Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Google scrambles to manually remove weird AI answers in search

theverge.com

Post reply on HN
Print view View original

331–340 of 387 posts

Re: Google scrambles to manually remove weird AI answers in search

#331

Post by fragmede »

noncoml wrote: post #267
jerf wrote: post #27

"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…

“No, not even you, dear person reading this. You are wrong about some basic things too.” But even when I’m wrong I’m not 100% off. Not “to help with depression jump of a bridge” or “use glue to keep the cheese on the pizza” kind of wrong.

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.

Re: Google scrambles to manually remove weird AI answers in search

#332

Post by fragmede »

mvdtnz wrote: post #185
verisimi wrote: post #171

Earlier quoted context omitted.

100% correct , 80% correct lol. The thing is that truth/reality is not a thing that is resolvable. Not even the scientific method has this sort of expectation! You can imagine getting close to those percentages, with regards to consensus opinion. That's just a question of educating people to respond appropriately.

No. Whether a person should eat a certain number of small rocks each day is not a matter of opinion, it's not a deep philosophical problem and it's not a question whose truth is not resolveable. You should not be eating rocks.

Really? No more salt for you then. Good luck with dehydration, hyponatremia, cramps, and cardiovascular issues.

Re: Google scrambles to manually remove weird AI answers in search

#333

Post by Tiktaalik »

Feels like there's a market for a bunch of Googlers to go off and take what they know about how Google works and make a new, barebones search engine that is essentially Google circa 2015.

Before AI, before we had to append "reddit" to get useful human knowledge.

Re: Google scrambles to manually remove weird AI answers in search

#334

Post by helsinkiandrew »

eldaisfish wrote: post #187
helsinkiandrew wrote: post #68

Earlier quoted context omitted.

The problem is that for some searches and answers Reddit or other social media is fine.

No, it isn't. Humans interacting with human-generated text is generally fine. You cannot unleash a machine on the mountains of text stored on reddit and magically expect it to tell fact from fiction or sarcasm from bad intent.

> You cannot unleash a machine on the mountains of text stored on reddit and magically expect it to tell fact from fiction or sarcasm from bad intent

I didn't say you could, but that a machine can't decode the mountains of text doesn't mean that the answer isn't (perhaps only) on Reddit. I don't think people would be that interested in search engine that just serves content from books and academic papers.

Re: Google scrambles to manually remove weird AI answers in search

#335

Post by vanviegen »

wizzwizz4 wrote: post #173
ta8645 wrote: post #169

Earlier quoted context omitted.

So what? You have no way to know for sure if the human you ask the same question, does either. The question that started this thread was related to verifiability. And i still think it is a spurious complaint, given that we have exactly the same limitations when dealing with any human agent.

The problem of other minds is no reason to throw everything out the window. Humans are capable of being conscious of their reasoning processes; token-at-a-time predictive text models wired up as chatbots aren't capable of it. Your choice is between a possibly-mistaken, possibly-lying human, and a 100%-definitely incapable computer program. You don't know either "for sure", but you don't know that the external world e…

When you ask an LLM to carefully reason step by step before arriving at its answer, that seems pretty much the same as conscious reasoning to me. Of course, when asked to justify a gut reaction after the fact, it will just come up with something that sounds plausible (and may or may not be true). Just like humans do.

Re: Google scrambles to manually remove weird AI answers in search

#336

Post by fragmede »

skydhash wrote: post #201
bombela wrote: post #189

Earlier quoted context omitted.

I sometimes wants a search engine, sometimes a question engine. Likewise at the store. Why not have both with a way to choose which one I want on the moment?

> I sometimes wants a search engine, sometimes a question engine. If you want a search engine, it's easy to use the results as a feedback to refine the query. But a question (answer?) engine would need to be an expert in the subject. And not parroting stuff. That usually means curation. You need something to do the work ahead to filter the wheat from the shaft. I don't see how LLMs can do that. LLMs can't be a search…

> I don't see how LLMs can do that.

If it's in the training data, then it should be able to do that. That is to say, a comment's points matter. and the subreddit it's on. and who said it, and how the rest of their comments do/where they are. The LLM could annotate the unredacted reddit dataset with metadata as to where to rate it on the words used, the accuracy of the information, the sarcasm quotient, the hilarity quotient, how condescending the comment is; all of that an LLM could generate metadata about and feed into itself to get better and better.

Re: Google scrambles to manually remove weird AI answers in search

#337

Post by vidarh »

threeseed wrote: post #132
ta8645 wrote: post #127

Earlier quoted context omitted.

> LLMs so far seem to be entirely unverifiable. I don't understand this complaint. Are they any less verifiable than a human?

I can ask a human to explain the steps they took to answer a question. I can ask a human a question 100 times and I don't get back 100 different answers. None of those applies to an LLM.

And for most cases that human explanation is likely with a disturbing frequency a complete fabrication after the fact. See experiments on split brain patients.

With respect to repeatability, yes, LLMs are currently frozen in time. That is not an inherent limitation, but it is one that is practical for a lot uses and a problem for some.

Re: Google scrambles to manually remove weird AI answers in search

#338

Post by vidarh »

bluefirebrand wrote: post #167
ta8645 wrote: post #144

Earlier quoted context omitted.

You can ask an LLM to explain itself.. it will give you a logical stepwise progression from your question to its answer. It will often contain a mistake, but the same is true for a human. And if your LLM is giving you 100 different answers, then it has been configured to do so. Because instead, it could be configured to never vary at all. It could be 100% reproducible if so desired.

> it will give you a logical stepwise progression from your question to its answer. No, it will generate a new hallucination that might be a logical stepwise progression from the question you asked to the answer you gave, but it is not due to any actual internal reasoning being done by the LLM

We have no clear evidence the same isn't true for humans, and some that it might be. See experiments with split brain patients, that have shown the brain halves will readily explain how they made decisions they provably never made.

Re: Google scrambles to manually remove weird AI answers in search

#339

Post by vidarh »

Dylan16807 wrote: post #280
ta8645 wrote: post #169

Earlier quoted context omitted.

So what? You have no way to know for sure if the human you ask the same question, does either. The question that started this thread was related to verifiability. And i still think it is a spurious complaint, given that we have exactly the same limitations when dealing with any human agent.

> You have no way to know for sure if the human you ask the same question, does either. The human might lie, but they generally don't. An LLM is always confabulating when it explains how it reached a conclusion, because that information was discarded as soon as it picked a word. The limitations are not in the same ballpark.

We have no evidence that humans can even know, much less that they generally don't. And we do have evidence that there are situations where the brain will readily construct an explanation after the fact which can't possibly be true (experiments with split brain patients where researchers tricked one brain half into thinking the other half, and so the brain as a whole, had made a decision while the action was taken by the researchers, and made it explain how it has made the decision)

There is no basis for claiming to know that people don't usually make up explanations like this other than when e.g. breaking the process apart and writing it down step by step during. But even then individual decisions are "suspect".

Re: Google scrambles to manually remove weird AI answers in search

#340

Post by fragmede »

astrange wrote: post #229
ben_jones wrote: post #118

Earlier quoted context omitted.

Institutional investors panic > board panics > executives panic > evps panic and dictates incentives to ship AI > directors, ems, and below, who actually know how shit works, take a submissive role because they have mortgages in Mountain View to pay. That’s how it happens.

Not sure if anyone below director can afford a mortgage in Mountain View.

I know that's hyperbole, but here's something for $600k. Let's say it goes for $700k. 20% of 700 is $140k.

An L4 engineer at Google makes like $300k. round down to $200k post tax. live in shared housing for $2000/month; that's $24k/yr, add $2000/month living expenses on top of that, is another $24k. Stick $2000/month into 401k, Save the rest (200-24-24-24 ) 128k and you'll save up $140k in a bit over a year.

bam; owner, not renter.

Now, whether this is better than renting, and putting the money into the stock market is totally other question, but even L4's at Google can afford a mortgage in Mountain View.

https://www.zillow.com/homedetails/50-E-Middlefield-Rd-APT-4...

Post reply on HN