“As AI search takes off, users will need to get into the habit of fact-checking their search results, he said.“
Against what? That people are treating a search technology that invariably delivers falsehoods as a revolutionary, positive advance is just bananas.
What an amazing statement. There is no universe where it’s a good thing that a search engine company forces people to do a second search to confirm their first, company-provided, result isn’t a straight-up lie.
@bhaggart i guess its kind of like when Wikipedia became a thing. You have to verify anything you read on there because it could be made up
@ivalaine @bhaggart @ech No. AI “search” doesn’t provide citations, just a list of sources (if you’re lucky). That’s the whole problem. If you read a fact on Wikipedia, there will be a citation for that fact. That’s what the [citation needed] joke is all about.
Also, Wikipedia is designed to be accurate. That’s what the whole culture of the site works toward. There is NOTHING in these language models to check accuracy. They aren’t designed to be accurate. There’s no way to correct or improve the accuracy of their results. They’ll never get better, because they’re stochastic parrots.
@PeoriaBummer @ivalaine @bhaggart @ech the issue is advertising money. Chapeau to Jimmy Wales for resisting all these years. He could have been a billionaire, but the resource would be gone.
@howardgreen @ivalaine @bhaggart @ech It’s an issue, but I’m not sure it’s THE issue. There are for-profit companies (NY Times) that sell based on their reputation for accuracy. It’s not ideal, but it’s better than the LLM products.
These “AI” companies are just trying to make a quick buck. Pump the price based on hype. Hide from all the externalities. Dump the company on suckers using the promise of infinite scale. Move on to the next scam. This has been the SV business model for the past 15-20 years.
@ivalaine Sure, but Wikipedia has a committed and experienced team of editors who use manual and automated tools to make sure it's quite hard to introduce wide-scale misinformation. Not perfect, but generally excellent, especially considering it's a largely unpaid team.
How would you do that for *the whole of Google's corpus of knowledge*? Maybe impossible, certainly out of reach for the time being.
@bhaggart @ech point is you have to verify those sources are correct. Not really that different? I assume if you ask chatGPT also provides list of “sources”?