"Study reveals Google AI produces false information" Topic

link

Call me completely not surprised.

I could argue anything artificial is fake

I am shocked. SHOCKED!

Most people rely on free information, but what you get for free isn't really information. If you aren't paying for the product, you are the product.

You would think it would present 'average' or consensus information, whatever most sources agree on.

It has no way to tell what is true or false, does it?

Must be fake ai news

So Parz, it's saying no better than the MSM? 😱

But since it accesses streams of MSM for its information, should we expect better output? 🤔

In addition, google's AI is using coding, designed on the whole, by people with the like views of those who own Google. We now have a double liberal whammy. 😳

Wow! One might say since over 70% of sources of its information are probably biased at best and fake at worst, it's amazing it reports anything correctly.

One might say: "Garbage in, garbage out."😉

Bill, there are ways around the main coding protocols of Google AI and have it search outside the sources that it designers wanted it to use.

And yet, some people swear by it.

And those people are WRONG!!!!!

AI is just another tool, albeit one I don't trust nor know how to use properly. Some will use it, some will abuse it, some will be injured by it. As long as it makes money for Google, they'll call it a success.

Yet my college just stated that every professor MUST use AI in their classes. Unbelievable.

I wouldn't be surprised if there's a lot of organisations- government and private- taking the same stance, Grattan. AI has a lot of potential- especially if you want to reduce your workforce (and increase your profits thusly)- and people will want to use it. Even if the potential hasn't been realised as yet. "Practice makes perfect" applies to virtual tools as well as physical ones, too.

Here's a thought. Once they AI most of the workers to reduce workforce costs, who is going to have a job, and thus the money, to buy the things AI is making, inventing, advertising or distributing?

Will CEO, CFO and similar executive positions also be "AIed", one day?

I'm using AI occasionally to do some of the work here on TMP. This is tedious stuff for a human being, it would be a pain to write code to do it, but AI handles it easily.

It's a tool. Use it wisely.

I enjoy working with an AI, and I am MOSTLY satisfied with the results. There is skill involved, and you have to invest some time in refining how you use the tool. Use any powerful tool carelessly and it is dangerous. Duh.

I asked on two different occasions if Smurfs were Satanic.
I got ambiguous answers.

This is nothing new, usually at the end of a search it will state the the AI can make mistakes. You can tell the AI it has made a mistake, not looked close enough and is being biased and to check again. From what I read in the article is that the test was only done on searches which is only one of the uses of AI.

And yet, some people swear by it. Yes, they know how to get the right results and verify them. For what I've been doing it's been a big help. If you are having a problem get help.

If you ask AI what is the best war game (without additional criteria) it will come back and give you results based on awards the games have been given. Writing prompts and knowing how to get what you want is more art than science.

The key thing to understand if you are going to use it: AI on Google Search, including the Gemini family of models and AI Overviews, can provide inaccurate information. This is often called "hallucinations". These AI models predict the most likely next words in a sentence. They are designed for fluency and pattern recognition, not for verifying facts, understanding context, or real-world logic. Just like any other tool, you need to know how to use it and its limitations. It's not magic even if the hype claims it is.

Errors can occur for several reasons: "Hallucinations" (Pattern Matching over Truth): The AI predicts the most probable sequence of words to answer a query. This can lead to incorrect or fabricated information, such as nonexistent citations.

Poor Quality Training Data & "Data Voids": The AI is trained on large datasets from the web. These datasets can include misinformation, satire, and sarcasm. When reliable, high-quality information is scarce, the AI may fill the gap with inaccurate information.

Failure to Recognize Satire and Misinformation: The AI can struggle to differentiate between reliable sources, satirical sites, or content from forums. However, you can point it towards sites where you know info is valid and tell it to stay away from specific sites and sources.

Misinterpreting Queries and Context: The AI may struggle to understand the context of a query. It may misinterpret language nuances or treat user speculation as fact.

"Ungrounded" Responses: In many cases, AI Overviews are "ungrounded." The links provided by Google to back up the answer do not support the information in the summary. You can ask for the sources it has used to help get it correct (example: stay away from Wikipedia, left/right leaning sources, etc.)

Pressure to Compete: The rapid push to release generative AI tools has led to the release of models that can exhibit low accuracy rates. That's not going to change anytime soon.

Key Findings on Accuracy: Accuracy Rates Vary: Even with a 90% accuracy rate, millions of search results per hour could be wrong.

"Misinformation" vs. "Hallucination": Some mistakes are pure hallucinations, while others are due to the AI pulling in false or misleading information from the web.

How to Stay Safe: Double-check: Google states that "AI can make mistakes, so double-check responses".

Critically Evaluate: Users should not consider the first AI-generated summary as the absolute truth, especially regarding health, financial, or safety. Some AI programs will ask you if you require additional details or suggest the next step for what you want to accomplish.

Yes, this is mostly AI generated but I did check, edit and add.

Wolfhag

I find AI (very) slightly better than gooogle as a search engine, in that it doesn't bring up pages and pages of garbage/product links before you actually find what you're looking for (if you can.) For trying to find specific information, it's so-so. As to carrying on "conversations" with it, forget it. I've seen lots of posts where people did that; It gets lumped into TL;DR
Not really impressed with AI…

YMMV… ;-)

There are lots of different AI tools, they address different problems. People seem to think AI is synonymous with generative products (which, unsurprisingly , are specifically designed to generate made up stuff, cluevis in the name).

My daughter works in medical research and says the AI agents are essential to what they do. Finding cures for cancer mainly.

Zephyr, I have many conversations with ChatGPT and now also with GROK. They are often fun, and almost always informative. The more AI knows about you the better it can "converse.' I think the danger is how WELL the machine can relate. I do NOT want a 13 year old who is confused about who he is to "befriend" a robot. And if you ask the AI about that it will agree. The kid needs a human who cares.

Wolf, Doc, Martin, precisely. 👍