How are any investors cool with the ROI of AI in 2024?
Google scrambles to manually remove weird AI answers in search
361–370 of 387 posts
Re: Google scrambles to manually remove weird AI answers in search
#362This approach to remove bad search suggestions manually reminded of a different approach Google once took, where they weren’t satisfied with manually tweaking search results but rather wanted to tweak the algorithm that produces these results when there were bad results. 'Around 2002, a team was testing a subset of search limited to products, called Froogle. But one problem was so glaring that the team wasn't comfort…
Spend say £500M (USD/GBP/EUR) on experts, per annum. Imagine typing a search and getting a response: "Give us 30 mins to respond - here's a token, come back at 17:35 with your token" ... and then you get an answer from an expert, which also gets indexed. The clever bit decides when to defer to an expert instead of returning answers from the index. I'll leave the finer details out.
So, replace Google with calling the DMV?
Re: Google scrambles to manually remove weird AI answers in search
#363"Achieving the initial 80 percent is relatively straightforward since it involves approximating a large amount of human data, Marcus said, but the final 20 percent is extremely challenging. In fact, Marcus thinks that last 20 percent might be the hardest thing of all." 100% completely accurate is super-AI-complete. No human can meet that goal either. No, not even you, dear person reading this. You are wrong about som…
I feel like there's some semantic slippage around the meaning of the word "accuracy" here. I grant you, my print Encyclopedia Britannica is not 100% accurate. But the difference between it and a LLM is not just a matter of degree: there's a "chain of custody" to information that just isn't there with a LLM. Philosophers have a working definition of knowledge as being (at least†) "justified true belief." Even if a LLM…
+1, AIs don’t really “understand” anything, like human anatomy and deformity rates and societal norms when tasked with image generation, so you get hands with weird numbers of digits and other topological errors which even a very unintelligent human wouldn’t make. AI doesn’t understand and build knowledge in interconnected layers, it can’t “think” and link things back to first principles, it can’t really reason about things, and it’s not going to get significantly better until we start approaching it differently. This generation of AI might be useful for some things, but it’s being applied wayyyy too broadly too quickly and I see a big pullback coming.
“Expert systems” are not a new thing, and they’re still not all that useful except in some very small niches. Phone trees that can keyword match FAQs are useful for a lot of low-effort callers who put zero effort into solving their problem on their own first, but frustrating for callers who are only calling because there’s literally no other way for them to resolve their issue. Unfortunately for consumers, the cost is very low for businesses to make everyone wade through junky phone tree systems and penalize anyone who tries to mash zero to talk to a real person, even if that’s the only thing which will actually help them.
Re: Google scrambles to manually remove weird AI answers in search
#364What's wrong with Weird Al's answers?
Re: Google scrambles to manually remove weird AI answers in search
#365Am I right in thinking this means they are using RAG to inject this rather than directly training the model on the source?
Re: Google scrambles to manually remove weird AI answers in search
#366Re: Google scrambles to manually remove weird AI answers in search
#367Earlier quoted context omitted.
"users would pay someone else to do the search." My notion isn't a rehash of Google Answers. Google pays the "someone else", not you.
You assume that a modern day tech giant would hire an army of experts, instead of just outsourcing it to the lowest bidder in the current third world country of choice.
I live in hope.
I don't think that my notion is too daft - I generally find that quality might be a profit generator.
When you are so far up the arse of your financials that you can't see sense ... you are probably profitable for now until ... you are not.
Re: Google scrambles to manually remove weird AI answers in search
#368"Achieving the initial 80 percent is relatively straightforward since it involves approximating a large amount of human data, Marcus said, but the final 20 percent is extremely challenging. In fact, Marcus thinks that last 20 percent might be the hardest thing of all." 100% completely accurate is super-AI-complete. No human can meet that goal either. No, not even you, dear person reading this. You are wrong about som…
> So 100% accurate can't be the goal. Obviously the goal is to get the responses to be less obviously stupid. I'm not sure I agree. I think you're right that 100% accuracy is potentially unfeasable as a realistic aim, but I think the question is how accurate something needs to be in order to be a useful proposition for search. AI that's as knowledgable as I am is a good achievement and helpful for a lot of use cases,…
I should have been more clear. I am referring to Google's goals. Humanity as an abstract concept or you personally may have other goals, but, well, perhaps I am a cynic, but I think Google's goals are rather more monetary and less idealistic than they would represent. They don't want or need (as other replies correctly point out) the AI to always be correct and accurate. Along with general cynicism with regard to any conceivable AI's ability to do that, it is also fair to point out that the web itself doesn't have that ability either. We can't even find an objective yardstick to measure an AI with that way. Google's goal is to make the bad press go away so people use the AI more so that in the indefinite but ideally near future this AI can be monetized somehow to justify the interstellar valuations being ascribed to this technology in the "if it isn't happening in two fiscal quarters or less it might as well not exist" US/Western financial markets.
Re: Google scrambles to manually remove weird AI answers in search
#369Earlier quoted context omitted.
So you think. Seems like hubris to believe you're not though. I'm blind to what I'm blind to, and while I'd link to think I'm never wrong, the reality is that I often am. The biggest personal growth for me was in not needing to be right.
I disagree. Not only for myself but for the vast majority of human kind. LLMs are just a statistical model. It can claim that it’s normal for pigs to have wings and fly to the moon if it’s in the training data. No human, free of a mental/cognitive disorder, will be that wrong.
Why would pigs have wings and fly to the moon be in the training data as a data source marked as serious? We can no true scotsman both sides here. No true human, free of mental/cognitive disorder would be that wrong, but neither would an LLM, with properly annotated training data would be that wrong either.
Re: Google scrambles to manually remove weird AI answers in search
#370My whole qualm with this AI integration into search engines: it's a search engine, not a question engine. I go to google to search the internet for something, not ask it a question. IMO, asking AI for something is a different task than searching the internet. It's sorta the same problem as if I go into a store and ask an employee where something is, and they reply with "well what are you trying to do?"
Right, but people use search engines to search for answers, don't they? Aren't answers something?
I ask Google questions so that it gives me websites where there are answers to those questions.
What foods are high in vitamin C? How long is a flight from New York to LA? How large is the moon? What are the symptoms of COVID-19?