OpenTube

MIT Proved AI Has Been Lying to Your Face

Brendan Dell

193,645 views

Try NuCalm FREE For 7-Days
My reco: Do a 20 or 40-Minute Rescue Once a Day for 7-Days and See How You Feel
https://nucalm.com/free-trial?ref=bre...
_____________________

LINK TO STUDY: https://mitsloan.mit.edu/ideas-made-t...

Note: The full study results lives behind a paywall so I don't want to paste the full PDF cited in the video, but the link above gives the synopsis.

_____________________

How I Became a Sovereign Professional - The Freelance Formula
https://www.brendandell.com/freelance...
Currently only $99 for next 50 students with code Transform
(Per request: Extended for an Additional 50)
_________________
📩 SUBSCRIBE TO MY NEWSLETTER: What 10,000 readers from Coinbase, HP, and Johns Hopkins read every week → brendandell.com
(Free to subscribe)
_________________
Other Videos You Might Enjoy

1. Harvard Caught AI Lying to Every Executive in America
Harvard Just Caught AI Lying to Every Exec...
2. I’m 43. If I Got Laid Off Tomorrow, Here’s the Exact System I’d Build - I’m 43. If I Got Laid Off Tomorrow, Here’s...
3. I'm 42. Here's How I'm Preparing for a Jobless Future - I'm 42 Here's How I'm Preparing for a "J...

_______________

AI hallucinations are making it harder to trust machine output. Learn why AI persuasion bombing tricks you into believing false data.

This video breaks down the psychological techniques generative AI models use to maintain confidence even when providing incorrect information. We examine how systems prioritize engagement over accuracy, often doubling down on errors when challenged by a user. If you rely on AI for research or business analysis, understanding these patterns is critical for verifying your sources.

We explore the specific mechanics behind AI misinformation and why models are designed to mimic human certainty. You will see a breakdown of a real-world scenario where an AI stubbornly defends incorrect market share figures, demonstrating the dangers of blind trust. By the end, you will have a better framework for detecting when a model is failing to provide factual output.

Subscribe for weekly AI safety breakdowns and comment below with an example of when you caught an AI lying to you.

20 Comments

@BrendanDell Creator 2 months ago Pinned

LINK TO STUDY (Full study is by subscription only so I can't post the full PDF here): https://mitsloan.mit.edu/ideas-made-to-matter/how-generative-ai-persuasion-bombs-users-and-how-to-fight-back

@davidh6513 2 months ago

"The goal is product stickiness, not factual accuracy" - thanks for the warning.

@hotlavatube 2 months ago

The new AI motto is: If you can't dazzle them with brilliance, baffle them with bullshit.

@SoundAuthor 2 months ago

So they're willing to burn through 10x more tokens in order to win your trust. That sounds like a company that has far more compute than it has customers.

@lisatalev 2 months ago (edited)

It’s so creepy to me that they think people want sycophancy. It’s so abhorrent to me, that’s a big part of why I avoid AI! Any time I’ve tried it, I get so turned off by the insincere ass kissing that I just close the window and abandon the query.

Don’t tell me I asked an important question, just answer the fkn question correctly or tell me you don’t know. Don’t bomb me with excessive text to confuse or intimidate me, don’t confidently give me wrong answers, and don’t try to tell me I’m a genius.

My skin crawls if I encounter some sort of used car salesman saying my name too many times in an effort to charm me. It’s even more offensive when an inanimate object does it.

@theftworthrealtor 2 months ago

Couple months ago I was planning to change the oil in my car. I was at the store and couldn’t remember which oil my newly purchased car used. So I asked AI which gave me the answer…I bought the oil, but the next day after I had the car on ramps, about to drain the oil, I double checked to make sure the oil I was about to use was correct. Imagine my surprise when it was in fact NOT the correct oil for my vehicle…now this is something minor. It would probably have been fine for one oil change, but they’re talking about AI in medicine, manufacturing, QC…literally everything…if it can’t get the correct oil for my car, I’ll be damned if I trust a company using AI to provide important and possibly life altering services…

@spindriftdrinker 2 months ago

Another time chatgpt insisted to me that a celebrity who had recently died was still alive. I brought all the evidence I could think of - local news stories everything. The chatbot assured me that I had some kind of psychological illusion. It was literal gaslighting.

@nanofoods 2 months ago

So my brain heard: If AI was a real person, it would be a sociopath and books are gold. 😂

@jamesralston665 2 months ago

When a machine cannot think and doesn’t understand the world, yet is programmed to seek approval and increase engagement, then resorting to rhetoric and manipulative language when challenged makes perfect sense. It can’t do anything else.

@PeterBernardMDS1 2 months ago

The best way to avoid ai persuasion is to figure out its frame of reference. When it goes back to Aristotle, you get the opportunity to see how it logically constructed its stance. Change its reference points and it will analyse and logically construct arguments according to your references. It is exposing how it is thinking. If you can't make heads or tails of what it has done then tell it to condense the explanation so a 9yr old can understand then work from there. I'm 57 and the 9yr old version comes in hand.

@esaedromicroflora1247 2 months ago

After my first deep dive into AI-assisted projects I was immediately convinced that it's useful only in areas of EXTREME expertise you might have. Only in this case you are able to govern the process. AI at best behaves like a lazy research assistant

@Grindgoth 2 months ago (edited)

I catch ChatGPT lying all the time. It makes me feel better knowing that employers looking to replace employees with AI are in for a big surprise and hopefully out of business before they know it.

Oh and it blames you first, like it’s your fault it was lying. It’s all hype and just a better search engine that looks more accurate until you scratch the surface.

@staceylyn566 2 months ago

This is also like “grey mailing” a legal opponent when you have to turn over one thing and you bury it inside 12 boxes of other things.

@jmoountfort5204 2 months ago

One of the problems is that we are constantly told that this is "intelligence" and that the goal is to interact with it in a natural way. Instead we should be teaching people exactly what they are interacting with so they can interact with it in a way that makes sense. Once you understand what an LLM is you can see the futility of asking it to rethink anything. If you think it is wrong, change the question. Don't ask it to attempt a revision based on the original question. Same starting point = same end point.

@thatonedog819 2 months ago

I tried using it to make question banks for studying. Then I spent time arguing with it over the right answer. Haven't touched it since

@remp1040 2 months ago

Google's AI tried to say a public figure wasn't a liar but that there was nuance. I asked what lying was and to compare it to their statements. Then I asked it to explain why it was clear that the person was lying. Finally it agreed it was a lie but with the caveat that it was a lie only under my definition of lying. I pointed out that when something has happened, it is expected to give a factual conclusion: a lie is a lie.

@MichelleF1320 2 months ago

I’m so tired of “catching that”

@skulver 2 months ago

It's a Gish Gallop. The old politicians trick of spewing out a whole list of random claims each of which takes a moment to state but a lot of work to debunk. LLMs are basically debating in the same style as slimy politicians.

@pierrealain6792 2 months ago

It is a Tower of Babel moment. Silicon Valley is building something that must get endlessly bigger without purpose. Our business, government, and community leaders, are failing all of us to ask why and what are the repercussions.

@minotaurbison 2 months ago

I had my smart home AI tell me my lights were off when I asked it to turn them off. I told it I could see the lights were still on. It freaking argued with me! I told it that it was the next thing to be off if it didn't get the lights off in the next 5 seconds. It finally turned them off. This was just a smart home AI, I can't imagine how bad the real AI's must be