trained this thing trained it to prioritize DEI over accuracy and truth. That is their priority.