AI-Based New Generation Working Culture

How generative AI tools redefine modern team productivity, communication paradigms, and operational agility.

Speakers: Yiğit Konur

22 Nov 2023

Video Transcript

⏱ 34 min read• 6,006 words

Read the full speech or switch languages

Let me share my story. About 3 or 3.5 years ago, I found OpenAI on Hacker News. I told Burak. "Let's build something with this." We thought it would be great for content. We played with the Playground but couldn't do much. We thought it wasn't a big deal. We focused on updating macOS instead. Everything changed after that. Later, I was chatting with a friend on Discord. He runs a startup for form backends. People send their forms to him. He had a problem. He was trying to filter spam. It’s a tough problem. Now, it's super easy. He started a Discord stream. He pasted an email and said, "Score these emails 1 to 100 for spam."

He hit enter and got a score of 80. I looked at it and agreed. It was definitely spam. I was blown away. My life hasn't been the same since. We've been building a lot around this at Woop. I have tons of automations on my computer now. I’m hooked. The world got hooked when they launched the chat interface. Sam Altman said they expected someone else to build ChatGPT. They offered GPT-3 cheaply, but nobody did it, so they did. It hit 1 million users in 5 days. That’s 15 times faster than Instagram. It kept that pace to reach 100 million users. That is no small feat. It’s not just for fun like Instagram.

You can’t take photos or flex your status. It doesn’t hit those primal instincts. Still, it’s fascinating. Since then, everything AI related has blown up. Take Auto-GPT. It’s a major AI project on GitHub. It hit 100,000 stars faster than almost anything in history. We have a slight echo here. Tons of projects followed. Everyone is focused on AI now. If your company isn't talking about AI, you’re at the wrong place. I’m being bold. So, what’s the deal, Yiğit? Is it just ChatGPT? What are we doing? Just rewriting or summarizing stuff? Why is the hype so real? The energy has dipped a bit lately.

It’s not the main topic anymore. Some fanatics like me are obsessed. Others gave up. Some don't care at all. Some just use it for quick tasks. But it’s deeper than we think. I found a dystopian example for you. A paper on Generative Agents came out last year. Stanford led the research. If you’ve played Pokémon Red, you’ll recognize this interface. It’s just a game. You might ask, what’s the point of this? Stanford created 25 unique characters. One is a painter, one is an engineer. They have distinct personas like you. They placed them in different environments with a sense of time.

All 25 have an LLM, like OpenAI. They have a reasoning engine. They have a logic flow. They shower, brush their teeth, and make breakfast. Their day starts just like ours. They’re living in a real universe. I’ll explain why they ran this experiment later. I printed this paper and read it. I was so hyped. These 25 characters are all unique individuals. They move around during the day. They make their own decisions. Don't let the basic graphics fool you. The backend is solid. There’s a fridge, for example.

The fridge can run empty. Someone has to go shopping and restock it. They chat clearly. They gossip and talk behind each other's backs. Like, "What do you think of Yiğit?" They might say, "Yiğit is just talking nonsense." They form negative opinions about him. It changes how they view him. They call this reflection. It has its own specific prompting pattern. By the end of the day, there’s so much action. There are choices, parties, and family ties. They even discuss their work with each other.

You can guess how timeless this is. It doesn't run on real-time. You can simulate 10 or 100 years in a single day. Just one click. This world doesn't follow our timeline. It moves as fast as your compute power allows. Isabella is throwing a party here. People RSVP, but some flake out. They’re building a truly realistic universe. Thinking about this gets pretty wild. It’s actually unsettling. Don't dig too deep. You’ll lose your mind. It’s all just cloud computing. Stanford ran this experiment to create social simulations. Think about LLM and ChatGPT. Look at what people are building with them. The goal is to test new governance models.

Communism never worked in the real world. Maybe it would have worked under different conditions. People always claim, "It would work if..." Now we can actually test that. We have a world and simulated humans with emotions. These humans interact and make decisions. They gossip, fight, fall in love, and get jobs. They earn money, get fired, and find new work. Just like us. They’re doing this entirely with LLMs. Think about that. It means simulating human outcomes infinitely. It’s a massive power. That’s why LLM is such a big deal.

We need to take this seriously. My marketing peers just want to write ads or summaries. No, man. This tech is coming for our jobs fast. If you don't internalize this, you're in trouble. I actually feed off the negativity. A little stress is good for you. I analyzed seven McKinsey papers on this before this talk. Most of us are in marketing. They analyzed the commercial impact of various fields. It hits 500 billion dollars. They checked how much AI can solve. Marketing is right at the top.

What do we do as marketers? Some handle search engines or ad platforms. Others do A/B testing. Some crunch Web Analytics data to make decisions. Are these things LLMs can't do? LLM stands for Large Language Model. It’s the core tech behind ChatGPT. It can definitely do them. It’s unlikely they can’t handle our simplified, machine-ready tasks. This won't happen overnight. Things won't just stop instantly. But trust me, it’s a slow fade. It’s already started in some fields. Look at the translation industry. Don't you think it’s been shrinking? Google Translate had a massive jump.

Some of you remember that. Google Translate took off overnight. The old-timers remember. If you're over 30, you know. What happened 6-7 years ago? The Transformer paper was published and implemented. That was the turning point. Maybe there was another context. Translation agencies started doing post-editing. People who used to translate from scratch now just edit machine output. They stopped translating sentence by sentence. Their entire routine changed. They had to adapt. Now, if someone says. "I won't sacrifice my pride as a translator," they are free to do that. They can translate sentence by sentence if they want.

But the results don't sound very healthy to me. Look at these statistics. There are many different fields here. I’ll tweet this out. Follow me on Twitter at Yiğit Konur. We have departments like marketing, customer relations, and product. And here are the sectors. Generative AI has the biggest impact on productivity in marketing and sales. That’s where Generative AI is currently providing the most value. What does productivity mean? It’s output. It’s maximizing work units per day. Most of us work in 30-40 person departments. Next year, those teams might run better with just 20 people.

Who gets cut? Those who don't adapt to AI. If that stresses you out, good. You should be stressed. Things won't end well for those who don't adapt. I’m not optimistic about this. It forces us to take real responsibility. That’s why we’re positioning WOP carefully. It’s a content suite for SaaS and e-commerce. It’s not just basic content from ChatGPT. We use data-driven workflows. We run hundreds of prompts to craft one piece of content. We combine those outputs. If we didn't feel this pressure, we’d be vulnerable. What if Amazon says they’re moving their budget to Semrush AI?

Semrush AI might offer them everything they need. That could actually happen. How would that affect Zero? We must keep our vision clear. We have to watch what’s happening globally. Look at the departments using Generative AI most. It’s marketing and product. Damn it, I have two careers. I moved from marketing to product. Both are under the AI influence. Let’s see what’s happening. It's mostly for drafting docs. Or personalized marketing. Or querying text docs. This is just the start. I shared that Stanford experiment to show the potential.

Look at companies doing Generative AI. There’s Jasper. There are like 140 content assistants and writers. They’re all just copies of each other. They all do the exact same tasks. But someone will innovate soon. Then that will get copied too. They’ll learn and iterate. Eventually, they’ll handle most of our daily work. At first, we’ll love it. We’ll think it’s great. Developers are already seeing this. There are amazing plugins out now. Like Cursor. It’s a great AI-based IDE. Or the new GitHub plugin.

Aykut sent it over. I’ve been checking these out. It’s a massive productivity boost for developers. A dev could work 3 hours instead of 8. Nobody would even notice. It’s that clear. Will you use it to slack off? Or will you use that productivity to 10x yourself? That’s the real question. That’s what Yusuf was saying. And he’s not even a marketing guy. The guy is a hardcore engineer. But look at his marketing work. He clearly copied it from somewhere. It’s so obvious. An engineer with no marketing experience couldn't pull that off.

How did he do it? He probably spent two hours talking to AI about it. It was a high-quality conversation. I’ll show you how to have those chats. I’m not here to scare you. But the market is moving incredibly fast. It started in early 2022. ChatGPT really accelerated things. Google launched a new LLM. Meta went open source. Facebook is doing great work. Salesforce is investing. Google is backing a company called Anthropic. It’s Google’s version of OpenAI. They released a product called Claude. It’s excellent. Definitely use it. In many scenarios, it beats ChatGPT. Bloomberg released LLMs trained only on financial data. People are using them to analyze quarterly earnings reports.

You’re in the market, reading reports under time pressure. Others are reading everything and acting on it. Companies are panicking because of this. One of the biggest risks companies see with AI is the herd mentality. AI doesn't always give the right answer. It can lie with total confidence. If you’ve played with it, you’ve seen it. It sounds so convincing. In the AI world, we call this hallucination. That’s how it’s defined. It’s a major risk. Cyber security is the second one. I haven't worked in many enterprise companies. It’s not a huge issue for me personally. Like intellectual property violations. Companies are currently focused on cyber security to solve these problems.

They want to mitigate risks. Is it because it’s critical? No. It’s because it’s easy. They email IT and block chat.openai.com for the whole company. You know companies like that exist. I get the regulation part. But you can just set up a local LLM instance. If you want to do it, you can. You don't have to use Zero. You can get consulting from other firms. OpenAI Enterprise is out. Azure offers this service. You can definitely find companies in Turkey to set this up.

Most companies just find it easier to ban it entirely. I think that’s terrible. It’s bad for management’s reputation too. If we don't use AI effectively at work. We’ll be the losers in this global shift. Look at where the world is heading. People are talking about Universal Basic Income. I suggest you look into it. In wealthy countries, as human labor becomes obsolete, they’re considering a fixed salary. This ensures people don't starve when there are no jobs left. They’ve tested this in the US and Finland. The poverty line in the US is around $1,000. Will we be the ones receiving Universal Basic Income? Or will we be the talent demanded by this changing economy?

This is crucial. That’s why I started on a pessimistic note. Nobody cares if I just say AI is amazing. I wanted to share my own mindset. Let’s talk about consuming AI content. We search Google or YouTube. We see videos claiming AI made them a fortune. They claim they make $8,000 a day. Yet they explain every detail on YouTube. As if their income won't drop if others copy them. That goes against how the global economy works. They claim you can make $23,000 a month. We should all quit our jobs after watching that. That’s the level of content being produced on YouTube right now.

It’s the same in Turkey. You see it on Instagram. There’s a series of ads claiming 99% of people use ChatGPT wrong. There are thousands of them. They all try to sell you a prompt. Prompts aren't meant for copy-pasting, guys. If you copy-paste prompts, you’ve outsourced your brain. Can you outsource your brain? Can you hire an agency for your brain? You need your brain to live. It’s ridiculous. We need to understand how AI works first. We really need to grasp the logic behind it. There are great academic papers on this. I’ll explain how it works without boring you. I can’t sit and read those papers either.

Reading 50-page papers kills my soul. I use AI to summarize them. That’s the only way I can read. AI helps you search smarter. I won't dive too deep. It’s all about embeddings. It can summarize things for you. Don't copy-paste random prompts from strangers. AI can write new content. It adapts to your specific needs. AI can classify data too. If you run an e-commerce site, you need to know this. It’s not just about generation. ChatGPT is just generation. There is so much more. It can cluster data. It can extract insights.

If you’re an e-commerce site, why not use it? Especially if you manufacture products. You have 10,000 reviews. Don't dump that on a junior dev. Don't make them extract insights manually. Managers love assigning that busy work. I never do that. Humans shouldn't waste time on this. It’s beneath human dignity. We need to let AI handle it. Say I produce goods in 8 factories. I have 10,000 reviews for each. I can identify all the pros and cons. I can generate a report for my manufacturer. I can tell them, "People say your sweaters stretch at 40 degrees."

Say you have 87 reviews on this. Your leverage increases. That’s the power of data. No wiggle room there. 87 reviews are set in stone. If your site lacks reviews, pull them from Trendyol. Pull them from Hepsiburada. Get them however you can. There’s so much to do. Don’t just view this as ChatGPT. Understand what this really is. I’m showing these screenshots for a reason. They’ve cheapened prompt engineering. Selling prompts for 50 bucks is a joke. These researchers are from OpenAI, Google Brain, or other LLM firms.

Like Cohen and others. I’ll share some examples. They won’t be boring. I’ll keep it clean. People always tell me, "We use AI, but the results suck." Let’s be real. If AI isn't working, it’s 99% your fault. The world has mastered this tech. You think it can’t solve your basic problem? That doesn't make sense to me. Let’s be honest with ourselves. If you can’t solve it with AI, you’re doing it wrong. You haven't researched it properly. I see this online constantly. Whenever I meet someone, I ask, "What are you doing with AI?"

What’s your company talking about? I keep asking this. Most answers are just nonsense. Let’s start with few-shot. It’s the simplest thing on earth. The projection is perfect. I’m here to help you out. You are an AI travel guide expert. You specialize in beach vacation spots. Give clear answers to questions. Question. Can you list places to visit in Izmir? Look, I want beach spots. Okay? Ephesus ruins? As an Izmir local, there’s no beach there. Here is the point of few-shot. It’s very basic. I write the same message. You are an AI that does this and that.

Question. Can you list places to visit in Izmir? For swimming. Alaçatı, Urla, Foça, Dikili, Karaburun. Then I ask. Can you list places to visit in Muğla? The first answer was wrong and poorly structured. I only want the names of the spots. No fluff. I don’t want any of that. I just want this specific part. I gave an example here. Answer. Alaçatı, Urla. I used commas and spaces. LLM models always try to continue the text above. That’s the logic. Just knowing that is enough. Going deeper is hard and unnecessary. My knowledge is pretty surface-level too. When I ask for Muğla, it gives me Bodrum, Marmaris, Fethiye. It follows my Izmir pattern perfectly.

Why? Yiğit already gave an example for İzmir above. When I ask for Muğla, it doesn't write a wall of text. It copies the format. Providing examples is crucial. It boosts your output quality by 10x. This is called few-shot prompting. Look it up. You'll see it's easy. Just write your own examples. It's not hard. Now, let's go deeper. It's about understanding how AI thinks. This is called chain of thought prompting. You can research this too. These are just jargon terms. Before I got into IT, jargon used to scare me off. Don't run away. There are simple ways to learn this. Forget the tutorials.

Just ask the AI. It explains it itself. That's the beauty of it. The only issue is AI training data isn't always current. It can be outdated. Use Claude for that. I recall Claude's data goes up to 2023. Let's move on. Ready? Let's make this fun. The left side is a weak prompt. The right side is a solid one. Let's expand the right one. I'm listing the rappers I enjoy listening to. I like Tupac, Biggie, Nas, Ice Cube, Snoop Dogg, Eazy E. I dislike Future, Migos, Lil Baby, DaBaby. I don't like them. Based on this list, suggest new rappers I might enjoy. You probably don't write prompts like this.

I shared this because it's a great example. If you just ask AI directly, it gives you an answer. You might like Kendrick Lamar or J-Cole. I personally don't like 4 or 5. Why? It says, "Yiğit, you're old school." It says you like artists with political and social messages. It gets straight to the point. Now, let's move to the second topic. I go to AI and ask the same question. I list the rappers I like. Then I add a paragraph. I tell it to think step by step. Analyze why I like or dislike them. Then find results that actually fit my taste.

Honestly, just saying "think step by step" is enough. I wrote all of this in Turkish. I don't usually recommend using Turkish with AI. I just used it so everyone could understand. So, what did it tell me? It said, "Yiğit. You like 80s and 90s hip-hop culture." It noted the strong social themes and storytelling. It identified that I hate mumble rap and trap. It correctly noted the repetitive hooks and heavy autotune. Based on that, it gave me better recommendations. It should have included MF DOOM. It shouldn't have included Logic. I forced it to reason through the prompt. Think of it like this. If I ask you a question, a quick answer is more likely to be wrong.

Give it some time. Tell it to slow down. Think out loud. Take it step by step. Tell it, "You like these things, Yiğit. So, you don't like those." That’s how the human brain works anyway. You need to create space for AI. Don't just fire questions at it. Force AI to think step by step. This applies to ChatGPT too. By the way, I’m using ChatGPT for all of this. You can combine chain of thought with these examples. This is super valuable. It’s the same setup on the left. You can also teach AI how to think. For example, I explain why I like or dislike certain rappers. I wrote this part myself.

Okay? I wrote this section. AI wrote this other part. Let's say you listen to Tupac, Biggie, and Yiğit together. These are controversial rappers. But you aren't into them. You care about the actual music output. Ice Cube, Eazy-E from NWA, Snoop Dogg, and so on. That’s why I’d recommend this to you. If you like Nas, you like New York rappers. You should listen to Rakim. I wrote this. This is an example of how to think. Now, I’m listing classical musicians. I gave an example. You can use rock, pop, or whatever you want. You could even write Batu Flex. It’s better if you don't. I mentioned classical artists here. I said I like Erik Satie, John Cage, and Philip Glass.

I said I don't like Wagner, Mahler, or Schönberg. See what it did here? In the white section, it’s mimicking my thought process. I just taught AI how to think by writing it out. It’s thinking just like me now. Look what it wrote. It says I like minimalist and experimental works. It identifies those as minimalist elements. It notes I dislike complex, heavy, romantic, or dodecaphonic pieces. Then it suggests what I should listen to. That’s it. I gave it an example and defined the logic. Then I asked a follow-up. You must create examples and thought patterns. The third technique is Tree of Thoughts.

I just started using this. It’s very simple. It’s a great way to understand how AI works. Imagine a panel in your head. Tell it you have a question. Imagine three experts for this. Each expert writes one step of their thought process. They share it. Then they move to the next step together. If an expert realizes they’re wrong, they leave the group. Then we ask our question. We write the prompt right here. This is the English version. Don't just copy-paste it. I’m showing you the logic behind it. Now, let’s get back to my music talk.

Yiğit is hunting for rappers again. There’s a three-person panel here. It says Yiğit, you didn't write these down. We don't know your thoughts. But you clearly love the late 90s and early 2000s. Three experts are debating this. This is so addictive. Or maybe just to me. It says Yiğit loves Nas, Biggie, and Tupac. So he likes lyrical rappers. One expert suggests Jay-Z. Okay? Then the second expert chimes in. If he likes Ice Cube and Eazy E, he’s West Coast. He should listen to Dre and Lamar. The third says if he likes Snoop and Ice Cube, he loves G-Funk. He should listen to Warren G. They start debating. Expert one evaluates expert two.

He says you're right. Jay-Z fits his lyrical taste. He gives a thumbs up. Number two says G-Funk is a solid call. Let's recommend those. Look, they’re actually talking to each other. I’m obsessed with this. I do this with like 10 people. I create different characters. Sometimes I go crazy. The presentation goes off the rails. It shows wild stuff. Sometimes I do this. This might be cringe, but I create a panel of 10 or 20 people. I give them random names. It’s a bit schizophrenic, but it works. They give examples. I might not like John or Mark’s suggestions.

I tell John and Mark they're out. Two new people are coming in. I tell the others why I swapped them. See? It works. John and Mark's outputs were just nonsense. The others see that nonsense. The next iteration gets better results. They eventually agree. Expert 1 and 2 say they've reached a consensus. They suggest Lamar, Jay-Z, and Warren G. I actually listen to all three. That's the story. What's the point? It's not about copy-paste prompts. It's about understanding reasoning. It's about teaching AI how to think. It's about asking the right questions. You don't even need me for that.

I want to improve my prompt engineering. I'll work on my soft skills. What should I do? I have AI translate this for me. GPT-4 does it automatically. I ask how to improve myself. Check out Perplexity AI. It's a great project. Seriously, it's amazing. You know what it does? It doesn't just give an AI answer. It asks follow-up questions. It doesn't annoy you. It uses multiple-choice questions. It really understands you. It even searches the web. It says "Developing soft skills as prompt engineers." It articulates the problem clearly. It runs a few searches. It reads all the results. Then it gives you an answer. Perplexity AI is free. You should definitely use it.

I’ve cut way back on Google searches. Seriously. It reads the info for me. Google used to be a nightmare. Ten results. Hope no one from Hürriyet is here. If you land on Hürriyet without an adblocker, your fan starts screaming. I don't know if they're running Flash or what. It’s awful. You get lost in the clutter. AI ignores all that junk. It reads it for you. It gives you the core message directly. It says, "To improve your prompt engineering skills. Do this." You need to break problems down. It gives me the output I want directly. That’s huge. So, just ask AI everything.

This sounds ridiculous. We went for hookah yesterday. Might sound a bit trashy to some. I live in Lisbon, so I enjoy these things back in Turkey. We had to pick a flavor. The menu was digital. I copied it. The names are weird. Hookah smokers know. I asked Perplexity for the ratings. I told it to pick the best one. It suggested a top-tier flavor. Keep that in mind. Try it if you go to Ukka. I’m talking about that level of integration. Install Perplexity on your phone. Get the app. Don't go to the website every time. There are great apps for this.

You can use them. It’s all about learning to think effectively. Yagiz showed me this site. It’s seriously great. Exactly. It’s all about critical thinking. It shows you these frameworks. It teaches you how to think deeply about a topic. It’s built for people. For example, there’s Six Thinking Hats here. I’ll dive into this more in a bit. You’ll see why I mentioned this site. It helps people, especially managers, with their decision-making processes. Antos. Very cool. That prompting technique I showed earlier? It’s actually an academic technique. Tree of Thoughts. Its background is actually critical thinking. It says to think by wearing six different hats.

Make the first hat positivity. Make the green hat creativity. Let the red hat be emotional. Let the white hat be data. Make the black hat the downside. Let the blue hat be the process. Think according to these. It creates six characters. Create them in your head. Put your decisions on the table using these. Tree of Thoughts comes from this, I think. It’s all about how you think. Projecting your internal monologue onto AI is the whole point. We’ll move to softer topics now. I have a weird way of using my computer. Or so people say. I love productivity hacks. The new macOS has a great dictation engine.

Check out the beta. It’s just me talking. It finishes with zero errors. I write entire paragraphs. This isn't huge for Americans. But Turkish dictation used to be terrible. It was broken for years. It still struggles if you mix Turkish and English. But it works perfectly in pure Turkish. People speak at 150 words per minute. Writing is only 50 words per minute. You can write 3x faster. I spend two hours a day on Discord. This is a game changer. I highly recommend it.

You talk to AI so much you get lazy. If you're lazy, you use AI less. That kills your learning curve. This is a big deal. There is a great app called MacWhisper. It’s for the Mac ecosystem. I recommend it. They built a solid product. You can even ask AI how to use AI. I’m a marketing manager. How should I use this? Perplexity searches for a Digital Marketing Manager Prompt Guide. It scans Google results. It reads them. It gives you exactly the output you need.

This is honestly super interesting. My favorite trick? Write a prompt. Feed it to AI. Let AI improve it. That’s a solid method. You can even write a prompt that optimizes other prompts. It’s smart to have one master prompt for all your prompts. I just say, "Optimize this." I do this all the time. For example, when we’re assigning tasks at the company. Yusuf mentioned CEOs using AI for decision-making. It’s the best way to stay neutral. Say we’re assigning tasks in the marketing department.

Let’s say we have Mehmet, Burak, and Oğulcan. I list their strengths and weaknesses based on my observations. I list the tasks. Then I ask AI who should handle what. Based on my input, it organizes everything perfectly. It really helps with decision-making. After optimizing, here is the most important step. Ask AI to critique your prompt in 10 points. Have it score each point out of 10. It might say, "Length is 7/10, clarity 8/10. Completeness 9/10." Then I say. "Fix this prompt so it gets 10/10 on every point. Propose a strategy and get my approval."

I tell it to optimize the prompt if it approves. We have a bit of a tough dynamic. Our moods shift sometimes. I don't even read the full output before moving on. I like your strategy. Now apply it, I say. I’m already prepping the next prompt while it’s still typing. GPT-4 is a bit slow. It says, "If you edit based on this strategy. I await your approval." It has such a sweet tone. What did we actually do? We forced a chain of thought. First, we made it critique. Then we asked it to explain how to apply those critiques.

Everything is clear now. It just needs to do the work. It created its own brief. Look, it cleaned everything up. You know what it did? It took Ali and Veli and assigned them digital marketing tasks. It simplified things. Ali is organized and an Excel pro, but weak on visuals. Veli is disciplined and great at Figma and Canva. It says to break down tasks like SEO, content creation, and visual design. It’s mapped out all the sub-topics. I only mentioned digital marketing tasks earlier. That’s the approach. You need to keep AI handy to do this. There are cool projects for using AI on Mac. One of them is ThinkBuddy.

It’s a project from Turkey. People you know might be behind it. Not an ad, though. A brilliant developer is leading it. They built a great tool. Use hotkeys to open an AI window and ask ThinkBuddy anything instantly. Ufuk is launching this next month. The best part is the voice feature. You can ask AI a question by voice. Just press a button. You might ask what it thinks about global warming. It takes your input directly. Speaking is so much easier than typing. It’s a great feature. It can also handle emails. Select an email and ask it to summarize. Or explain it like I’m five.

Or ask it to rewrite. It summarizes long emails instantly. Some people send way too much text. It gets annoying. It gives you three options based on your selection. You just click to take action. Coming soon. It’s 7 dollars a month. We’ll offer special pricing for Turkey. This is my weekend project, by the way. Most friends here didn't know. Now they do. Let's wrap this up. I definitely went over my allotted time. Beyond that, here are 6 or 7 quick points. You absolutely must speak to AI in English. It searches the English corpus much better. English is the language of the internet, after all.

You have to accept that. If you don't get the right output, it’s on you. You’re the one doing it wrong. You need to improve your approach. Now that you get these concepts, dive deeper into prompt engineering. Test what I’ve shared. Apply it to your own problems. Want to go further? Look into AI agent'ları. They let you program AI. I won't go into deep detail. But get your hands dirty. Check out embeddings. It’s about clustering documents. Searching through them. If you run an e-commerce site, you can redesign your product categories with AI. Here is the bad news. Companies have no one who knows AI.

Nobody is learning. Our company is great at this, thankfully. My friends' companies are too. But unfortunately, that’s the reality. People are ignoring it. They seem happy that way. Your dev team won't bring this to the table. As marketers and product people, you must bring it up. You know the value proposition for the user. It’ll hurt, but we have to learn. Ask Perplexity. There are concepts in AI you need to know. Have you ever had a long chat with AI? Sometimes it forgets the top of the chat. Why? You need to know that. I mentioned Claude a few times. GPT-3.5 Turbo has an 8K limit. Maybe 16K. Claude has a 100K context size.

You can feed it 70,000 words of info. Give it a massive book. Ask it questions about it. It’s insane. I do this with Claude. I go to a website. I do Cmd+A. Select all. Cmd+C. Copy. I paste it into Claude. I ask, what does this site do? Or what is this article about? I read stuff on Hacker News sometimes. I don't get it. I copy it. I ask it to explain like I'm 13. It works great. Or there's a YouTube video about psychology. I don't quite get it. I copy it.

I ask questions about the video. What did they mean here? I can feed it the whole transcript. It’s perfect. This is a researchable topic. Our company has an AI bir culture. We record our meetings. We transcribe the audio. We write docs from these transcripts. For example, we created an API dokümanı from a meeting with our designer. We turned it into text. We asked what we discussed. We asked it to break it down. Make an API dokümanı. That feature was actually built using that API dokümanı. It had some bugs. Sorry, Recep. But we fixed it. It was super easy.

Writing an API dokümanı is hell. I hate it. AI is a great excuse to avoid tasks you hate. It’s fun. There are cool projects. GPT for Sheets is great. Burak’s presentation on the Zeo YouTube channel is also great. If you prefer starting with a video instead of tinkering. I gave you the keyword. So you can search it. This field changes so fast. Sometimes it really bums me out. I see something new on Twitter and get FOMO. The FOMO is real. You have to stay on top of it. You might need to change who you follow on Twitter. You might need to flood your YouTube with AI. Things change seriously fast. If I went offline for six months, I’d be depressed.

I thought about that recently. Being offline for six months would be a disaster. Everything changes so much. The world shifts in just a month or two. You definitely need to follow Hacker News. If you're a dev, check trending GitHub projects. There’s a site called Futurepedia. It lists products built with AI. You can follow everyone Yusuf follows on Twitter. People in Turkey stalk Yusuf’s followers like crazy. I tell everyone to follow who Yusuf follows. You can check my follows too. They’re all right here. Thanks for listening. I hope I didn't create too much of a pessimistic vibe.

This transcript was generated using AI speech recognition and verified against subtitle style rules.

Yiğit Konur

Content author

Yiğit Konur

both

Contact us

How we handle your data

Open as a full page