🔍 Vector Search Tuning: The Hidden Knobs

🔍 Vector Search Tuning: The Hidden Knobs

Vector search has a lot of tuning knobs, but which one should you change first when search quality isn't where you want it to be? Key Points:

• Vector Search Tuning Knobs: hnsw_ef, candidate depth, RRF k, quantization, reranking, and more. Each knob affects search quality, but understanding their interplay is crucial.
• Tuning Strategy: Start with the most impactful knobs, such as hnsw_ef and candidate depth, and iteratively refine other parameters to achieve optimal search quality.
• Actionable Takeaway: Regularly monitor search quality and adjust tuning knobs accordingly to ensure optimal performance.

🔗 Resources:

• Original post ↗
• Qdrant Engine
• Vector Search Tuning Knobs
  • Brief description: Vector search tuning knobs

🚀 GSAP Agent Skills Demo

Made this crazy demo to show off the GSAP agent skills I've been iterating on. It got a little out of hand. Powered by @greensock inspired by the old glory days of Flash Key Points:

• GSAP Agent Skills: GSAP agent skills are a powerful tool for creating complex animations and interactions. This demo showcases their capabilities.
• Agent Skills: Agent skills are a set of pre-built animations and interactions that can be easily combined to create complex behaviors.
• Actionable Takeaway: Use GSAP agent skills to create engaging and interactive experiences.

🔗 Resources:

• Original post ↗
• Greensock
• GSAP Agent Skills Demo
  • Brief description: GSAP agent skills demo

🚨 Cache Hits and GPU Waiting

A cache is only worth what it hits. The community's having a KV cache moment. Here's the corner we work in: In long-context serving, a cache hit can still leave the GPU waiting for data. When KV lives outside GPU memory, how you bring it back matters. That's the problem we set Key Points:

• Cache Hits and GPU Waiting: Cache hits are crucial, but they can also lead to GPU waiting if the data is not readily available.
• GPU Waiting: GPU waiting occurs when the GPU is waiting for data to be fetched from memory or other sources.
• Actionable Takeaway: Optimize cache hits to minimize GPU waiting and improve overall performance.

🔗 Resources:

• Original post ↗
• Tencent AI News
• Cache Hits and GPU Waiting
  • Brief description: Cache hits and GPU waiting

🤖 AI Agents at Scale

One month = 130,000 engineering tasks done by AI agents. @DoorDash is already operating Agents at serious scale. Once you reach that point, the model isn’t the only thing that matters. You need: → Agent identity → Tool access → Permissions → Communication → Key Points:

• AI Agents at Scale: AI agents can perform a large number of tasks, but they require careful management and optimization.
• Agent Identity: Agent identity is crucial for ensuring that AI agents are properly authenticated and authorized.
• Actionable Takeaway: Establish clear agent identity and access controls to ensure secure and efficient operation.

🔗 Resources:

• Original post ↗
• Fetch.ai
• AI Agents at Scale
  • Brief description: AI agents at scale

🎙️ reSpeaker Clip: Open Wearable AI Recorder

Meet reSpeaker Clip — an Open Wearable AI Recorder built for real-world conversations. Clip it. Wear it. Capture it. Build with it. But it’s more than a recorder. 🔓 Open SDK & Firmware 🎙️ Dual-Mic Audio Capture 📡 Wi-Fi 6 + Bluetooth 5.3 💾 2GB Onboard Storage From Key Points:

• reSpeaker Clip: reSpeaker Clip is an open-source wearable AI recorder designed for real-world conversations.
• Open SDK & Firmware: The device comes with an open SDK and firmware, allowing developers to customize and extend its capabilities.
• Actionable Takeaway: Explore the reSpeaker Clip's features and capabilities for building innovative AI-powered applications.

🔗 Resources:

• Original post ↗
• Seeed Studio
• reSpeaker Clip
  • Brief description: reSpeaker Clip

🛠️ Site of the Day: IKEA “Taking Apart the 30”

Site of the Day - IKEA “Taking Apart the 30” (30th Anniversary) MRM Spain — @mrmspain /nk.studio — @NKStudio Mach Studio — @StudioMach ⚙️ ScrollTrigger, SplitText, useGSAP 🛠️ React site → https:// family.ikea.es/demos/desmonta ndo-los-30/ … showcase → https:// gsap.com/showcase Key Points:

• Site of the Day: IKEA's "Taking Apart the 30" website showcases innovative use of GSAP and ScrollTrigger.
• GSAP and ScrollTrigger: The website utilizes GSAP and ScrollTrigger to create engaging animations and interactions.
• Actionable Takeaway: Explore the website's code and learn from its implementation of GSAP and ScrollTrigger.

🔗 Resources:

• Original post ↗
• Greensock
• Site of the Day
  • Brief description: Site of the day

🚀 Browserbase vs Obscura

okay, this is genuinely impressive. browserbase charges $99 a month for 500 browser hours. obscura runs unlimited, free, on your own machine. → 30mb memory vs 200mb+ chrome → native rendering, no chromium needed → real js via v8 engine → puppeteer and playwright compatible Key Points:

• Browserbase vs Obscura: Browserbase and Obscura are two tools for browser automation, but they have different pricing models and features.
• Obscura: Obscura is a free, open-source tool that runs on your own machine, offering unlimited browser hours and native rendering.
• Actionable Takeaway: Consider using Obscura for browser automation tasks, especially for large-scale or long-running tasks.

🔗 Resources:

• Original post ↗
• The Whizz AI
• Browserbase vs Obscura
  • Brief description: Browserbase vs Obscura

🌐 The Digitized World

If you think about it... Our world is at the verge of becoming 100% digitised. We already spend a huge part of our lives behind screens. AI is bringing intelligence into the digital realm. Crypto is bringing ownership, money and coordination into it. And we’re quickly Key Points:

• The Digitized World: The world is rapidly becoming digitized, with AI and crypto playing key roles in shaping this new reality.
• AI and Crypto: AI is bringing intelligence to the digital realm, while crypto is enabling ownership, money, and coordination.
• Actionable Takeaway: Stay informed about the latest developments in AI and crypto to navigate this rapidly changing world.

🔗 Resources:

• Original post ↗
• Ivaav Music
• The Digitized World
  • Brief description: The digitized world

🚀 Monad API Hub

Introducing Monad API Hub One place for developers and agents to access data, search, and AI APIs. No API keys. No subscriptions. Pay per request with USDC on Monad. Explore API Hub: https:// app.monad.xyz/agents/api-hub Key Points:

• Monad API Hub: Monad API Hub is a platform that provides access to data, search, and AI APIs, with a pay-per-request model.
• No API Keys or Subscriptions: Monad API Hub eliminates the need for API keys and subscriptions, making it easier to access APIs.
• Actionable Takeaway: Explore Monad API Hub for accessing a wide range of APIs.

🔗 Resources:

• Original post ↗
• Monad
• Monad API Hub
  • Brief description: Monad API hub

🤖 FLOP and GenLayer

AI agents need three things to operate at scale: food (compute) to power them, memory to keep them running, and a place to settle disputes when things go wrong. This is the thesis behind FLOP and GenLayer. Great take from the godfather of crypto derivatives, @CryptoHayes . Key Points:

• FLOP and GenLayer: FLOP and GenLayer are two concepts that aim to address the challenges of operating AI agents at scale.
• Compute, Memory, and Dispute Resolution: AI agents require compute power, memory, and a mechanism for resolving disputes to operate effectively.
• Actionable Takeaway: Explore FLOP and GenLayer for insights into operating AI agents at scale.

🔗 Resources:

• Original post ↗
• GenLayer
• FLOP and GenLayer
  • Brief description: FLOP and GenLayer