- Modern Chaos
- MC.22: Alibaba's Qwen-VL Model, Google's Generative AI for Maps, and Microsoft's Future of Work Insights
MC.22: Alibaba's Qwen-VL Model, Google's Generative AI for Maps, and Microsoft's Future of Work Insights
This week, Alibaba outperforms with Qwen-VL despite export controls, Google Maps introduces a new generative AI feature, and Microsoft's report highlights AI's role in workplace efficiency. Plus, a look at Whisper's audio transcription tool and Nat Bullard's comprehensive analysis on decarbonization.
modern chaos issue 22
Modern chaos is a newsletter exploring tech and AI through the journey of a dev agency turning into a startup studio. We share our notes, analysis, and experiments, plus a bit of Emacs tips.
Last week, I conducted a poll to understand what you expect from the newsletter.
The responses were mixed, but one thing is certain: nobody wants a short form with only links to tools and news. It's a relief because I enjoy curating the content and explaining why I find it interesting.
0% | SHORT: Just links and tools with brief reviews, no polls
50% | MIXED: Links and tools with brief reviews, a poll on a topic, and results from the previous poll
50% | FULL: A summary of long-form content with commentary, results from the previous poll, and links with brief reviews
The poll for this newsletter concerns AI agents and the security trade-offs we are willing to accept if they become proficient at delegable tasks.
It’s linked to the post by Joseph Thacker, which shares his ideas from a security expert's standpoint. Make sure to respond, as it will be an interesting topic for next week!
In this issue:
🗣️ Whisper: Transcribe 2.5 hours of audio in 98 seconds.
🛍️ Alibaba: The Surprisingly Effective Qwen-VL Model
🐞 Google: AI-powered patching approach resolves 15% of bugs
🗺️ Google Maps: New Generative AI Feature Coming
🖥️ Alvea AI: On-Demand User Interface
📊 Tom Tungunz: AI Design Patterns
🔒 Joseph Thacker: Required Security Changes for Secure AI Agents
📜 UK House of Lords Committee: Final Report on Generative AI Models
🚀 Microsoft: Future of Work Report
🌍 Nat Bullard: A comprehensive 200-slide analysis on decarbonization
Updates & tools
Whisper: Transcribe 2.5 hours of audio in 98 seconds
A CLI for transcribing audio files using Whisper on your device, compatible only with NVIDIA GPUs & Mac. As our devices evolve, they acquire an increasing range of capabilities. Entrepreneurs need to consider designing products that can seamlessly transition to a local-first model. Ignoring this aspect invites competition based solely on this factor. MORE
Alibaba: The Surprisingly Effective Qwen-VL Model
Even with the lack of computational power due to export controls, researchers from Alibaba have been able to produce powerful models. The model Qwen-VL-Max outperforms both GPT-4V from OpenAI and Gemini on Chinese benchmarks. More
Google: AI-powered patching approach resolves 15% of bugs
They built a custom AI application for fuzzing and integrated it with other existing automated and human systems. As a result, it’s able to scale the throughput of finding bugs and providing fixes. My takeaway: stitch some AI into your existing processes to optimize specialized tasks, you might be surprised at how well it accelerates your operations. MORE
Google Maps: New Generative AI Feature Coming
Simply say what you're looking for, and Google LLM will analyze Maps' 250 million places and insights from 300 million contributors, quickly making suggestions for where to go. MORE
Alvea AI: On-Demand User Interface
In my view, conversational UIs fall short as the optimal mode of interaction with LLM. The construction of on-demand UIs, ranging from simple forms to intricate web components tailored to the user's immediate task, stands as a promising alternative or supplement to chat interactions. MORE
Tom Tungunz: AI Design Patterns
Implementation patterns are emerging and it's important to have the mental models in place to choose the best one depending on your app's maturity state. Tom Tungunz presents the current architecture patterns of most solutions in the market with well-thought-out charts. MORE
Joseph Thacker: Required Security Changes for Secure AI Agents
AI agents are rapidly emerging, yet the challenge lies in providing them secure access to our accounts. Joseph proposes a dual-method solution. The first involves setting up an account akin to an employee's, complete with email and phone number for multi-factor authorization. I've yet to see startups working on this problem. MORE
What's your stance on AI agents having access to your accounts?
How comfortable are you with this idea, especially if it could amplify tenfold your productivity on a specific task?
UK House of Lords Committee: Final Report on Generative AI Models
A fitting title for the report might be "Brace for Impact". The report states that LLMs will have impacts comparable to the invention of the internet. The expressed goal for the UK is not to mimic the regulatory frameworks or strategic decisions of Europe or the USA, but to carve out a unique national stance on AI. It remains to be seen how distinct this approach will be. MORE
Microsoft: Future of Work Report
🔍 Key metrics — Generative AI enabled individuals to accomplish routine writing tasks 37% quicker. In a mock consulting project, BCG consultants generated outputs of over 40% superior quality with the aid of LLMs. Microsoft's Copilot users finished tasks in 26-73% of the time compared to non-users, and 73% concurred that Copilot expedited their task completion. MORE
Nat Bullard: A comprehensive 200-slide analysis on decarbonization
Impressive effort. Nat delivers granular insights into decarbonization, viewed through the prism of climate, capital markets, and technology. One slide reveals a significant increase in AI mentions in public company presentations and earnings calls - 40,000 in 2023 compared to under 20,000 in 2022. This not only underscores the growing focus on AI but also highlights its role in climate solutions, such as mitigating the warming impact of contrails from flights. MORE