YouTube2Text

مبقتش حمل المواقع المعمولة بالذكاء الاصطناعي — Transcript

by غريب الشيخ || Ghareeb Elshaikh · 2,502 words · 367 segments · language en · Watch on YouTube

Full transcript

  1. 0:00Any website made with AI today, you
  2. 0:02don't need any effort to spot it. And
  3. 0:04know that it was made with AI. There
  4. 0:06are a few things if we put them
  5. 0:08together, you'll know on your own.
  6. 0:09Writing with a specific font. And that
  7. 0:11purple or eggplant color. Cards with
  8. 0:14rounded edges. Buttons that have a glow
  9. 0:17or blink on and off. Not to mention, of
  10. 0:19course, the icons next to the text. And
  11. 0:21the stars, rockets, and things like
  12. 0:23that. The truth is, the look isn't bad
  13. 0:25in itself. But our brains have been
  14. 0:27programmed, from seeing it everywhere,
  15. 0:29that it's something cheap and lazy.
  16. 0:31Without any effort at all, and because
  17. 0:33of its abundance, almost all websites
  18. 0:35made in the last two years look alike,
  19. 0:37especially dark mode. As someone who
  20. 0:39uses it on my laptop, I don't like it
  21. 0:41anymore. Because I see it on every site
  22. 0:43now, with or without reason. Alright.
  23. 0:45Does AI not know how to do anything
  24. 0:47else? Is this its limit, then? And why
  25. 0:48does it do that anyway? If it's
  26. 0:50mimicking existing websites, there are
  27. 0:53so many styles. And how do you
  28. 0:54practically overcome this problem?
  29. 0:56That's what we'll find out, God willing
  30. 0:57, in today's video. So let's get
  31. 0:59started. Peace be upon you and God's
  32. 1:00mercy and blessings. Welcome to a new
  33. 1:02video. The issue of all AI sites
  34. 1:05looking alike has a term that’s
  35. 1:06repeated a lot. It's the "Sea of
  36. 1:10Sameness." Sometimes they call it "AI
  37. 1:14Slop." Most likely, if you have an idea
  38. 1:18for a website now and open something
  39. 1:19like Claude Code to build it, or any
  40. 1:21other model, the first version it gives
  41. 1:23you will be in that style we talked
  42. 1:25about. And the truth is, this doesn't
  43. 1:27happen by coincidence. No, this has a
  44. 1:28reason. And the main culprit is
  45. 1:31something you'll find strange: a famous
  46. 1:33framework called Tailwind CSS. Let's
  47. 1:35see how Tailwind CSS did this to us.
  48. 1:39Tailwind is a framework for CSS, the
  49. 1:41language responsible for formatting,
  50. 1:43color, and shape of elements on
  51. 1:45websites. It has a library called
  52. 1:48Tailwind UI, consisting of a collection
  53. 1:50of buttons, cards, and forms that you
  54. 1:52can copy and use in your design. That
  55. 1:55was back in the day, when we used to go
  56. 1:57and get these things and use them,
  57. 1:58before AI ruined everything. Anyway,
  58. 2:00the guy who made the library, when he
  59. 2:02chose a default color for these
  60. 2:04elements he was displaying, chose the
  61. 2:06color called Indigo 500. It's a color
  62. 2:09between blue and purple, and he chose
  63. 2:11it. Maybe randomly, or he saw it as a
  64. 2:14neutral color. You can fit it with any
  65. 2:17design without feeling like it's out of
  66. 2:20context. And what happened is the
  67. 2:22snowball effect. Tailwind UI became one
  68. 2:25of the most famous front-end libraries
  69. 2:27in the world. And thousands of
  70. 2:29tutorials, templates, and open-source
  71. 2:32projects on GitHub used these
  72. 2:33components exactly as they were in the
  73. 2:36default setting. With the color purple
  74. 2:39—the buttons, the purple gradients,
  75. 2:41everything, it's all purple. With the
  76. 2:44rise of LLMs, which are AI models that
  77. 2:46are essentially the best imitators on
  78. 2:48Earth, the kings of patterns and
  79. 2:50repetitions. These models started
  80. 2:53training on all the material available
  81. 2:54on the entire internet. And the data
  82. 2:57and projects available online have this
  83. 2:59pattern repeating very often. Whenever
  84. 3:03it sees this look repeated in designs,
  85. 3:05it considers it the best choice, the
  86. 3:07one people love. So it recommends it,
  87. 3:10generates it, and leans toward it. Even
  88. 3:12the guy, Adam Wathan, the man who
  89. 3:14created Tailwind itself, wrote a tweet
  90. 3:16saying he wanted to officially
  91. 3:17apologize to everyone. For making all
  92. 3:20Tailwind UI buttons that specific color
  93. 3:22. And that what he did led to every
  94. 3:24interface generated by AI on the face
  95. 3:26of the. Earth becoming purple. Of
  96. 3:28course, he was joking, though. But it's
  97. 3:29true, you know. That is the reason. Of
  98. 3:31course, AI companies aren't just
  99. 3:33sitting around watching this happen.
  100. 3:34They certainly caught on to something
  101. 3:36like this. And they noticed the issue.
  102. 3:37And saw that it was repeating. So they
  103. 3:39started putting instructions in the
  104. 3:41models 'prompts to avoid the color
  105. 3:43purple. But unfortunately, that didn't
  106. 3:46solve the problem. Because we ended up
  107. 3:48with a new, also repetitive theme. The
  108. 3:50creamy background. The one with serif
  109. 3:53fonts. Big headings and italicized
  110. 3:55subheadings. And the accent color is
  111. 3:57that brick-like or orange shade. So we
  112. 3:59moved slightly away from the purple
  113. 4:01pattern. And entered another pattern.
  114. 4:03It will repeat in the same way dark
  115. 4:05mode and the purple pattern were
  116. 4:07repeated on websites before. And if you
  117. 4:10went and told the AI not to use the
  118. 4:12creamy color. It will indeed avoid it.
  119. 4:15But it will then move you to another
  120. 4:17repetitive color theme. So why are we
  121. 4:19constantly shifting from one pattern to
  122. 4:21another like this? The problem lies in
  123. 4:23the training method of the models
  124. 4:24themselves. These models are trained
  125. 4:27using something called RLHF. Which is
  126. 4:29known as Reinforcement Learning from
  127. 4:32Human Feedback. Meaning another model
  128. 4:34evaluates these answers and gives them
  129. 4:37a score. And the base model learns to
  130. 4:40produce the answers that get the
  131. 4:41highest evaluation scores. And the
  132. 4:45model that evaluates is trained on the
  133. 4:47average human opinion. So when it gets
  134. 4:50a strange, distinct, or different
  135. 4:52answer. It gives it a lower evaluation
  136. 4:55score. Not because it's bad, no. Just
  137. 4:58because it's anomalous, or outside the
  138. 5:00well-known and customary patterns. You
  139. 5:03could say that we are training the
  140. 5:05models to kill uniqueness. And if you
  141. 5:08apply this to design, a unique design,
  142. 5:11which we define as being different. Is
  143. 5:14statistically unusual and also not
  144. 5:17repetitive. So the AI model will not
  145. 5:19lean toward it. That's why any design
  146. 5:21you ask from AI will always end up in
  147. 5:24that "average" range. And it won't give
  148. 5:26you something "wow" on the first try.
  149. 5:28So back to the question, is this all AI
  150. 5:30can do? Doesn't it know how to do
  151. 5:31anything else? No, it knows, and it
  152. 5:34knows very well, too. The problem is
  153. 5:36that you aren't giving it a design
  154. 5:38direction to follow. So they will
  155. 5:40immediately go for the thing they know
  156. 5:42best, the eggplant color. Someone might
  157. 5:44tell me, so what? Even if it’s
  158. 5:45eggplant colored or brick colored, it
  159. 5:47makes no difference. It’s working and
  160. 5:49it’s correct. What more could I
  161. 5:50possibly need? And by the way, part of
  162. 5:51what you’re saying is very true. As
  163. 5:53long as it serves the purpose and does
  164. 5:54what you need, then it’s fine. Even
  165. 5:56if, by design standards, it looks bad.
  166. 5:58But the problem is, at certain stages,
  167. 6:01design isn’t just a look for the user
  168. 6:04to see. Design is sometimes the thing
  169. 6:07that reflects and speaks for your
  170. 6:09product or business. Even before the
  171. 6:12user thinks of anything else. It is the
  172. 6:14first thing their eye lands on. So when
  173. 6:16this look becomes accessible, so that
  174. 6:18anyone can do it. With the click of a
  175. 6:21button, it becomes repetitive and worn
  176. 6:23out, losing its ability to convey a
  177. 6:24message or say something about your
  178. 6:26product or brand because everything
  179. 6:28looks the same; a pharmacy site looks
  180. 6:30like a delivery site or a travel agency
  181. 6:32site, with no identity or clear
  182. 6:34branding. The part that bothers me even
  183. 6:37more is the unjustified, purposeless
  184. 6:39clutter. Try asking AI for a website
  185. 6:42without specifying anything. You will
  186. 6:44find every section has a heading. And
  187. 6:45under the heading, a subheading. And
  188. 6:47under them, three cards full of text.
  189. 6:49And it looks like you don't know where
  190. 6:51to go or where to start. So where do
  191. 6:52you begin your journey? If you are
  192. 6:54buying a product on the site. I mean,
  193. 6:55if you want to. Guide the user within
  194. 6:57this site to a specific thing, you
  195. 6:58won't be able to. And I am not telling
  196. 7:00you here that you must turn into a
  197. 7:01designer. I mean, there is a minimum
  198. 7:04you must know if you are building a
  199. 7:06product with AI. Like knowing what a
  200. 7:08visual hierarchy is. And that only one
  201. 7:11thing on the page should grab the
  202. 7:13user’s eye. And knowing what spacing
  203. 7:15between elements means. And knowing
  204. 7:17that choosing a font can change the
  205. 7:19entire feel of the website. Just
  206. 7:22knowing these things. The knowledge
  207. 7:24itself makes a huge difference. Because
  208. 7:26then you can tell the AI to move in
  209. 7:27that direction. Or go in that path or
  210. 7:29implement these things. And it will do
  211. 7:32it for you based on its vision. And
  212. 7:34here I must say that, undoubtedly,
  213. 7:36these tools opened the door for many
  214. 7:38people. Who found it difficult to do
  215. 7:40these things before AI. Especially if
  216. 7:41someone is making something simple like
  217. 7:43a landing page for their course. Or,
  218. 7:44for example, a teacher creating a page
  219. 7:47for their students to interact with. It
  220. 7:49saved them all the hassle that used to
  221. 7:51happen before. And certainly, not every
  222. 7:54site needs me to stop and look at its
  223. 7:56design and all that. I mean, for
  224. 7:58example, I have tools that I made and
  225. 8:00use. I don't like their design. But as
  226. 8:02long as their purpose is working, it
  227. 8:03makes no difference to me. But what we
  228. 8:06are talking about here is that if there
  229. 8:07is the ability to improve the design,
  230. 8:09why not? So the solution to the matter.
  231. 8:11Where does it lie? The solution is in
  232. 8:14something called productive friction.
  233. 8:16The idea is that you get back to having
  234. 8:18a hand in it, in some way, at certain
  235. 8:19stages. So you can produce something
  236. 8:22that looks humanized. Or make it look
  237. 8:24like a human created it. Or as we call
  238. 8:27it, "human-in-the-loop." First thing:
  239. 8:29don't tell the AI what not to do or
  240. 8:31what to avoid. Tell it what to do. This
  241. 8:34is actually the biggest mistake people
  242. 8:36make. When you tell it, for example, "
  243. 8:38don't use the color purple." Or "don't
  244. 8:41make it look like AI." You haven't told
  245. 8:43it what to do instead. So, it will
  246. 8:45default to the closest thing it knows.
  247. 8:48And it gives you the same problem, just
  248. 8:49in a different color. So, by...Telling
  249. 8:51it not to go in a certain direction.
  250. 8:53You aren't really guiding it. You're
  251. 8:54actually confusing it more. You have to
  252. 8:56give it a specific direction and style
  253. 8:58to follow. Also, avoid vague
  254. 9:00descriptions. Like, for instance, don't
  255. 9:02just tell it to make the site "modern"
  256. 9:04or "professional." None of that will
  257. 9:05solve the problem. It sees those words
  258. 9:07and descriptions you use, like "
  259. 9:09professional" and so on, as neutral
  260. 9:11terms. And "professional" to you is not
  261. 9:13the same as it is to the AI. So you
  262. 9:15must be specific in describing exactly
  263. 9:17what you want. Second thing: to create
  264. 9:19a good design, it starts before the
  265. 9:21coding and execution phase. Meaning you
  266. 9:25need a design system and colors that
  267. 9:28align with your brand. And that is very
  268. 9:30easy to do now with something like
  269. 9:32Claude Projects. You can go in and talk
  270. 9:35to it. Give it ideas. And it will build
  271. 9:37you a complete visual identity. Which
  272. 9:39you can then use later in the execution
  273. 9:42phase. And make it stick to, for
  274. 9:44example, this button style. And this
  275. 9:46menu style. And these fonts. These are
  276. 9:48very important things. Third thing: use
  277. 9:51design skills/prompts. There is a skill
  278. 9:53called "Front-end Design." This is an
  279. 9:55official skill from Anthropic found
  280. 9:57inside Claude. And it instructs the
  281. 9:59model. It tells it in its system
  282. 10:01instructions file. It tells it to
  283. 10:03imagine itself as a design manager at a
  284. 10:04famous small studio. And you are
  285. 10:07required to create a visual identity
  286. 10:09that can never be mistaken for someone
  287. 10:11else's. And the client coming to you
  288. 10:13has rejected previous proposals. They
  289. 10:15feel they were just templates. Or
  290. 10:17ready-made designs, and they are paying
  291. 10:20money now for a distinct perspective,
  292. 10:22so you want to take a creative risk and
  293. 10:24give this guy your best work, making it
  294. 10:26work in two stages: the first stage is
  295. 10:28to create a design plan before writing
  296. 10:30any code, including colors, fonts, and
  297. 10:32everything, then the second stage is to
  298. 10:34review this plan yourself, and if it
  299. 10:36finds any part that could be done for
  300. 10:38any other project and doesn't reflect
  301. 10:41the specific business we're talking
  302. 10:42about, it removes it. The best thing
  303. 10:46about this skill is that it warns it
  304. 10:48about styles that look too much like AI
  305. 10:50. Not 100%, of course. It tells it, for
  306. 10:53example, about three or four looks that
  307. 10:55clearly look like AI, which are. The
  308. 10:57cream background, the terracotta accent
  309. 10:59, the eggplant background, and the
  310. 11:01green and orange stuff. The second
  311. 11:04important skill is something called "UI
  312. 11:06UX Pro Max." This is an open-source
  313. 11:08skill. The idea behind it is that it
  314. 11:11gives it a database to search through.
  315. 11:14It contains 79 styles, 192 color
  316. 11:16palettes for 192 different products,
  317. 11:18and a wide combination of fonts. So,
  318. 11:21there's quite a lot of variety in there
  319. 11:23. The most important thing is that it
  320. 11:25changes depending on the field you are
  321. 11:26talking about. If you tell it, for
  322. 11:28example, that you want a spa website,
  323. 11:30it will give you a different system and
  324. 11:32visual identity. Than if you tell it
  325. 11:33you want a banking application. It
  326. 11:35gives you the patterns, style, and
  327. 11:36colors appropriate for that business.
  328. 11:38It also gives you a list of things you
  329. 11:40should not do in this field. And a
  330. 11:42checklist before. Delivery, which you
  331. 11:44use to verify things, such as
  332. 11:46responsiveness or how the site looks on
  333. 11:48mobile. But keep in mind that these
  334. 11:50skills are not a magic wand. You still
  335. 11:53need to combine these things together
  336. 11:55or assemble them. Meaning you should
  337. 11:58have a specific direction already, and
  338. 12:00a design system you created with cloud
  339. 12:02design, for instance, and use these
  340. 12:03skills to execute the design for you.
  341. 12:05It is a workflow or a sequence of steps
  342. 12:07that are carried out. It makes your
  343. 12:10website look significantly different
  344. 12:12from the websites that AI produces by
  345. 12:14default. One last thing is that when
  346. 12:17you want to build any website, you must
  347. 12:18prepare its content first. If you have
  348. 12:22already written all the site's content,
  349. 12:24it will reduce the gaps that AI sees
  350. 12:26and tries to fill with filler. The AI
  351. 12:30won't go and create three cards and
  352. 12:31stuff them with text by default unless
  353. 12:33there is no content on the site,
  354. 12:35because it is trying to help you. So
  355. 12:37when you have written all the sentences
  356. 12:39for it, the design will be built around
  357. 12:41that content and will truly serve the
  358. 12:43idea you want to present. If you do
  359. 12:46these few things, hopefully, your site
  360. 12:47will likely turn out different from the
  361. 12:49rest of the sites around it. I hope
  362. 12:51this video has been useful to you in
  363. 12:52one way or another. And don't forget to
  364. 12:54like at the end because it makes a big
  365. 12:56difference in the video's reach. See
  366. 12:58you in a new video, God willing, and
  367. 13:00stay safe. Yeah.

About this transcript

This page contains the full transcript of مبقتش حمل المواقع المعمولة بالذكاء الاصطناعي by غريب الشيخ || Ghareeb Elshaikh, generated from the public captions YouTube serves with the video. The transcript has 2,502 words across 367 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.