From glue-on-pizza recipes to recommending "blinker fluid," Google's AI sourcing needs work.
See full article...
See full article...
MVP is a four-letter word.The joys of modern software development- release now when it's functional, correct for errors later.
Lol this is comedy gold, thanks Google
I am of the opinion (especially in this situation as a former firefighter myself) that fire code or any safety-critical use-case is inappropriate for general AI searches/summaries. Possibly Retrieval Augmented Generation (RAG) built around your use-case domain would be accurate enough, but that’s not what you are likely using. Your result validation is good, but for every person who does that, there are 10 who don’t, or who aren’t as diligent as you are. So please be careful about how you present this idea to others in your industry so people don’t get killed because of misuse/misunderstanding of this technology.So far, I've found AI searches to be helpful, but only as a starting point. Myself: I use them for difficult to find fire code references in my job: What comes back from the AI, I'll then reference in CAN/ULC or NFPA to verify the accuracy - It still saves me a ton of time, but my own liability, as well as that of my employer, would skyrocket if I ever took the AI's response as the only one to put into my work for our customers.
And by then, it'll be too late... it'll be garbage all the way down.As more and more of the training data is fake posts and web pages generated by AI, this is going to get a lot funnier until it's not, and adults in charge wake up.
And I'm not convinced these so-called adults will wake up or that they even exist. There is just too much money to be made, consequences be damned.And by then, it'll be too late... it'll be garbage all the way down.
This might explain why Gemini is having problems with math'ing dates.Since a year (except for a leap year) has 365 days, three months would be (365 days/year) * (3 months) = 1095 days.
Are the people having unrealistic expectations of LLMs to which you refer actually CEOs of large tech companies?This is a great example of:
- garbage in, garbage out. Even the LLM says it's from a Reddit post.
- people having unrealistic expectations about LLMs. Perhaps this will convince everyone that they're parroting what they're fed and have no understanding or self consciousness.
- google shooting themselves in the foot. It's one thing to give a result like the Reddit suggesion as a link to the original post on Reddit. It's another one entirely to get it in this overview where it sounds like it's endorsed by Google.
The problem self-magnifies too. As AI discourages and demonetizes human content creators, the corpus of training data becomes more and more AI-generated as you said. The “internet grey goo” concept isn’t new, but I’d take it even a step further and compare it to genetic diversity. We all know what happens to a population that lacks an injection of new genetics…what happens when people stop making new original content and all that’s left is just one big pile of stagnant content statistically rearranged over and over?As more and more of the training data is fake posts and web pages generated by AI, this is going to get a lot funnier until it's not, and adults in charge wake up.
This is part of why AI is dangerous in my opinion. People currently say that it's fine so long as you verify the answer, but very few people will continue to do that long term if they don't spot immediate problems themselves.I am on the opinion (especially in this situation as a former firefighter myself) that fire code or any safety-critical use-case is inappropriate for general AI searches/summaries. Possibly Retrieval Augmented Generation (RAG) built around your use-case domain would be accurate enough, but that’s not what you are likely using. Your result validation is good, but for every person who does that, there are 10 who don’t, or who aren’t as diligent as you are. So please be careful about how you present this idea to others in your industry so people don’t get killed because of misuse/misunderstanding of this technology.
Former firefighter, as well I use it solely as a starting point. I.e. "What NFPA code references lumens per square foot for strobes in an office setting" to get me to the right NFPA code and section - I've heard enough horror stories about AI generated content that precisely zero of it is used in anything given to the client.I am on the opinion (especially in this situation as a former firefighter myself) that fire code or any safety-critical use-case is inappropriate for general AI searches/summaries. Possibly Retrieval Augmented Generation (RAG) built around your use-case domain would be accurate enough, but that’s not what you are likely using. Your result validation is good, but for every person who does that, there are 10 who don’t, or who aren’t as diligent as you are. So please be careful about how you present this idea to others in your industry so people don’t get killed because of misuse/misunderstanding of this technology.
Given that they're using reddit as a source, I'm surprised that we didn't get a bunch of dick jokes in with the blinker fluid suggestions.I am not surprised considering LLM don't actually know anything. As for training with reddit and other social media...garbage in garbage out still holds true it seems.
We are nearly there....the only content left is just one big pile of stagnant content statistically rearranged over and over?
As I've basically just mentioned in another comment: I agree, but I don't think the usage you describe is really possible for most people, and that's the problem.Former firefighter, as well I use it solely as a starting point. I.e. "What NFPA code references lumens per square foot for strobes in an office setting" to get me to the right NFPA code and section - I've heard enough horror stories about AI generated content that precisely zero of it is used in anything given to the client.
I am of the opinion that, right now: AI is handy for speeding up the start-up portion of the process, but at least in my line of work: Then end result must still be 100% human generated - and verified.