Murat Yener

Coding with Claude 3.5 Sonnet is ~10x more efficient than GPT-4o or any other LLMs out there. The new Artifacts feature enhances user experience by generating and running code next to your chat. Wild - Try it for free: claude.ai #ai #anthropic #claudeai #llms #coding

6

1 Comment

The Lies We Tell Ourselves. Senthu Sivasambu embarks on a quest to uncover an objective characterization of monolithic software, revealing the superficial facade of monolithic and microservice architectural styles in this enlightening article. He leverages a unique analogy to present the findings. 💡 "Pagodas will always be pagodas" . No surface-level distinctions here—just a fresh perspective. #microservices #monolithics

7

**Unleashing the Power of RAG: A Workshop Recap** ✨ Last Saturday, I had the incredible opportunity to delve into the world of Retrieval-Augmented Generation (RAG) at a workshop led by Faizan Ahmed. This powerful approach combines the strengths of large language models (LLMs) with external knowledge sources to revolutionize question answering. Imagine being able to ask complex questions about massive datasets and receiving comprehensive, informative responses! RAG makes this possible by building graph-based text indexes and generating insightful summaries. I'm excited to explore how RAG can be applied to various fields, from research to customer service. Stay tuned for more updates on this transformative technology! #RAG #LLMs #AI #NLP #workshop #datascience #MachineLearning #HeadstarterAI 🚀🤖🧠

2

1 Comment

Highlighting: "The amount of training data for each language was not the only factor at play: The study found that the chatbot had particular difficulty with low-resource languages that were structurally different from English." ----- AI training for LLMs reflects what you put into it. And if you train only in English you don't reflect the rest of the world. Open AI and most AI solutions made in English speaking countries have major failings outside of English. ----- RestOfWorld: "We tested ChatGPT in Bengali, Kurdish, and Tamil. It failed. Outside of English, ChatGPT makes up words, fails logic tests, and can't do basic information retrieval." (6 Sept 2023) (Andrew Deck) "ChatGPT is being used all over the world, generating Amazon listings in China and call center scripts in the Philippines. But while ChatGPT thrives in English, Spanish, Japanese, and other dominant languages, it struggles to produce the same quality of text in languages like Bengali, Swahili, Urdu, and Thai — languages that have millions of speakers, but appear much less often online. When Rest of World tested ChatGPT’s ability to respond in underrepresented languages, we found problems reaching far beyond translation errors, including fabricated words, illogical answers and, in some cases, complete nonsense. Take Tigrinya, a language which has over 7 million speakers, with the vast majority located in Eritrea and the northern part of Ethiopia. Tigrinya shares a similar script to Amharic, a more dominant Ethiopian language, but there are significant differences between the two. When asked to list examples of African countries, ChatGPT mixed up Tigrinya and Amharic, adding characters that don’t exist in Tigrinya. It created an output to this simple question that is challenging to read for native speakers of both languages. ChatGPT also inserted countries that aren’t on the African continent, including Jordan and Canada. Researchers told Rest of World that proper nouns — names, places, and institutions — are a persistent weak point for ChatGPT, and that this problem is common across many underrepresented languages. “If you ask ChatGPT in Tigrinya or Amharic the simplest and most frequently asked questions, it gives you gibberish, a mix of Tigrinya and Amharic, or even made-up words,” said Asmelash Teka Hadgu, co-founder and chief technology officer of Lesan, a startup developing machine translation products for Ethiopian languages. “Chatbots like ChatGPT are utterly broken or useless for these languages.” Many of these languages are what AI researchers call “low-resource.” AI language models are largely trained on data scraped from the internet. While languages like Bengali are some of those most spoken in the world, they are less represented online, so there is less digitized text available to train models tailored to them." RestOfWorld: https://lnkd.in/g2hPqKUc #ai #fakeai #genai #degenai #generativeai #degenerativeai

24

8 Comments

Pro-tip for those attempting to productionize gen-ai services based on text input and output: scrutinize the text input severely and relate that to the precise wording in the prompt. it is incredible *how* subjective text evaluation really is and I have learned just how *good* but subjective LLMs are. I find myself being confused about who's right in the evaluation: myself or the LLM! One thing stands out though: when I go back and read my prompt I often realize that the wording I used is actually not that clear, even though in everyday parlance another speaker *would seemingly know exactly what I was asking or saying*. It's a wild world.

2

Instead of asking, 'Which diagramming tool should we use for our system architecture?', reframe the question to address the core issue. For example: 'We need to ship functional software on schedule. How can we easily visualize and access up-to-date system architecture information?' You'll quickly see that static diagramming tools won't cut it for modern, complex systems. You need a dynamic tool that embraces complexity, supports evolving architectures, and integrates seamlessly with your system design process. https://lnkd.in/eYmBUVEV

5

Check out Gradio for quick LLM App demo creation

4

1 Comment

GitLab's built-in code search is super limited. I wrote a quick guide on how to make it an order of magnitude better using Sourcebot

26

Pondering Hybrid vs Native mobile apps? And if you go Hybrid, then React vs Flutter? These articles are *excellent* and highly recommend them for any technologist out there. Plus they provide ideas for your scorecard when making such decisions. Have you seen the stats on what is popular? Are you surprised? Article 1: Hybrid vs Native https://lnkd.in/g-uMKuqq Article 2: React vs Flutter https://lnkd.in/gAu2fEey Statista: cross platform mobile frameworks used by developers 2019-2023 https://lnkd.in/gi4GTMum

11

4 Comments

State of the Frontend in 2024! https://lnkd.in/gQ9-YwMS TL;DR: 1. React and Next.js are here to stay. 2. Redux and Redux Toolkit is still kinda popular. 3. Vercel, Github Actions, Node.js , Typescript and Tailwind css lead in their spaces. 4. GraphQL is waning in popularity 5. Dedicated QA teams are shrinking, devs expand their roles further into testing. 6. ChatGPT and Copilot are popular choices for AI tools, but Sonnet 3.5 is viewed as a better code generation model. Curious to see how much of a spike Cursor (uses Sonnet 3.5) will enjoy! These are just some items that really caught my attention, more details in the report...

16

1 Comment

LinkedIn Caption: Tired of struggling to keep your developer documentation up to date? 😵‍💫 Doc-E.ai simplifies the process by turning messy docs into clear, actionable content faster than you can fix a bug! 🛠️ Stop drowning in wires and outdated docs—let us help you maintain accurate, real-time documentation effortlessly. 🚀 Empower your developer team to focus on what they do best while we handle the documentation chaos! 📚💡 #DocEai #DeveloperTools #DeveloperEngagement #DocumentationSimplified #DevRel #TechSolutions

4

2 Comments

Join me for an interesting session on #devex and how to think about organizational and engineering metrics.

15

One of the key reasons I chose to join RelationalAI more than a year ago to continue my passion and journey to "Democratize Graph / Knowledge Graph for All" is our deep talents (computer scientists / engineers) in #database, #knowledgegraph, #logicprogramming, #machinelearning, #rules, #optimization, #highperformancecomputing #julia, and intersection of all the above. Below is just one of the examples of research work by RelationalAI's talented team with lasting impact to the industry.

20