Good Morning! Microsoft's making waves with Drasi, their new open-source project that's about to make event-driven systems a whole lot easier to manage. Google's not sitting idle either - they've just dropped Gemini 1.5 Flash-8B, a leaner, meaner version of their AI model that's ready to tackle your production needs. And if you're a ChatGPT fan, get ready for a game-changer: Canvas is here, bringing a whole new dimension to how we interact with our favorite AI assistant.

— Forrest Knight & Meghanadh Vasireddy

Microsoft Open-Sources Drasi

Event-driven systems are awesome, but they can be a real headache to manage at scale. Detecting relevant changes across components? Nightmare. Ensuring real-time responses? Even worse. Current solutions often involve cobbling together multiple tools, resulting in fragile architectures that are a pain to maintain.

Enter Drasi, Microsoft's latest open-source project. This lightweight data processing system simplifies change detection and reaction in complex systems. Here's the cool stuff:

• Sources: Connect to various data sources, monitoring for critical changes in real-time.

• Continuous Queries: Constantly evaluate incoming changes using Cypher Query Language.

• Reactions: Execute automated actions when changes trigger predefined criteria.

The best part? Drasi watches for events in logs and change feeds without copying data to a central lake or constantly querying sources. It's designed for extensibility, so you can create custom integrations to fit your needs.

Drasi's been submitted as a CNCF Sandbox project, so expect some community love soon. Ready to simplify your event-driven architecture? Check out drasi.io and dive into the GitHub repos!

Read More Here

Gemini 1.5 Flash-8B: Faster, Cheaper, and Production-Ready

Remember Gemini 1.5 Flash from I/O? Well, the team at Google DeepMind has been busy cooking up an even more streamlined version. Flash-8B is smaller and faster, but don't let its size fool you – it's packing some serious punch.

What's New:

• 50% price drop compared to 1.5 Flash

• Double the rate limits (now 4,000 RPM)

• Lower latency on small prompts

The best part? Flash-8B nearly matches its bigger sibling across many benchmarks, especially shining in chat, transcription, and long-context translation tasks.

Here's where it gets really interesting. Flash-8B boasts the lowest cost per intelligence of any Gemini model:

• $0.0375 per 1M input tokens (<128K)

• $0.15 per 1M output tokens (<128K)

• $0.01 per 1M tokens on cached prompts (<128K)

Whether you're building high-volume multimodal apps or tackling long-context summarization, Flash-8B is worth a look.

Read More Here

ChatGPT Canvas

ChatGPT's been our go-to for quick coding help and writing tasks, but let's face it - the chat interface can be a bit limiting for larger projects. Enter Canvas, the first major UI update since ChatGPT's launch two years ago.

What's new:

• Side-by-side collaboration: Canvas opens in a separate window, allowing you to edit and refine alongside ChatGPT.

• Smart context understanding: Highlight specific sections for targeted feedback, with ChatGPT considering the entire project.

• Shortcut heaven: Quick actions for both writing (e.g., adjust length, change reading level) and coding (e.g., review code, add logs, port to different languages).

Canvas automatically triggers for suitable prompts, but you can also summon it by including "use canvas" in your message. It's rolling out to Plus and Team users now, with Enterprise and Edu access coming next week.

Read More Here

🔥 More Notes

• Patent troll Sable pays up, dedicates all its patents to the public!: Cloudflare defeated patent troll Sable in court, proving that Cloudflare did not infringe Sable's patent and that the patent was invalid. As a result, Sable agreed to pay Cloudflare $225,000, grant Cloudflare a royalty-free license to Sable's entire patent portfolio, and dedicate all of Sable's patents to the public.

• Cloudflare Study: 39% of Companies Losing Control of Their IT and Security Environment: The Cloudflare study found that 39% of companies are losing control of their IT and security environment due to increased complexity from remote work, security threats, cloud apps, and compliance. Companies need to acknowledge this complexity and focus on continuous learning to adapt.

• Working Turing Machine: The model has 4 ( 2²) possible symbols and 8 (2 ³) possible states, so in total 32 possible symbol-state combinations. Each instruction has 7 bits (3 for the state, 2 for the symbol, 1 for moving left/right and 1 for stopping), so the "source code" takes 732 = 224 bits or 14 bytes. That means you can make 2^224 ≈ 2.6910⁶⁷ programs!

📹 Youtube Spotlight

Grail Demo from CHM tape

Was this forwarded to you? Sign Up Here