DEV Community

DRIX10
DRIX10

Posted on Originally published at blogs.drix10.com

Computer Vision and AI Applications #247

#tech #ai #coding

🤖 AI/ML - Skills vs MCP vs RAG vs Memory

Skills vs MCP vs RAG vs Memory: What Each One Actually Does Four questions, four concepts. How should it work? Skills (the playbook). What can it connect to? MCP (the tool/data standard). What knowledge does it need now? RAG (retrieval into the answer). What should it retain?

Key Points:

  • Skills (Playbook): A set of rules or guidelines that define how a system should work, providing a framework for decision-making and problem-solving.

  • MCP (Tool/Data Standard): A standardized format for data and tools, enabling seamless integration and exchange between different systems and components.

  • RAG (Retrieval into the Answer): A mechanism for retrieving relevant information from a knowledge base, allowing a system to answer questions and provide solutions.

  • Memory (Retention): The ability of a system to retain and recall information, enabling it to learn and improve over time.

🔗 Resources:

🚀 AI/ML - FrontierSmith Accepted to NeurIPS

FrontierSmith has been accepted to NeurIPS as spotlight. Since releasing, we kept receiving lots of discussion about the work from lots of data team. I believe data (especially synthetic data) becomes the really central topic! I’m proud of the FrontierCS team and it’s my great

Key Points:

  • FrontierSmith Accepted to NeurIPS: FrontierSmith's work has been accepted to NeurIPS, a prestigious conference in the field of artificial intelligence.

  • Synthetic Data: Synthetic data is becoming increasingly important, especially in the context of data teams and AI research.

  • FrontierCS Team: The FrontierCS team is to be commended for their work, which has received significant attention and discussion.

🔗 Resources:

📚 Education - Career Update

Overdue career update time! I’ve moved to Toronto and will be starting graduate school (MS/PhD) @UofT & @VectorInst , advised by @florian_shkurti ! I will remain @ Stack AV until the end of the year to wrap up some projects, after which I’m open for internships!

Key Points:

  • Career Update: The author is moving to Toronto and starting graduate school at the University of Toronto and Vector Institute.

  • Graduate School: The author will be pursuing a master's or PhD degree, advised by Florian Shkurti.

  • Internships: The author is open to internships after completing their graduate school projects.

🔗 Resources:

🚀 AI/ML - DCI Accepted to NeurIPS

DCI has been accepted to #NeurIPS2026! Huge congratulations to @zhuofengli96475 , @GhxIsaac , @CongWei1230 , @lupantech , @DongfuJiang , and all collaborators! Paper: https:// arxiv.org/abs/2605.05242 Code: https:// github.com/DCI-Agent/DCI- Agent-Lite …

Key Points:

  • DCI Accepted to NeurIPS: DCI's work has been accepted to NeurIPS, a prestigious conference in the field of artificial intelligence.

  • Collaborators: The authors would like to extend their congratulations to all collaborators on this project.

  • Paper and Code: The paper and code for this project can be found on arXiv and GitHub.

🔗 Resources:

🤖 AI/ML - Tencent ARC Releases GAE

Tencent ARC releases GAE: a geometry-native latent space for 3D-consistent world generation GAE compresses 3,072-channel geometry features into a compact 128-channel latent that decodes into RGB, depth, camera trajectories, and point clouds from the same generated state. It

Key Points:

  • Tencent ARC Releases GAE: Tencent ARC has released GAE, a geometry-native latent space for 3D-consistent world generation.

  • GAE: GAE compresses 3,072-channel geometry features into a compact 128-channel latent that decodes into RGB, depth, camera trajectories, and point clouds.

  • 3D-Consistent World Generation: GAE is designed for 3D-consistent world generation, allowing for the creation of realistic and detailed 3D environments.

🔗 Resources:

🚀 AI/ML - Opus 5.5 from @AnthropicAI

Opus 5.5 from @AnthropicAI on ARC-AGI (Verified): - ARC-AGI-2: 93.3%, $0.41/task - ARC-AGI-1: 98.5%, $0.16/task Opus 5.5 outscored Opus 5 by 2.9 percentage points on v2 and 1.0 on v1, at roughly 80% lower evaluation cost.

Key Points:

  • Opus 5.5 from @AnthropicAI: Opus 5.5 has been released by @AnthropicAI, with significant improvements in performance and efficiency.

  • ARC-AGI-2 and ARC-AGI-1: The performance of ARC-AGI-2 and ARC-AGI-1 has been improved, with significant reductions in evaluation cost.

  • Opus 5.5 vs Opus 5: Opus 5.5 has outperformed Opus 5 by a significant margin, with improvements in both performance and efficiency.

🔗 Resources:

🚀 AI/ML - Introducing XOR

Introducing XOR An open source, multimodal Jev-like decision model. Being developed with data sovereignty & enterprise grade in mind. ❶ Multimodal: great for text + visual classification ❷ Much bigger 260k context window ❸ Qwen3.6 based https:// huggingface.co/juspay/xor ↓↓↓

Key Points:

  • Introducing XOR: XOR is an open-source, multimodal decision model being developed with data sovereignty and enterprise-grade in mind.

  • Multimodal: XOR is designed for multimodal classification, allowing for the combination of text and visual data.

  • Context Window: XOR has a larger context window, allowing for more complex and nuanced decision-making.

  • Qwen3.6: XOR is based on Qwen3.6, a powerful and efficient decision-making framework.

🔗 Resources:

🤖 AI/ML - Physical AI Data Race

The Physical AI data race just hit a new frontier: nursing homes. Seniors are strapping on head-mounted cameras and folding paper by hand, and every crease becomes training data for robots.

Key Points:

  • Physical AI Data Race: The physical AI data race has reached a new frontier, with the use of head-mounted cameras and folding paper to generate training data for robots.

  • Nursing Homes: Seniors in nursing homes are participating in this effort, providing valuable data for the development of AI-powered robots.

  • Training Data: The creases in the folded paper are being used as training data for robots, allowing them to learn and improve their performance.

🔗 Resources:

🚀 AI/ML - Microsoft @Surface Powered by @snapdragon X2 Plus

From #SnapdragonSummit: Microsoft @Surface powered by @snapdragon X2 Plus. Built with advanced on-device AI and all-day battery life in a sleek, portable design to help people stay productive wherever work happens.

Key Points:

  • Microsoft @Surface Powered by @snapdragon X2 Plus: Microsoft @Surface is powered by @snapdragon X2 Plus, a powerful and efficient processor.

  • Advanced On-Device AI: The device is built with advanced on-device AI, allowing for improved performance and efficiency.

  • All-Day Battery Life: The device has all-day battery life, making it ideal for use on-the-go.

  • Sleek, Portable Design: The device has a sleek and portable design, making it easy to carry and use.

🔗 Resources:

🚀 AI/ML - Risk from Secret Labs

If you believe the risk is coming from 1-3 people in a garage with no money and no compute, you just don't understand this technology and haven't learned anything this summer. Risk comes from the asymmetry of capabilities created by secret labs training frontier agents and

Key Points:

  • Risk from Secret Labs: The author argues that the risk from AI and machine learning comes not from small groups of people, but from the asymmetry of capabilities created by secret labs.

  • Asymmetry of Capabilities: The author believes that the risk comes from the fact that some groups have access to advanced capabilities and resources, while others do not.

  • Secret Labs: The author suggests that secret labs are a major contributor to this asymmetry, as they are able to train and

Read More & Connect

Interactive version: blogs.drix10.com

Written by Drishtant Ghosh (Drix10), a technical founder and engineer working across AI systems, developer infrastructure, and cybersecurity.

Top comments (0)