This content explains how AI detectors work by analyzing specific writing patterns and provides practical strategies for humanizing AI-generated text to avoid detection, emphasizing that simple word replacement is insufficient.
Mind Map
Nhấn để mở rộng
Nhấn để khám phá sơ đồ tư duy tương tác đầy đủ
So by the end of this video, you'll have
full understanding of how AI detectors
work. So what exact elements of our
writing these AI detectors pay attention
to. So if you've been using Chajbd or
some other AI tool to support your
writing or to write some paragraphs for
you now, you don't want to be flagged
for uh AI generated content, you want to
know exactly how to redo these
paragraphs. And let's be honest, you
don't want to be using these AI
humanizer tools because they are
completely useless. they will uh heavily
affect the quality of your writing. So
you have to know how to work on this
yourself and how to restructure and then
I'll show you an example of AI generated
paragraph and I'll work on it and try to
read it so that it doesn't get flagged
for AI generated content. So first let's
quickly go through uh the different
factors or different elements of your
writing that uh AI detectors uh pay
attention to. So it's not just about the
words as some people think. It's not
just about these favorite words that AI
uses. It's about so much more than that.
So the first metric and these two are
the most commonly cited statistical
metrics or text perplexity and
burstiness. So basically perplexity is a
measure of how surprised a language
model is by our text because human
writing tends to have a higher
perplexity because it's more
unpredictable. AI content on the other
hand is more statistically uh expected
and therefore has less perplexity and
burstiness is pretty related is is uh
about variation. So so human writing
again it it has variation in sentence
length structure word choice while uh AI
tends to be just uniform. Then sentence
structure and grammar. So consistency
and polish these are things that are
taken into account uh because AI
generated text is usually grammatically
flawless uh and stylistically
consistent. Quite often it's just too
consistent. Then AI does have its
favorite words like I said it may repeat
words, phrases or sentences and it just
does it uh too often. And another one is
very subtle. It's about transitions.
It's good to use these expressions, but
AI likes to overuse these expressions,
these connectors like therefore,
however, in conclusion. Again, this
overlaps with the the favorite
expressions that it just like to to
overuse. Vocabulary and word choice very
related again overlaps. It's about
favorite expressions. At the same time,
it's using safe vocabulary. So, very
rarely does it use any colloquial words.
Um, for our academic writing is not a
big problem. So here I just chose
whatever is relevant for academic
writing of course and then when we talk
about fixing things again I'm talking
about fixing for academic writing
because you can see some guidelines
online for how to fix how to humanize
content and we will talk about things
like adding humor personal anecdotes
making it inconsistent and open-ended
but it's usually talking about our
everyday language. So maybe if you're
writing a blog here, we're talking about
academic text of course. So it so the
bar is much higher. But about the
vocabulary and AI, it's learned to
predict what comes next. So again, it
takes this most predictable path
essentially when writing our content for
us and then later gets our content gets
flagged by AI for the exact same reason.
Terms of coherence and depth of AI
writing. Again, it's not always easy to
spot, but it does focus on the surface
uh level meaning quite often lacks
originality argumentation. And then it
uses the statements that do seem fine
for academic writing, but again it
really depends on the context of where
they appear. So they appear meaningful,
but they are not always meaningful. So
like it's important to consider all
perspectives. It can be a meaningful
statement or it can be just there for
the sake of it. And this often the
latter often happens in AI writing. And
then finally it may compare our writing
to know other known outputs. So their
language corpora and uh both for AI
language and human language and some of
these detectors they make use of this uh
huge database of language of any
possible use of language that there
exists. So so again that's another
factor. So you can see that there are
more factors than just the words that AI
uses. This is what people tend to think.
So they think that they will just
replace these favorite expressions of uh
AI or CHBD and the text will become
undetectable and then they're surprised
that it's still uh very much detectable.
So it's there is so much more. It's
mainly if I had to pick one, it is
really about the structure. You have to
introduce variation in the structure.
We'll go through it. So now I'll go
through some ideas for how to humanize
that text. and I I'll continue to talk
about some issues with AI uh created or
generated content and then I'll show you
an example of AI generated content how
it's flagged 100% AI generated and then
how I'll work on some uh some changes
and to make it less or to make it
undetectable. So the first way to fix AI
generated content and humanize it is to
introduce intellectual hesitation which
is a normal thing that is expected in
academic writing anyway. So basically AI
likes to write in absolutes in uh
present information as factual
statements as something that's just sure
and scholars on the other hand or humans
for that matter they hedge they suggest
they suspect. So so it's about using
this kind of language which again in any
case is something that you have to do in
academic writing. So you don't like you
don't want to be presenting things as
facts. Instead you're talking about uh
there uh being a possibility that
something happens. it appears that
something is true. Uh it is suspected,
it is believed, it is likely. So this
kind of thing, this is something that AI
rarely rarely does. So it will say
social media has an impact on uh
whatever self-esteem and it's not
something that we like to do like I said
in any case, but especially if we want
to avoid being flagged for AI content.
Another one is to add subtle critique.
Critique critique critique
critique critique. Jesus. A subtle crit
critique critique critique
critique
pronunciation critique. Another one is
to add subtle critique. And I'm laughing
because I just spent a little bit of
time trying to trying to learn how to
pronounce the word the word critique.
This is again pretty obvious because in
academic writing you do have to present
different uh points of view. Anyway, so
AI will present things uh it will just
factually present a statement that
somebody does or says something whereas
as a human we want to point out maybe
some inconsistencies or point to some
other points of view. It's not something
you can do to every sentence. So we'll
go through every sentence in a second
but it's something that generally should
be present. And now this one is very
important. This one will be uh important
at the level of sentences and and just
has to be uh present everywhere as we
make our way through this AI generated
context. So to introduce uh variation in
sentence opening uh openings and syntax.
So AI will often start sentences in the
exact same way even if we don't
initially notice it. So things like this
study this shows it is important. So
there are these little repetitions and
again they are very uh to our basically
human eye it's it's very difficult to
spot but they are almost exact uh in
their structure and their length which
then adds to the rest of the structure
of the sentence and overall the whole
sentence tends to have the same kind of
number of beats same number of sometimes
even similar number of syllables words
length and so on and so forth. So by
adding this this variation and how we
introduce our sentences we are breaking
down that structure already. So we can
use introductory clauses, dependent
clauses, inverted structures rather than
just uh repeating that same exact
structure of a sentence. So things like
although widely cited, this study has
received little empirical follow-up.
It's already a little bit different or
central to this argument is the notion
of so just uh reversing the order.
You'll see when I go through individual
uh sentences what I mean just reversing
the order and introducing this
variation. So now you know what factors
are being considered by AI detectors.
how you know roughly in theory how to
humanize the content. And now let's have
a look at how we can actually humanize
the content. So let's have a look at
exact example sentence generated by AI
and see how we can change it and
humanize it so it's undetectable. But
before I continue, have a look at my
website and the range of services that I
offer that are relevant to any aspect
and any stage of your study. They
include Zoom tutorials, data analysis
services or writing support. Okay. So,
what we can see here now is a paragraph
that I asked Chad GBT to produce. If the
sound quality is slightly different,
it's just because I switched computers.
Hopefully, it's not too bad. But this is
the paragraph that we're going to to
work on. Now, just to show you that it
it is being flagged for AI generated
content. Let me paste it here to this
free tool, detect AI. Uh, sure enough,
100% generated by AI. So, we'll have to
work on it a little bit. So I pasted it
to Microsoft Word and let uh let us
begin the work. So the first sentence,
self-esteem plays a critical role in
shaping a communicative experience of
migrants using English second language.
Firstly, if I liked it, I could just
keep it. Uh as I said, what is really
more important are not individual
sentences, just how they are glued
together. So uh occasionally, if I
really like something, I can change it.
uh as long as I make sure that the
following sentence does not follow the
exact same pattern like I said the
number of beats and and uh generally the
length. So uh here I will uh tone it
down a little bit. So I'll say as I said
uh I don't like and nobody does in
academia to present things as factual
statements. So rather than it plays a
critical role, I'll say it uh can play a
critical role. And at the end, we'll
just add the reference because again,
it's something that normally would be
expected to be referenced anyway. So,
let me just say it's some somebody
called Jackson 2020. And this already
kind of breaks down the sentence a
little bit. And the second sentence is
high self-esteem fosters confidence,
which is essential for engaging in
conversations, expressing needs, and
participating social, educational, and
professional context.
I don't like some of it. For example,
participating in social educational
professional context seems a little bit
vague. So, so this is another in my
opinion a little bit of a giveaway. Not
so much, but it's just something I don't
like. In any case, uh it's talking about
uh confidence, which is essential for
engaging conversations. So, we're
talking about conversations, which is
fine. So, why then talk about social and
participating in social education
professional context? So I feel like we
just have to simplify it. So that's
generally my approach. I have to
understand the sentence first and then
think of how I can possibly uh talk
about it in a way that I just like more
than what it says now. And usually this
includes deleting some things that are
just not needed. Remember what I told
you about surface level meaning and
things it says that kind of make sense
but then they don't really. So I'll
start by again toning it down. I'll say
it um has also been shown that high
self-esteem can increase so so again
toning down can increase uh
uh
confidence which is a chance for
engaging blah blah blah and then
participating there is just too much
information can increase
confidence and as a result um their
willingness or migrants doesn't matter
their willing ness to to
participate in various
various
conversational contexts exchanges. It
doesn't matter. It's not perfect either,
but basically the the
the message is the same. Hopefully, you
agree there's too much about the
conversation and expressing needs. It's
just about confidence and being willing
to participate, to engage in
conversation. So, now let's continue.
Again, it's not a bad word and I use it
conversely. I just feel like here I'll
include some variation again to to
influence the overall length even of the
sentence. I'll say low self-esteem I'll
also uh reverse the order as I said
earlier. So low low self-esteem on the
other hand and there is a lot to unpack
here again. So it may hinder
communication but by increasing anxiety,
fear of judgment and reluctance to speak
which doesn't make that much sense
because again it's uh or maybe it makes
some sense anyway. Ultimately affecting
language development and social
integration which is again a little bit
about communication. It's just about
simplifying. I want to simplify this. Uh
so low self-esteem on the other hand may
affect social integration mainly due to
increased anxiety and fear of judgment.
It doesn't have to be this way. Maybe I
could keep u some elements but I feel
like again it communicates the point
that I was trying to make. Migrants who
perceive their English skills as
inadequate may may internalize feelings
of inferiority reinforcing inclusion and margin
margin
marginalization whatever. Um let's see
what we can do here. The sentence is
good. I like it. I would normally like
it. But then remember and this is
probably the what will cause us problem.
There is another sentence that's again
very similar in length in structure even
though it's a different sentence. So
then it says therefore supporting
self-esteem is vital not only for
linguistic competence but also for
promoting migrants overall well-being
and inclusion almost every time it lists
things is almost every time is two
things uh divided by and so it's just too
too
similar. Uh the previous sentence was
also about something and something and
then there is this almost exact same
structure. So I want to to avoid that. I
want to again get the overall sense of
what it means and combine it maybe into
now something that's a little bit longer
maybe slightly uh clumsy even so um as
and also rever reverse the order so as
low self-perceived or selfassessed we
could say English skills skills are
believed remember about toning down
uh to lead to inferior iority complex to
lead to
inferiority complex and exclusion
marginalization is I mean arguably
slightly different but for now it will
do an exclusion so as it's believed to
lead to these these things is vital it
says we have to do it so it is crucial
to support
self esteem building to promote the
migrants linguistic
and social inclusion. Uh because that's
what it says, promoting their well-being
and inclusion and the language
linguistic competence. So, this should
do. Hopefully, it's not too uh short
because I think it has to be 80 words.
It Yeah, that's enough uh for this uh
for this tool that we were using. So now
this
0%. It did take a little bit of of work
and I cut out some sequences when I was
thinking. Uh but that's basically what
happens. That's what happens. So I think
my main message to you would
be read it. Try to understand what it
says and try to say it in your own
words. In fact, what I would normally
do, I would probably have it as separate
paragraphs. I try to literally forget
what it said exactly. I just want to
remember the meaning and then say it in
my own words, but it's very hard because
you're influenced by what you just saw.
And it does have good sentences, but
basically that's what happens. I hope
that you enjoyed the video and learned
something new today. If you did, please
like the video. This really helps
YouTube position my video and lets
YouTube know that it's a good video with
quality content. Uh, feel free to ask me
questions in the comments. Feel free to
share it with others and consider
Nhấn vào bất kỳ đoạn văn bản hoặc mốc thời gian nào để nhảy đến phần đó trong video
Chia sẻ:
Hầu hết transcript sẵn sàng trong dưới 5 giây
Sao Chép 1 Chạm125+ Ngôn ngữTìm kiếm nội dungNhảy đến mốc thời gian
Dán URL YouTube
Nhập link bất kỳ video YouTube để lấy toàn bộ transcript
Form Trích Xuất Transcript
Hầu hết transcript sẵn sàng trong dưới 5 giây
Cài Tiện Ích Chrome Của Chúng Tôi
Lấy transcript ngay mà không cần rời khỏi YouTube. Cài tiện ích Chrome để truy cập transcript của bất kỳ video nào ngay trên trang xem, chỉ với một cú nhấp.
