- Modern Chaos
- Posts
- MC.5: Workday Play, AGI Horizons, MetaPrompter Magic, Whisper Turbo Mode, and Sequoia's $200B question 🌳
MC.5: Workday Play, AGI Horizons, MetaPrompter Magic, Whisper Turbo Mode, and Sequoia's $200B question 🌳
This week, delve into the near horizon of AGI, the turbocharging MetaPrompter for ChatGPT, and the lively ethos of a fun-driven workplace. Plus, the noteworthy performance of Mistral-7B-OpenOrca, and Sequoia Capital's $200B pondering on generative AI's real value.
modern chaos issue 5 - generated by Ideogram.ai
Modern chaos is a newsletter exploring tech and AI through the journey of a dev agency shifting from services to product design. We share our notes, analysis and experiments.
Hi!
This is the fifth issue of Modern Chaos. I started working on Alter, the first project to pass our App Grill Process. My work began by designing the desktop app on Figma. The user experience looks promising. We’ll start developing the first version soon, I will keep you posted!
Cheers
In this issue:
🎉 Having Fun at Work: More Than Just a Motto
🧠 Why AGI is Closer Than You Think
🚀 MetaPrompter: Turbocharge ChatGPT
🧪 Promptfoo: Prompt Testing
💭 Think-verify-revise Framework: Logical Chain-of-Thought in LLMs
🎨 qrGPT & QRious: Cool Looking QR Codes
⚡ Whisper Turbo: Cross-Platform, GPU Accelerated Whisper
🌪️ Mistral-7B-OpenOrca: 98% of Llama2-70B-chat’s performance
🌳 Sequoia Capital: The $200B question
Work & Experiments
Having Fun at Work: More Than Just a Motto
Many companies boast values that rarely move past breakroom posters. At our web agency, when we say "Have Fun," we mean it. And here's how this mantra has become an action-filled journey for us. READ IT
Tech updates & tools
Why AGI is Closer Than You Think
Samuel Hammond explains why he has short timelines to Artificial General Intelligence (AGI) and why we should have too. He makes strong points, one being that AGI is about mimicking the human brain and that’s something that could be brute force. READ IT
MetaPrompter: Turbocharge ChatGPT
Not everyone is skilled in writing great prompts. Joseph Thacker wrote a small tool to improves its prompts by running them through a Metaprompter before sending them as the main prompt to ChatGPT. I use meta prompters myself, It’s super valuable for finding the right experts/references for a given task. READ IT | GITHUB
Promptfoo: Prompt Testing
Promptfoo supports a wide range of LLM APIs and allows you to evaluate LLM outputs side by side. No more guessing and it can be integrated in your CI system to make your LLM product more resilient to failures. GITHUB | WEBSITE
Think-verify-revise Framework: Logical Chain-of-Thought in LLMs
While I’m unsure about the real improvement over the Tree-of-Thoughts approach, I’m amaze at the people creativity for optimizing zero shot prompting. I’d like to write an historical blog post about the different strategies we’ve seen since the start of GPT. Paper
qrGPT & QRious: Cool Looking QR Codes
The Twitter gallery of QR codes made with QRious is worth a look. There’s also an open source project with a similar goal. QRious is offering 20 generation free. Is it the start of a new QR codes trend? TWITTER | QRious | qrGPT
Whisper Turbo: Cross-Platform, GPU Accelerated Whisper
A lightning-fast Whisper implementation, designed to run entirely client-side in your browser and faster than real-time. Definitely a project to watch for any project looking for a live speech-to-text solution. GITHUB | DEMO
Mistral-7B-OpenOrca: 98% of Llama2-70B-chat’s performance
This is the first 7B model to score better overall than all other models below 30B. The model is a finetune of Mistral with the OpenOrca dataset. Super impressive when we know that the the finetuning costed only 400$. TWITTER | DEMO
Sequoia Capital: The $200B question
“What are you going to use all this infrastructure to do? How is it going to change people’s lives?”, David Cahn, Sequoia Partner. I think he nailed it: generative AI’s biggest problem is not finding use cases or demand or distribution, it is proving value. How GPT and others can provide enough value for users to stick to the product? READ IT
Reply