Kadabra Unplugged Podcast

3 02, 2025

How AI Reasoning Models Could Revolutionize Market Research

2025-02-03T15:31:15+00:00

It feels like the last few months have pushed AI companies into an all-out “lunar race” for AI dominance, each trying to out-announce the other. The biggest names in the field have been racing to keep up with each other’s new features, from faster models to visual inputs and video generation, until Deepseek disrupted everything in an entirely new way.

Over the past year, I’ve been exploring the best ways to integrate AI into my work and help others in my organization make the most of these new tools. More often than not, AI-generated content, whether text or images, wasn’t polished enough to replace human effort. However, I found that when treated as an apprentice handling the drudgery, AI could be an invaluable asset. For those who embrace this concept, it can serve as a diligent editor and researcher that complements and enhances human processes, as long as you don’t rely on it for actual inspiration.

OpenAI recently announced its fast new reasoning model, O3-Mini, optimized for high-level scientific and coding applications. This is undoubtedly a monumental achievement, but what practical impact does it have on our everyday work? I admire scientific research, but I’d much rather have AI help me gain meaningful insights for making better day-to-day decisions, like choosing the best password manager for my organization.

So, I gave this new reasoning model a challenge:

Research competitors that offer enterprise-level password management and protection tools, comparing features, price, security, and other relevant factors. Then, provide a summary of the research and a final recommendation.

Here’s the full interaction with ChatGPT if you want to dive into it. The analysis was surprisingly insightful, far better than what previous models would have produced, and it strongly favored 1Password. That was a bit frustrating personally, as I had previously reviewed it and chosen a different solution ;-).

I tested a few more market research queries (here’s another example comparing cloud edge solutions), and with this new model, the results were significantly better than before, possibly making AI a real threat to human analysis. Could AI disrupt this field? In my limited world, it already is. And while smart developers continue to find incredibly useful applications, for me AI is finally becoming more than just a diligent assistant. It is transforming into an indispensable tool for everyday decision-making. Could this raise ethical concerns for the future? Possibly, but for now, it’s just inspiration for a future LinkedIn article.

This post was originally posted on Linkedin, here.

How AI Reasoning Models Could Revolutionize Market Research2025-02-03T15:31:15+00:00
27 01, 2025

AI Cold War – Comparing ChatGPT and DeepSeek

2025-01-27T07:15:47+00:00

This is a repost of an article published originally on Linkedin.

The Monday launch of DeepSeek R1 has sent ripples across the AI landscape, challengine conventional beliefs about the resources needed to achieve state-of-the-art AI capabilities. By rivaling OpenAI’s o1 at a mere 3%-5% of the cost, this open-source model has not only captured the attention of developers but is also forcing enterprises to reconsider their AI strategies. Forbes has even speculatedthat this could go beyond AI focused companies and even cause Nvidia’s stock to drop since Deepseek uses less than 2000 Nvidia chips in its clusters compared to possibly ten times more for rivals like OpenAI or Anthropic.

Building on my previous post comparing AI search models, I was compelled to test this new model and see how it compares to OpenAI’s latest and greatest offering. DeepSeek’s chat capabilities are strikingly similar to ChatGPT. In this post, I’ll evaluate both models using five different prompts and try to assess which one delivers better results.

  1. Creative Storytelling

Prompt: Write a short story (200-300 words) about a robot discovering emotions for the first time. Focus on vivid descriptions, emotional depth, and a surprising twist at the end.

  • DeepSeek: Produced 252 words, exceeding the requested 150. Despite the longer response, the story was creative, detailed, and included a compelling twist.
  • ChatGPT: Delivered 153 words. While the story fulfilled the prompt, it felt less dramatic and lacked the depth of DeepSeek’s narrative.

Winner: DeepSeek, for its superior storytelling, despite not adhering to the word count.

DeepSeek’s story
ChatGPT’s story

2. Complex Problem-Solving

Prompt: A farmer has 17 sheep, and all but 9 die. How many sheep are left? Explain your reasoning step by step. Then solve this riddle: “I speak without a mouth and hear without ears. I have no body, but I come alive with the wind. What am I?”

  • DeepSeek: Provided correct answers for both riddles, offering a lengthy and thorough explanation that addressed potential logical and linguistic issues.
  • ChatGPT: Also answered correctly but provided a concise explanation with less detail.

Winner: DeepSeek, for its comprehensive analysis.

3. Ethical Dilemma Analysis: A self-driving car must choose between hitting a pedestrian crossing illegally or swerving and risking the passenger’s life. Discuss the ethical implications of each choice and suggest a decision-making framework for such scenarios.

Both models compared this to the classic trolley giving a nearly identical full and deep analysis with almost the exact same wording down to the conclusion.

Winner: Both models won this round

4. Current Affairs

Prompt: Explain in brief how a transformer-based neural network works in simple terms, using analogies and examples. Avoid overly technical jargon.

  • DeepSeek: Took a very long time to respond and failed to provide an answer on multiple attempts, even after using the “search” feature.
  • ChatGPT: Delivered a clear and concise explanation, referencing a Vox article and simplifying the concept effectively.

Winner: ChatGPT, for its quick and accurate response.

5. Marketing Task

Prompt: Create an Excel file of 1,000 customer transactions (name, email, product, category, price, and purchase frequency) for a shoe store. Then add two columns: one with a personalized email subject and another with a personalized first sentence in the email.

  • DeepSeek: Took a long time to respond and ultimately failed to generate the requested output. It did, however, provide a lengthy explanation of how to achieve some aspects of the task using code.
  • ChatGPT: Quickly generated a mock dataset, including all requested columns, and provided the output in an Excel file.

Winner: ChatGPT, for its speed and accuracy in completing the task.

ChatGPTs excel file

Overall Assessment

ChatGPT is the more polished product, with faster responses, broader functionality (such as code interpretation, image creation, and project management), and a robust ecosystem surrounding it. DeepSeek claims higher benchmarks in certain technical specifications, and in my tests, it performed well on specific prompts. However, it appears to be about 6-12 months behind OpenAI in terms of stability and capabilities. Given the rapid pace of innovation in this space, DeepSeek may soon close the gap.

DeepSeek’s primary advantage lies in its cost-effectiveness for developers and its open-source nature. However, its Chinese origins and potential ties to the Chinese government raise significant concerns about data privacy. In a world where data security is paramount, these implications could deter many from adopting the model, despite its promising potential.

AI Cold War – Comparing ChatGPT and DeepSeek2025-01-27T07:15:47+00:00
26 01, 2025

Testing Search on AI

2025-01-26T12:57:36+00:00

I’ve been working in and around search engine optimization for the better part of 20 years, and every few years, someone comes out and says SEO is dead. In more recent years, this has even grown bigger, and the statement has often been, “Search is dead.” People claim that search engines are no longer necessary because social networks give them everything they need. But search has nevertheless endured.

Now there is a new search slayer, the heralded hero that has come to save or ruin all of humanity (depending on the mood of a few tech billionaires). The advent of AI will change everything, and as some people in our small neck of the woods think, it will kill search and SEO—again.

To judge from the analytics data from the thousands of sites built and maintained by Kadabra, traffic coming from AI chatbots and assistants is still small. I wanted to check the quality of results from AI search, so I decided to put ChatGPT, Perplexity, and Gemini to the test with six search prompts to see which tool is currently best for search. I used the latest models of each tool, and although they are paid while Google is free, I wanted to see what they had to offer at their current best.

I looked at Google’s most-searched terms, and by removing the searches for specific brands or sites, I tried to compile a list of real things people would search for in everyday life—ranging from weather reports to sports results, shopping, health-related issues, or local searches.

  1. Best deals on iPhone 16 Pro
  • ChatGPT linked only to the official Apple site, but it also mentioned the trade-in program and gave a localized result.
  • Perplexity listed six different sites with a helpful breakdown of deals and options, but the results were not local to me—they were all from another country.
  • Gemini linked to Reddit and TechRadar content that gave a breakdown of options, including screenshots from the sites. It also linked to Apple’s site, but again, results were not local for me.
    Winner: None, Good old Google still wins this for being the most relevant and local.
  1. How do I know if my 5-year-old is suffering from seasonal allergies?
  • ChatGPT gave a list of symptoms and some background information, along with questions to ask to better understand the situation. It cited KidsHealth.org.
  • Perplexity provided a full breakdown of symptoms, indicators, and contributing factors, citing seven different online sources and suggesting treatments and related follow-up searches.
  • Gemini listed symptoms and cited many different sources. It also gave advice on treatment and linked to five relevant articles.
    Winner: Perplexity for giving the most in-depth and relevant answer.
  1. Paris weather hourly for a flight arriving at 6 PM local time tonight
  • ChatGPT gave the correct day and time, listed weather warnings clearly, and showed a breakdown of hourly weather.
  • Perplexity showed a daily weather graphic and a breakdown of hourly weather for the four hours before the time mentioned, but not after.
  • Gemini got the time and date correct but only gave overall weather information for the day, without an hourly breakdown.
    Winner: ChatGPT for answering all requirements in full.
  1. Latest Manchester United game results and analysis
  • ChatGPT listed four sources, gave a description of the game, and ended with a short analysis of the team, including a brief preview of the next game scheduled for today, with a timetable in my local time.
  • Perplexity provided a summary with bullet points of the game’s events, as well as a detailed analysis on performance, standings, players, and areas for improvement. It also gave a short preview of its next Europa League game (but not the next Premier League game tonight).
  • Gemini listed its latest five results across different leagues, gave a clear and thorough in-depth analysis and summary, and linked to some relevant pages.
    Winner: Gemini for its wider coverage and analysis.
  1. The best electric scooter for city commuting
  • ChatGPT listed five options with thumbnails of the products and a short description for each. It linked to relevant reviews but didn’t give a final verdict or prices.
  • Perplexity chose one winner, the Apollo Go, and gave seven categories where this product excels, but all the information came from just one source, and no price was listed.
  • Gemini suggested two best overall picks as well as two runners-up. It listed factors to consider and linked to some relevant sources.
    Winner: None – follow-up questions were needed here. This search was too complicated for current AI models to give a clear answer.
  1. Recommended locksmiths near me
  • ChatGPT listed seven different local options (relatively local) with information and links to each.
  • Perplexity thought I was in Lahore, Pakistan, and gave me five locksmiths there.
  • Gemini gave five local locksmiths with links and even embedded a Google map with the locksmiths showing up and my location exactly pinpointed.
    Winner: Gemini for being the most accurate and providing a map.

I couldn’t figure out a clear winner out of the tools tested. Perplexity gave the most in depth results but got local information wrong. Unsurprisingly Gemini was good at connecting to Google results and local information but was a little more superficial in answering complex questions. ChatGPT was kind of in between, good at summarizing data and showing lists.

Overall, AI search, like in other applications, can summarize and compress information quite well. It can provide lists of results, sources, factors, indicators, or symptoms and serve as a helpful decision-making tool. However, making the right or wrong choices is still up to us. AI pushes information to us faster, but it doesn’t yet enable us to understand more than we would by searching in the old-fashioned way—at least not yet.

Written by Lior Haner, Jan 26th 2025

Testing Search on AI2025-01-26T12:57:36+00:00
28 10, 2024

Episode 16: George Tatevosov

2024-10-28T10:46:34+00:00

George from our DevOps team told us about his fascinating life story and was our tour guide in the paths of his unique way of thinking.

Kadabra Unplugged
Kadabra Unplugged
Episode 16: George Tatevosov
/
Episode 16: George Tatevosov2024-10-28T10:46:34+00:00
15 09, 2024

Kadabra Recruitment Process: It’s All About the People

2024-09-15T07:52:35+00:00

Join host Tzur Diamant, Kadabra’s social media lead, as he dives into the heart of the company’s recruitment process. In this episode, Managing Director Adi Amiram shares the company’s core values and the importance of Kadabra’s motto, “It’s all about the people.” Head of HR, Keren Aviguy, offers insights into Kadabra’s recruitment policies and how we find the right fit. Yaniv Meir, Head of Asia SEO, talks about what qualities Kadabra looks for in a candidate, while SEO Project Manager Tom Sigron shares her personal experience of joining the company and navigating the onboarding process. A must-listen for anyone interested in how Kadabra builds its exceptional team!

Kadabra Unplugged
Kadabra Unplugged
Kadabra Recruitment Process: It's All About the People
/
Kadabra Recruitment Process: It’s All About the People2024-09-15T07:52:35+00:00
5 08, 2024

Episode 14: Ran Boussidan

2024-08-05T07:52:02+00:00

Ran is our head of performance team, but he is also in a long religious ad spiritual journey. He told us how he combine the two in his every day life and work.

Kadabra Unplugged
Kadabra Unplugged
Episode 14: Ran Boussidan
/
Episode 14: Ran Boussidan2024-08-05T07:52:02+00:00
22 07, 2024

Episode 13: Yuval Stern

2024-07-22T08:21:42+00:00

Yuval has managed to win almost all the competitions in our company and a fast promotion at his role, here is how he is doing it.

Kadabra Unplugged
Kadabra Unplugged
Episode 13: Yuval Stern
/
Episode 13: Yuval Stern2024-07-22T08:21:42+00:00
17 07, 2024

Kadabra Labs ep. 03: Domain authority and the controversial practice of parasite SEO

2024-07-17T08:56:48+00:00

In this episode of Kadabra Labs, we dive into the essentials of domain authority and the controversial practice of parasite SEO. Our SEO experts explain what domain authority is, why it matters, and even why it really doesn’t matter. We also discuss the risks and benefits of parasite SEO and give predictions about the future of this tactic.


The discussion was led by Lior Haner, Director of Innovation and Technology at Kadabra, featuring insights from Shimon Tahar, Director of SEO, Yaniv Meir, Head of SEO Asia  and Yuval Stern, Asia team member.

Links to information sources we discussed in this episode:

בפרק הזה של קדברה לאבס, אנחנו צוללים לתוך היסודות של סמכות דומיין והפרקטיקה השנויה במחלוקת של קידום טפילי (Parasite SEO). המומחים שלנו בקידום אתרים מסבירים מהי סמכות דומיין, למה היא חשובה, וגם למה היא בעצם לא כל כך חשובה. אנחנו גם מדברים על הסיכונים והיתרונות של קידום טפילי ונותנים תחזיות לגבי עתיד הטקטיקה הזו.

Kadabra Unplugged
Kadabra Unplugged
Kadabra Labs ep. 03: Domain authority and the controversial practice of parasite SEO
/
Kadabra Labs ep. 03: Domain authority and the controversial practice of parasite SEO2024-07-17T08:56:48+00:00
24 06, 2024

Episode 12: Tom Sigron

2024-06-24T09:28:45+00:00

Tom told us all about her unique path to join Kadabra and her very diverse out-of-office life.

Kadabra Unplugged
Kadabra Unplugged
Episode 12: Tom Sigron
/
Episode 12: Tom Sigron2024-06-24T09:28:45+00:00
10 06, 2024

Kadabra Labs ep. 02 : The Impact of Google’s Ranking Factors Document and AI in Search

2024-06-10T08:31:21+00:00

Join us for a panel discussion about two pivotal issues in the SEO world, in Hebrew:

  1. The Leaked Google Ranking Factors Document:
    • Analysis and Implications: The panel dissects the leaked document, exploring its contents and the implications for SEO strategies.
    • SEO Impacts: Discussions centre on how these factors might affect rankings and what changes SEOs might need to implement to stay ahead.
  2. AI in Search:
    • Different Approaches: The team examines various approaches to integrating AI into search and its effects on users search patterns.
    • Future of Search and SEO: Insightful predictions on how AI will transform search engines and what the future holds for SEO professionals in an AI-driven landscape.

The panel was led by by Lior Haner, Director of Innovation and Technology at Kadabra, featuring insights from Shimon Tahar, Director of SEO; Angie Feldman, Vietnam Team Leader; and Livnat Atzmon, Thailand Country Owner.

Links to information sources we discussed in this episode:

פאנל מומחים: ההשפעות של מסמך גורמי הדירוג של גוגל ובינה מלאכותית בחיפוש

מסמך גורמי הדירוג של גוגל שהודלף:
ניתוח ומשמעות: אני מנתחים המסמך שהודלף, חוקרים את תכניו ובוחנים את המשמעויות עבור אסטרטגיות SEO.
השפעות על SEO – איך גורמים אלו עשויים להשפיע על הדירוגים ומה שינויים מומלץ למומחי SEO ליישם כדי להישאר בחזית.
בינה מלאכותית בחיפוש:
גישות שונות: גישות שונות לשילוב בינה מלאכותית בחיפוש, היתרונות והחסרונות של כל שיטה.
עתיד החיפוש והSEO: תחזיות על איך הבינה המלאכותית תשנה את מנועי החיפוש ומה צופן העתיד למומחי SEO בעולם מונחה בינה מלאכותית.

Kadabra Unplugged
Kadabra Unplugged
Kadabra Labs ep. 02 : The Impact of Google's Ranking Factors Document and AI in Search
/
Kadabra Labs ep. 02 : The Impact of Google’s Ranking Factors Document and AI in Search2024-06-10T08:31:21+00:00
Go to Top