Grok3 is goated it’s just so much smarter that gpt o3 mini. Thanks Elon Musk

It’s been a day since I installed IOS 26, They definitely fixed the photos, Phone and iMessages app. The call filtering and screening features are also great feature to have especially here in India but the other the UI elements not a huge fan

Finally got back to writing after a long break! It's about Native Sparse Attention the hottest topic in town! Read more about it here: adityakm24.medium.com/native-sparse-…

Another weekend, another learning project! Here is an implementation of Llama3.2 from scratch - colab.research.google.com/drive/1_QYB6u_…

Hello there! SpokespersonNavy

Went to a studio to get a photo printed and the guy was generating photos using Perplexity 💀

Why is no one talking about this? This is why I don't use an AI browser You can literally get prompt injected and your bank account drained by doomscrolling on reddit:

RAG is not dead! However, we are in an interesting phase of exploring unique ways to index and retrieve information. This vectorless RAG framework uses a tree structure index in place of vectors. Reasoning models will enable methods that mimic human-like search. Early days!

How did a team of 4 research engineers beat $100B labs like OpenAI and Anthropic in establishing the best coding agent? It starts with having a killer engineer on your team. In our case, we have Abhay Singhal 🧵

RAG-Anything: #1 on GitHub Trending! A huge thank you to the open-source community for your incredible support—clearly, the demand for multi-modal RAG is immense! Easy-to-use: github.com/HKUDS/RAG-Anyt… Try RAG-Anything: The ultimate All-in-One RAG Framework 🌟 Key Features of

Just published a new blog diving into a fascinating (and slightly controversial) topic in AI research: Hierarchical Reasoning Models AKA HRMs. Can a tiny model really outsmart the giants? adityakm24.medium.com/how-hierarchic…

The exponential takeoff in action. And we need new Evals soon!

Wow! CUDA IS NOW WORKING IN MAC! Got 127 Tokens per second running Qwen3.5 35B. Magic!

God speed! NASA Artemis

Kimi is cooking! Kimi K2.6, an open-source model boasting 4,000+ tool calls over 12-hour runs, parallel agent swarms!🤯 Already beating oUS 4.6 AND got-5.4-xhigh

My life atp

Someone asked me what will I do if I lose my software job due to AI.... My Plan: 1. No job so I get +3000 2. Unemployed wife, +3000 3. Wife will get lovely behan +1500 4. Parents will get pension, +6000 (3000 each) 5. Wife gets pregnant +21,000 6. Will register 1sq ft land on

An algorithm just wiped 7% off Micron's stock. 🤯 It's called TurboQuant by Google, and it solves the biggest bottleneck in LLM inference: the KV Cache. Full story here: adityakm24.medium.com/turboquant-an-…

My weekend plans are ruining #OpenAI just released GPT-5.5!