Doc67 wrote: ↑Wed Feb 28, 2024 9:10 am When trawling the internet, how does it know which information it is considering using is factually correct and what is incorrect. Given the ocean of bullshit out there, can it fact check everything before using it?

What you put in is what you get out. If you train it on fake news that's what it will know.

But these language models that we're interacting with today have been trained in steps. You start with large datasets for language structure, general conversation etc. Then you use smaller high-quality datasets to make it more factual. In the end you have humans interacting with it, correcting it and rating answers etc.

While some can surf the web and process results, it will probably not "learn" from those results, but simply use them as context when generating a response.

Remember, it's all about predicting the next word, the word with the highest probability gets spit out, and then move on to the next until the sentence is deemed finished. It's not actually thinking, reasoning or having opinions about the future of AI.