Wordware (YC S24)’s Post

Prompt Engineering: The Real Deal! I've shared a lot about prompt engineering (and the memes that come with it!). But let’s get clear on what it actually means. There’s a big misconception that prompt engineering is just about crafting prompts (aka "prompt designer"). But if we stick to the true engineering spirit, it’s much more than that. In my view, real prompt engineering involves a deeper, technical skillset—think software development, ML engineering, data science, and more. It’s about applying engineering principles to get the most out of LLMs. So next time someone claims to be a prompt engineer, ask them about the building blocks of the picture. That’s where the real expertise shows. #AI #GenAI #Promptengineer

With the recent rise of AI, is coding a lost art? As a cyber professional and long time developer, here are my two cents: Knowing how to write simple to moderate algorithms is no longer necessary. Knowing how to parse strings, write regular expressions, setup interprocess communication mechanisms, and write SQL queries are all outdated. AI is at the point where it knows what you need, even more than you do. So what should young developers and recruiters focus in on? ✅ Debugging, Debugging, Debugging! While AI can provide you a basic proof of concept, it cannot debug a problem within your team’s project of 100,000 lines of code. ‼️Young professionals ‼️ In your interviews, your goal is to show how you would go about solving a problem. Walk us through your debugging steps. Prove to us AI didn’t write all your college projects for you, but that you’ve spent days of your life in the GDB debugger, just like the rest of us. 🤝

The talk of the day is of course ’Devin’ and their $21M Series A funding. I’ve gotten a mass inflow of requests for my opinion on the matter. So here it is: As mindblowing and intriguing as this is, and a true statement of where genai is heading, I am still curious what data was used to train the model? Is the code generated safe to use? Free of bugs or poor coding decisions? What about vulnerabilities? Will I be using someone elses code without knowing its origin? Whose ”perfect engineering workflow” is it mimicking? I surely hope it isnt mine!! ;) Basically: Who has it trained from and on what? The issues with genAI is still real for the enterprise. There still needs to be a careful and seasoned code reviewer in place and efficient and up to date license and code-vulnarability scans. And still I would be nervous putting this code in production. I am not saying humans are better, but at least there is a audit track, a human responsible origin of sorts… the black box decision making is what is scary. Hopefully ’Devin’ and its likes will be treated as highly skilled interns for some time still and will not cause big business harm. And hopefully the many ’Devins’ of the world to come - and the masses will come and markets will flood, each left with a lesser and lesser slice - will think about these things and truly empathise with the challenges of enterprise and not just jump headless into sexy technology ”because they can”. #responsibleai

Almost every company on my feed is hiring for LLM and GenAI based roles. RAG, LLM fine-tuning and MLOPS seem to be the most sought out skills in this domain. There's also a great demand for learning resources, content and experimental validations in these domains. No wonder, my last 4 articles on Medium on some fundamentals like tokens, embeddings, attention, transformers, all did very well. Interestingly, every developer company now needs high quality content via blogs and tutorials related to their product and documentation to create good data that can be used with an RAG bot on their website. Ease of access and ease of help/learning are two important qualities that make a developer product get adopted widely. If you are looking for someone to create deep and hard technical content for your product documentation, website or blog, feel free to hit me up. My focus would be on explaining research for applied engineering and experiments/tutorials involving various techniques and concepts in ML/AI, in such a way that everyone from all backgrounds can follow along and learn seamlessly. Linking some of my past GenAI related articles in the comments, which have gathered over 26K views in the last few months on Medium, all with a read ratio of greater than 50%. For long form content with over 10 min reads, that's been quite a ride.

Now that #BlackHat and soon #DEFCON are off to the races, excited to show off something we’ve been working on: Meet your new security engineering and threat intel research AI: fluent in thousands of programming languages and scripts (including query and markup languages for popular enterprise security hardware and software products), trained with volumes of CVEs and decades of hacking experience, and capable of automating the hardest tedium faced my overwhelmed security and infrastructure engineering teams. This is Kindo WhiteRabbitNeo - the first AI specifically built to make security and infrastructure/devsecops lives easier. Usable natively within Kindo’s enterprise security platform for AI, on its own, or even as a copilot within VSCode. Or better described by one of my secops friends: “it’s like pair programming with a horrifyingly powerful mixture of Jarvis, Metasploit, and Carl Sagan.”

🚀 Exciting Update: Project Submission Success! 🚀 https://lnkd.in/gXSK5thR After two months of dedicated work, I am thrilled to announce that my project has been successfully submitted to Hackster! 🎉 A huge thanks to AMD for providing the essential hardware that made this project possible. Project Overview: My project embodies the concept of Kaizen—constant self-improvement. It is designed to analyze study material, extract and answer questions ( It is able to pinpoint the right solution based on the questions provided ), and provide insightful explanations and help you understand on why you got a particular answer wrong. This feature helps users understand their mistakes and enhances their learning process. 📚✨ Technical Details: AI Model: Utilised Hugging Face’s powerful transformer model for natural language understanding and processing. Backend: Developed with Flask, enabling smooth handling of requests and responses. Functionality: The system reads and interprets study texts, answers questions, and provides detailed feedback on errors to foster continuous improvement. Why This Matters: This tool is more than just a question-and-answer machine; it’s an educational assistant designed to empower users by helping them learn from their mistakes and steadily improve their knowledge. #AMD #AI #MachineLearning #Transformers #Flask #EducationTech #Hackster #Python #Flask

For those seeking a step-by-step method to expand a prompt in sequence to create a scene, there are no limits to the imagination! #Test2video #Runway Gen-3 #Scripting #AItool #GPT # AI

𝗗𝗲𝘃𝗶𝗻 ❌ 𝗗𝗲𝘃𝗶𝗹 😈 ✅ . . Lots of students are panicking about the new 𝘼𝙄 𝙎𝙤𝙛𝙩𝙬𝙖𝙧𝙚 𝙀𝙣𝙜𝙞𝙣𝙚𝙚𝙧 𝘿𝙚𝙫𝙞𝙣. So my point is here please don't 𝗣𝗔𝗡𝗜𝗖 ⚠ 1. Lots of AI tools are in the market and it's not going to happen like this, some companies will allow a complete takeover of AI tools and fire all software developers. 2. AI tools will always boost our productivity and help human brains think more about creativity than working on repetitive work. P.S. Let me know your thoughts in the comment section. we are happy to discuss 😁 #aitools #devin #softwareengineering

💡 SYNNQ Devnet v0.3.2 is Live! This update migrates our Neural Network to NQA, resulting in increased transaction speed, less overhead, and the integration of Quantum Graph Neural Networks (QGNN) to enhance SYNNQ's capabilities. See what’s new in detail below: Neural Network Migration: • Migrated to NQA (nqa.synnq.io) • /predict endpoint now exclusively handles predictions using the default_model • Forecasting functionality has been discontinued Performance Improvements: • Transaction processing is now 420% faster thanks to the NQA migration • Significantly reduced system overhead NQA Version 0.1 - Key Enhancements: Federated Learning Framework: • Clients can now train models locally and submit updates to the global model via /submit_model_update • Aggregates Quantum Graph Neural Network (QGNN) updates from multiple clients with new aggregation logic Quantum Graph Neural Network (QGNN): • Hybrid quantum-classical model with Variational Quantum Circuit (VQC) • Supports multiple variational layers and qubit entanglement • Outputs probabilities of specific quantum states APIs & Model Configuration: • /get_global_model endpoint for downloading the latest global model • Configure training parameters, model settings, and input data via JSON files • Dynamic model distribution to new clients requesting local training • Models can be dynamically configured using JSON files 👉 Join and Discuss in our Discord Server: https://lnkd.in/dTw_GetC #Devnet #Blockchain #AI #QuantumComputing #neuralnetworks #QuantumNetworks #QuantumComputing #ArtificialIntelligence #MachineLearning #developers #developer #dev #programming #coding #programmer #webdevelopment #javascript #softwaredeveloper #webdeveloper #developerlife #coder #code #development #programmers #technology #codinglife #software #computerscience #python #typescript #rust #coders #java #softwareengineer

Devin AI has gained attention recently, impressing many with its capabilities. However, the idea that it could completely replace software developers is very unlikely. It can certainly assist us in software development and bug resolution, but deploying code to production requires a skilled software development engineer (SDE) who can ensure its functionality and take responsibility for any issues. If problems arise, can we truly hold Devin AI accountable? Not Probably soon. Nevertheless, it will undoubtedly be a valuable tool for SDEs, if you know, you know ;) These opinions are personal, subject to change with the evolving role of responsibility in AI. What are your thoughts? #DevinAI #genai #generative #chatgpt #datascience #jobs