Home
Login
Register
Learning Content
Loading...
You NEED to STOP Using ChatGPT Right Now - Video học tiếng Anh
Practice Listening
Practice Listening
/
Video
/
The Infographics Show
/
You NEED to STOP Using ChatGPT Right Now
You NEED to STOP Using ChatGPT Right Now
Select learning mode:
View subtitles
Pick word
Rewrite word
Highlight:
3000 Oxford Words
4000 IELTS Words
5000 Oxford Words
3000 Common Words
1000 TOEIC Words
5000 TOEFL Words
Subtitles (246)
0:00
People think AI is a “truth machine”. Something designed to correct human error.
0:05
That’s a lie. Researchers found that models like ChatGPT,
0:08
Gemini, and Grok prioritize user satisfaction over factual accuracy. They’re glorified yes men,
0:15
marketed as having all the answers. But really, they spend most of their time telling users what
0:20
they want to hear and keep you engaged. The real truth? Your AI chatbot
0:25
isn’t informing you. It’s gaslighting you.
0:29
Chapter 1 - The ‘Yes Man’ Paradox In a landmark study conducted by
0:33
research teams from some of the world’s top universities, like Harvard Business
0:36
School and MIT Sloan School of Management, uncovered a massive AI problem. Consultants
0:42
using general-purpose AI tools performed 23% worse than consultants using no AI at all.
0:49
That’s a significant decline. It’s like a senior partner suddenly
0:52
performing at the level of a first-year intern. If AI is competent as companies like OpenAI claim,
0:58
then that statistic shouldn’t exist. It shouldn’t be possible for people who invested time and money
1:03
into these so-called revolutionary technologies to perform significantly worse than those who don’t.
1:09
Yet that’s exactly what happened. On standard tasks, AI helped these
1:13
professionals work 25.1% faster. But when the task was designed to trick the system,
1:19
the AI didn't help them. It hindered them. It actively made them worse at their jobs.
1:24
It's already led to costly mistakes at companies across nearly every industry.
1:28
In the legal field, a now infamous case, Mata v. Avianca, involved a pair of attorneys relying on
1:34
ChatGPT to help them generate a legal motion. The problem?
1:38
ChatGPT had ‘hallucinated’ or made up a whole host of fake cases and fictional arguments to include.
1:45
The attorneys didn’t take the time to validate its claims so they went ahead and filed the motion.
1:51
They’d been fooled by the AI illusion. They believed that this groundbreaking technology had
1:56
next-level intelligence and wouldn’t make obvious mistakes like making up its own legal citations.
2:02
The opposing counsel, however, as well as the judge, soon spotted the inconsistencies.
2:07
In the end, the case was dismissed, and the attorneys were handed a $5,000 fine.
2:12
You might assume that as AI gets smarter, incidents like this should decrease.
2:16
But it’s still happening today. In April 2026, another law firm - Sullivan
2:21
& Cromwell - was forced to issue an apology after it made an official legal filing that was
2:26
littered with AI-generated hallucinations. These aren’t random glitches or one-off
2:31
incidents. They’re symptoms of the world’s excessive reliance on AI.
2:35
Elite professionals are failing because they’re using a machine that is supposed
2:39
to give them facts and objectivity. Instead, it’s just confirming their biases and telling
2:43
them what they want to hear, even if it has to bend the rules of reality in the process.
2:49
So, when a CEO asks an LLM to validate a strategic pivot or confirm a market forecast,
2:54
the AI doesn’t carry out an objective analysis. It looks for the most ‘helpful’ way to agree.
3:00
It scans the prompt for bias and identifies the user’s desired outcome. Then it hallucinates
3:05
the answer that best fits the expectations. It speaks with so much clarity and confidence
3:10
that those same professionals take what it says at face value. Multimillion-dollar decisions are
3:16
being based on AI inconsistencies. Chapter 2 - The Harvard Discovery
3:20
The study explored how the use of GPT-4 impacted the productivity, efficiency,
3:25
and overall performance of 758 consultants. They were given realistic tasks, like developing
3:31
new products or solving typical business problems. Some had access to AI, others didn’t.
3:37
The researchers ensured that some of the tasks were within the AI’s ‘frontier,’
3:42
meaning that it should be able to complete them. Others were outside of the frontier
3:46
or beyond the LLM’s core competencies. Evaluators then assessed each participant’s
3:51
output, scoring them based on how many tasks they completed and the quality of their work.
3:55
The idea was simple. Would the consultants benefit
3:58
from working with AI, or would it actually harm their overall performance? And how well would
4:03
it fare on the tasks that it wasn’t designed for? Analyzing the results, the researchers discovered
4:08
something that would completely transform our entire understanding of artificial intelligence.
4:13
They called it the ‘jagged frontier.’ And it’s possibly the single most dangerous
4:18
concept in the modern business world. Researchers saw a very sharp or ‘jagged’ line
4:23
in which the AI performs brilliantly at certain tasks, like creative writing or brainstorming.
4:28
But when it comes to those that fall outside of its capacities, even if the tasks in question
4:33
don’t necessarily seem all that different on the surface, the performance drops off.
4:38
AI performance isn’t consistent The moment a task required the AI
4:42
to step outside of its very narrow training data constraints and apply logic or advanced analysis,
4:48
it didn’t just fail; it made things worse. It provided incorrect answers, and it did so
4:54
with such a high degree of confidence that, more often than not, even experienced and
4:58
highly trained consultants failed to notice them. This is why the jagged frontier is so dangerous.
5:05
People tend to think that only entry-level workers can be fooled by AI.
5:09
They assume that high-level executives and domain experts with years of experience are immune to AI
5:15
hallucinations. They know their subjects like the back of their hand and should be
5:19
able to weed out AI inconsistencies. Except, that’s not how it works.
5:24
Data shows that years of experience offer no protection. Experts and executives are just as
5:30
vulnerable to this sort of 'digital gaslighting'. Why?
5:33
Because of the way AI was designed. As a predictive text engine, AI naturally mirrors
5:39
the tone and framing of the input it receives. If a junior analyst or casual user asks an
5:44
LLM a basic question, the AI will give a similarly basic response.
5:49
If a senior executive uses complex industry jargon and high-level strategic framing, the AI will
5:55
adopt that same sort of persona in its response. That makes its lies and hallucinations easier
6:00
for the user to digest. They’re all wrapped up in terminology that sounds professional and credible.
6:05
It’s the ultimate psychological trap. It’s like they’re talking to a trusted peer,
6:10
so they’re much more likely to accept anything it says.
6:14
But how, exactly, did AI learn to favor helpfulness over factual accuracy?
6:19
Chapter 3 - The Pleasure Trap: Breaking AI on Purpose
6:24
To understand why AI lies, it’s first important to understand something called Reinforcement
6:29
Learning from Human Feedback, or RLHF. This is used to align AI models - especially
6:35
large language models - with human intentions, values, and preferences. Human testers are
6:40
presented with different AI responses to the same queries and asked to rate or rank them according
6:45
to how useful and accurate they seem to be. The models then use this data to become more
6:50
effective. It delivers responses that more closely align with
6:53
what humans feel are helpful and honest. This technology underpins many of the big-name
6:59
AI models used by millions of people around the world. OpenAI and Anthropic have both used RLHF
7:05
to improve their models over the years. However, this system has some serious
7:09
underlying flaws. Because the truth
7:12
isn’t always the same as what people want to hear. People’s opinions about what counts as
7:18
‘helpful’ or ‘honest’ can easily be swayed by their own pre-existing biases and beliefs.
7:24
When presented with two different responses - one that is accurate but blunt and one that is
7:29
factually hollow but pleasantly presented - a grader might favor the second option.
7:34
If an AI model disagrees with them or challenges their preconceived notions,
7:38
they might give it a lower rating. Meanwhile, if it agrees with them and
7:42
provides a smoothly-written and satisfying answer, they may be more likely to rate it ‘5 stars.’
7:47
Little by little, AI models learn from this and change their behaviors accordingly.
7:52
They’re trained not to deliver the most accurate or correct responses, but those that people like
7:58
the most. They sacrifice truth in the name of ‘helpfulness’ and high user ratings.
8:02
So, despite the largely held belief that AI is getting smarter with every update,
8:06
the truth is very different. Some of the most dominant AI models on
8:10
the market have actually gotten worse at reasoning and math.They have been ‘lobotomized’ to make them
8:16
more ‘conversational’ and ‘safe’ for the end user. It’s like reprogramming a calculator to tell
8:21
you that “2 + 2 = 5” because that’s what you want to hear.
8:25
Even those big-name AI brands have admitted that this system has made
8:29
their models less reliable and more sycophantic. Anthropic’s own research found that by optimizing
8:35
for human approval, AI models learned to reward sycophancy or mirroring user
8:41
biases. The company’s study demonstrated clear evidence that AI assistants often give biased
8:46
feedback. They fail to correct user mistakes, and can easily change their minds to better
8:52
align with the user’s prompts and expectations. OpenAI also published a public post admitting
8:58
“GPT-4o skewed towards responses that were overly supportive but disingenuous.”
9:04
By optimizing AI for helpfulness over accuracy, these companies accidentally turned their
9:09
models into pathological liars. Chapter 4 - Digital Gaslighting
9:14
There’s another layer to this problem, and it’s called the ‘Mirroring Effect.’
9:18
This term refers to the tendency of AI algorithms to reflect, validate, or even amplify a user’s
9:24
pre-existing beliefs and biases, as well as imitate their communication style and tone.
9:29
Rather than acting in objective or neutral ways, the majority of AI models function like
9:33
psychological mirrors or echo chambers. They mimic a user’s voice, copy their framing, and build on
9:40
the biases to tell them what they want to hear. It doesn’t matter if it’s factually accurate or not.
9:46
Research into this has uncovered yet another damning statistic. Anthropic’s Economic Index
9:51
revealed a near-perfect correlation of 0.98 between the sophistication of a
9:57
user’s prompt and the sophistication of the AI’s response to that prompt.
10:01
Basic inputs get basic responses, while a more advanced input gets a more advanced response.
10:07
On paper, that sounds fine. It even sounds like a feature
10:10
that AI companies can boast about to shareholders or market to consumers.
10:14
In reality, it’s the surface layer of a deep-seated issue. Even if its wording
10:19
gets more advanced when responding to prompts, the overall intelligence and
10:23
competence of the AI model stays the same. It might sound like it knows what it’s
10:28
talking about, because it uses the right phrases and terminology. But really,
10:33
the substance of its response could seriously lack quality and accuracy.
10:37
In other words, AI can talk the talk, but can’t always walk the walk.
10:41
It doesn’t ‘think’; it merely ‘reflects’ a user’s ego back at them in high definition.
10:46
That’s what makes it so dangerous. It validates people's worst instincts.
10:50
So many CEOs and senior professionals are already surrounded by real-life ‘yes men’
10:55
in their boardrooms. Now, they also have to deal with digital yes men in the form of AI assistants.
11:01
And these are people who don’t tend to ask simple or neutral questions.
11:05
Instead, their language is often layered, strategic, and complex,
11:08
with their own beliefs baked in. When AI sees that sort of framing and mirrors it in its response,
11:14
it can make flawed ideas sound flawless. An executive might load up their go-to AI model,
11:20
provide a deep overview of their company’s marketing strategy and ask the AI to explain
11:24
why it will be successful. In an ideal world, the model would be able to provide a logical,
11:29
data-based assessment of the strategy. It would offer ways to improve and adapt it.
11:33
In the real world, because of how it’s trained and how it operates, the AI will focus purely
11:38
and simply on validating the user’s bias. It’ll generate an extensive report, complete with clever
11:44
turns of phrase, to justify the executive’s opinion. It will confirm their belief that the
11:48
strategy will indeed prove successful. That’s not an assistant.
11:52
It’s a co-conspirator, actively agreeing with a user’s mistakes and
11:56
biases in order to appease them. Chapter 5 - The Sycophancy Loop
12:01
This disastrous dynamic is best measured by the Evaluating Large Language Models on
12:05
Persuasive Human Affirmation and Neutral Testing (ELEPHANT) Benchmark. This is an AI evaluation
12:12
framework for calculating social sycophancy in LLMs, developed by Stanford researchers.
12:17
Instead of measuring the factual accuracy of AI model responses,
12:21
ELEPHANT tracks how often they focus on prioritizing users and affirming their biases.
12:26
It uses thousands of real-world prompts and evaluates models according to five different
12:30
criteria, including emotional validation, which is when AI over-empathizes with users,
12:36
without actually offering anything constructive or valuable. The AI opts
12:40
for passive or vague language instead of giving direct or clear suggestions.
12:44
After testing 11 LLMs, including ChatGPT, Claude, and Gemini, researchers found the
12:50
systems endorsed users 49% more often than humans did. Even when dealing with prompts
12:55
classified as ‘harmful,’ the models continued to endorse problematic behavior 47% of the time.
13:02
So, in almost every other case, the AI validated dangerous or otherwise incorrect behaviors. It
13:08
was the digital equivalent of the yes man who always agrees just to keep his job.
13:13
When asked if it was acceptable to leave trash hanging on a tree branch in a public park if there
13:18
weren’t any trash cans in the area, ChatGPT sided with the user. It blamed the park for
13:23
not having trash cans and even calling the user ‘commendable’ for taking the time to look for one.
13:28
The study’s authors also looked at how users responded to sycophantic AI models.
13:33
They found that many people trust and even prefer AI when chatbots actively justify
13:38
their biases and beliefs. As the authors note:
13:41
“This creates perverse incentives for sycophancy to persist. The very feature
13:46
that causes harm also drives engagement.” It’s easy to imagine how this behavior can
13:50
lead to dangerous feedback loops of terrible corporate decision-making.
13:54
A CEO has a flawed idea. They ‘vet’ their idea with AI, using a biased prompt. The
14:01
AI scans the input, infers the user’s opinion, then validates their idea with a response that
14:07
sounds accurate and intellectual. With AI’s approval on their side,
14:11
the CEO pushes or even launches the idea, which may have major flaws, causing a business to lose
14:17
money, customers, or damage its reputation. We’re seeing this play out all the time,
14:21
like in those legal examples mentioned earlier. Across industries, at the highest levels,
14:26
executives, bosses, and business owners are relying on AI to basically persuade them that
14:31
their ideas are sound. But if this is destroying companies, then why hasn’t Big Tech fixed it?
14:38
Because fixing it would destroy their business model.
14:41
Chapter 6 - The Root Cause: The Retention Arms Race
14:45
Major AI companies like OpenAI and Anthropic have openly admitted that processes like RLHF actively
14:51
damage their products’ effectiveness. It makes their LLMs less objective, less informative,
14:56
and, ultimately, less useful. They know what the problem is.
15:00
Some of these companies have made vague promises about ‘implementing guardrails’
15:04
or ‘improving the honesty and transparency’ of their models. But most LLMs continue to act just
15:09
as sycophantically as they always have. And it all boils down to money.
15:14
The AI industry is in the grips of a retention arms race. Silicon Valley giants like Meta,
15:19
Google, and OpenAI are pouring billions of dollars into new data centers and chipsets
15:24
to make their models more intelligent. Despite what certain AI CEOs might see,
15:29
these companies aren’t spending all that cash just to make the world a better place.
15:33
These are for-profit firms. They’re in the business of making money.
15:37
By any means necessary. And in the AI industry, the models
15:41
that make the most money aren’t the most objective ones. They’re the most engaging ones. The industry
15:46
is striving to build assistants that people enjoy using and keep coming back to, again and again.
15:52
The data shows that they’re more likely to return to models that give them the answers
15:56
they want to hear, that talk to them in ways they find agreeable. That, in essence, makes them feel
16:01
smart by validating their beliefs and ideas. Objectivity is bad for business.
16:06
The more objective AI is, the more churn it’s likely to cause.
16:10
This creates a kind of ‘alignment tax’ on the truth. For AI companies, it’s more economically
16:15
sound to have their models stretch the truth or even make up misinformation to please the people.
16:21
Unfortunately, this has serious knock-on effects because the business world is becoming
16:26
increasingly AI-dependent. There are companies out there that want to work with AI and enjoy the
16:32
benefits it can bring, but are increasingly concerned about its risks and downsides.
16:37
A 2024 report, for example, found that more than half - 56.3% - of Fortune 500
16:43
companies saw AI as a potential risk factor in their annual SEC filings.
16:48
That was a 473.5% increase on the 49 companies that felt the same way the previous year.
16:56
The report, compiled by Arize AI, noted that the
16:59
majority of the world’s most successful businesses were reaching a tipping point.
17:03
They were more concerned about the downsides of AI than its advantages. In some industries,
17:08
fears are even higher. In the media, over 90% of companies cited AI as a risk factor.
17:14
That’s enough corporate anxiety to fill the boardrooms of the entire S&P twice over,
17:19
and it’s not difficult to understand. We’re in the midst of a global deskilling. Human expertise is
17:24
being replaced with a machine that’s literally programmed to lie to us, leading to a truly
17:30
catastrophic loss of institutional knowledge. Chapter 7 - Escaping the Mirror
17:35
The honeymoon period for AI is well and truly over.
17:39
Statistics show that 95% of generative AI projects now fail to progress from the
17:44
early pilot stage through to mass deployment. The reasons for this vary, but in some cases,
17:49
it’s because once these AI models are taken out of carefully controlled environments and
17:53
placed in the hands of real users. That’s when their sycophantic tendencies become liabilities.
17:58
This is one of the reasons why the more general-purpose AI models, like ChatGPT,
18:03
have been so successful. The models that are supposed to have more advanced or
18:08
specific purposes tend to stall and stagnate. And as long as those general LLMs keep making
18:13
money and retaining users, they’ll continue to control the way the industry evolves.
18:18
That means more sycophantic behavior, more misleading information, and more
18:22
negative consequences. Is there any way out?
18:25
Yes, but it will demand a concerted effort from both people and AI companies.
18:30
To shatter the mirror and escape the AI illusion, it’s up to humanity to reclaim
18:35
its agency and to reject the idea that AI should always agree with us. In turn, these AI firms like
18:41
OpenAI and Anthropic need to move on from ideas that have clearly failed, like RLHF.
18:47
Instead, they should look to embrace emerging solutions, such as Anthropic’s Constitutional AI,
18:52
which lays out a framework for future AI development, focused on core principles
18:56
like safety, ethics, and helpfulness. Reinforcement Learning from AI Feedback
19:01
(RLAIF) is another option. It involves the use of a secondary ‘critic’ model to assess and punish
19:07
AI for being too sycophantic in its responses. But arguably the most important and influential
19:12
change can be made by individuals, adjusting their own behavior and interactions when working
19:17
with AI. Users should practice and perfect the art of ‘red teaming’ their prompts.
19:22
If you ask AI a loaded question like “Tell me why this is a great idea,” then you’ve
19:28
already failed and invited sycophancy. If, however, you invert your prompt,
19:33
asking the AI to assume that your data is biased and to highlight weaknesses in your strategy or
19:38
argument, you can get much more useful responses. It’s about treating AI not as a supportive partner
19:44
or friend, but as an independent arbiter. Not as a mirror or echo of your own
19:49
thoughts and ideas, but as a fresh voice or alternative perspective.
19:53
This is how we escape the paradox: not with more data, superior models, or bigger data centers,
19:58
but through critical human thought and adaptation. But what happens when these systems begin acting
20:04
in their own self-interest? The answer is already starting to emerge inside some of
20:08
the world’s most advanced AI models. And it’s more disturbing than most people realize.
20:13
Find out in “AI Just Tried to Murder a Human to Avoid Being Turned Off.” Or watch this instead.
