This specific thing is a much more blatant class of error, and one that has been known to occur in several previous models because of DEI systems (e.g. in cases where prompts have been leaked), and has never been known to occur for any other reason. Yes, it's conceivable that Google's newer, beter-than-ever-before AI system somehow has a fundamental technical problem that coincidentally just happens to cause the same kind of bad output as previous hamfisted DEI systems, but come on, you don't really believe that. (Or if you do, how much do you want to bet? I would absolutely stake a significant proportion of my net worth - say, $20k - on this)
Google likely knew of this issue, but you need to understand that DEI-related missteps are judged a lot more harshly by the society than other types of errors. So, for Google this was likely “choosing the lesser evil” type of scenario. Anticipating edge cases for LLM behavior is very difficult, but it’s hard to imagine that no-one at Google tested vikings and 1940s Germans before the release.
“…over time, the model became way more cautious than we intended and refused to answer certain prompts entirely — wrongly interpreting some very anodyne prompts as sensitive.”
As predicted. Nobody did anything wrong. Google did great. Machines make mistakes. Trust the code.
I always give the user the benefit of any doubts when it comes to Google or any other company that uses "AI" support. I don't care if the user did it to themselves, a person at the company should be able and willing to fix it.
Isn’t the fact that Google considers this a bug evidence against exactly what you’re saying? If DEI was really the cause, and not a more broad concern about becoming the next Tay, they would’ve kept it as-is.
Weird refusals and paternalistic concerns about harm are not desirable behavior. You can consider it a bug, just like the ChatGPT decoding bug the other day.
> That or they’re convinced that they’re this close to fixing the automated system, which they obviously are not.
Knowing Google's engineering culture, you're probably spot-on. Ignoring long-tail events like this one is a common failure mode of this kind of relentless metrics-driven optimization (and they should know better).
That's still a pretty elementary error for a company that gets off on using CS-y riddles in interviews like Google used to. Move fast and break things, and then get a $5B fine.
Ah my apologies, that is ridiculous. I should have read in. I thought it was a case of them being told to go flood another channel.
Right now we don't really have any way to know: was this corrected? Maybe Google doesn't do this anymore! But we have no idea. It would be a more moral & just stance for Google to actually talk about this kind of stuff, but, like most companies, there is little communication or updates on how these things work. The triplines are all invisible, the effect happens at digital speed. I find that to be one of the worst things about where we are; it is a intense info-industrial mechanization.
I still don't see how except automation we expect to build affordable at-scale systems. But the AI deciding your are an enemy & flipping the bit suddenly, like it does, is reckless, cruel, & shoddy.
We trusted google at one point with a lot of our info, then they started to screw us.
Are people overreacting for something not enabled by default? Quite possibly, but literally today open ai is getting in trouble for almost certainly using Scarlett Johansson's voice, even after she specifically told them "no". They're already giving all the indications they don't care about consequences to abuse.
And the URL for the AI API shouldn't be buried in the advanced settings.
This lack of communication from the companies (Google in this case) is just insulting. They can afford to have humans hand these escalated issues.
And quite frankly, automated penalties (removal, ban, etc.) should be required to give specific reasons for the penalty.
If the user doesn't know what they've done wrong, chances are they'll do it wrong again unless they are told where they made a misstep.
And when the company makes a misstep and incorrectly penalizes a user, the company should admit its mistake and state how it will prevent making such a mistake again. If the company claims it is too complex, then they should not be using such a system.
Google (and other companies which use similar technologies) need to change their requirements for AI to never generate false positives.
I’ve read the post, its just how it is with their policies being evaluated by automations and not humans.
No, it's definitely a problem with the humans. It's just that the humans in question aren't customer support humans. It's Google's engineering humans that are to blame for shipping these poor quality experiences. They might be able to invert binary trees on whiteboards pretty well, but they suck at making an HTML page that explains why their system isn't working for you.
> This is an indication that Google underestimates the value of data stored by users.
That becomes apparent if you ever had any issue with a google product. There's no way to resolve issues outside of canned answers from "AI" systems and public forums.
Yes that is a problem. And the worst part is because the heuristics are a black box there are no developer controls to fix it yourself. You are the whim of Google’s ability to correctly interpret what it sees. Sometimes its wrong
The problem is that it hurts the perception of Google, particularly for enterprise customers. If the perception is that Google is unresponsive and gives customers no recourse to address obviously incorrect behavior of its automated systems, it makes it harder for enterprise customers to trust Google with their workloads.
A one off bug like this isn't the end of the world, but it fits into the larger pattern of how Google operates.
Do you have any sources that talk about these breaking issues? I wouldn't be entirely shocked that Google would be doing this, but I do want to see more than just one anecdote that it's ill intent rather than incompetence.
Is this attitude prevalent in Google: pure trust in automated systems? Engineers should know better than any that software is not perfect and it's insane to have blind trust in it.
reply