sabarinathan

熱画像の「ぼやけ」を鮮明に再現──クーガーの超解像AI技術が世界最高峰CVPR Workshopに採択 prtimes.jp/main/html/rd/p… via @PRTIMES_JP @couger_jp @couger_en

This 2-hour Stanford lecture breaks down how models like ChatGPT and Claude are actually built, clearer than what many people in top AI roles ever get exposed to. Save this and set aside two hours today. It might end up being the most valuable thing you learn all week.

You can now train Qwen3.5 with RL in our free notebook! You just need 8GB VRAM to RL Qwen3.5-2B locally! Qwen3.5 will learn to solve math problems autonomously via vision GRPO. RL Guide: unsloth.ai/docs/get-start… GitHub: github.com/unslothai/unsl… Qwen3-4B: colab.research.google.com/github/unsloth…

Simple idea, but impactful for future model design Paper: arxiv.org/abs/2603.15031 @Kimi_Moonshot

🚀 Interesting paper: Attention Residuals (AttnRes) A small but powerful idea to improve Transformers 👇

Why this matters: Better feature usage More stable training Stronger reasoning performance My takeaway: We already use attention across tokens… Now extending it across layers makes a lot of sense

This paper proposes: 👉 Use attention across layers Instead of equal addition, the model learns which layers to focus on

Today’s Transformers use residual connections → All previous layer outputs are added equally But… not all layers are equally useful 🤔

In practice, many production AI systems will likely combine both approaches depending on the use case. The key is choosing the right architecture for the problem. Good read: junia.ai/blog/perplexit… #AI #AIAgents #LLM #MachineLearning #AgentSystems

🚨 Interesting shift in the AI Agents ecosystem Recently I came across an article mentioning that Perplexity is moving away from MCP (Model Context Protocol) and relying more on direct APIs for its agent workflows.

That said, MCP (Model Context Protocol) is still a powerful idea. It standardizes how AI agents interact with tools and services. But real-world systems often prioritize simplicity and performance.

Cost control becomes easier too. You can measure every call, implement caching strategies, and optimize usage without additional tool-layer overhead.

Direct APIs also give predictable performance. You control: • Payload size • Serialization • Request patterns Which helps avoid unpredictable agent behavior.

Another advantage: observability. With direct APIs you can easily use: • HTTP logs • Metrics • Tracing • Retry strategies This makes production debugging much easier.

Direct APIs also improve security. Instead of routing everything through a tool layer, systems can use direct authentication with the service, making access control easier to manage.

Why direct APIs are often preferred in production AI systems: • Lower token usage • Less tool description overhead • Smaller context size This helps reduce LLM cost and latency

Building production AI systems teaches an important lesson: Sometimes direct APIs are more practical than complex agent tooling. Here’s why 👇

I spent more test time compute and realized that my micrograd can be dramatically simplified even further. You just return local gradients for each op and get backward() to do the multiply (chaining) with global gradient from loss. So each op just expresses the bare fundamentals of what it needs to: the forward computation and the backward gradients for it. Huge savings from 243 lines of code to just 200 (~18%). Also, the code now fits even more beautifully to 3 columns and happens to break just right: Column 1: Dataset, Tokenizer, Autograd Column 2: GPT model Column 3: Training, Inference Ok now surely we are done.

@AmazonHelp @AmazonHelp @amazonIN @AmitAgarwal @ajassy Please check my message in DM and resolve the issue soon. I would like to order a product for my family.

@sabarinathan_7 We're sorry to know about your issue with the account. We would like to escalate this for you. Please contact us via DM using the 'Send us a Private Message' option below, without sharing your order or account details, and we will do our best to assist. -Hafeez twitter.com/messages/compo…

@AmazonHelp @amazonIN @AmitAgarwal @ajassy My order 405-4873672-8028367 was canceled and my account was restricted to digital purchases only, citing “multiple returns/refunds.” I’m requesting clarification, as this does not seem accurate. I’m an active Amazon Prime member.

@AmazonHelp @amazonIN @AmitAgarwal @ajassy @AmazonHelp @amazonIN @AmitAgarwal In Oct 2025, I received an email from Amazon Canada about alleged multiple refunds and replied clarifying it was not my account—but never received a response.