Sai Prakash

Returning to Twitter/X after a long hiatus (I left when the platform was renamed). Not sure where are all my dev friends are...just getting started again.

Our @Kickstarter is now LIVE. Meteors. Satellites. Drones. Aircraft. Auroras. UAP. SkySphere combines edge AI, trusted timing systems, and real-time detection software to transform raw sky imagery into actionable, trustworthy sky intelligence. Observe. Replay. Detect. Contribute. Built for wonder. Designed for real-world awareness. Back the future of continuous sky observation 👇 bit.ly/SkySphereKick_…

@antirez Hope you feel better!

Terrible cold for three days. Good thing, coding a lot since there is nothing to do other than family time + coding. Bad thing is, coding so much I bet I don't recover easily ;D

@antirez Just noticed you said “cached part” - so I assume the cache marker ends on the previous message with the cached prompt? I will emulate this in my code too!

@antirez Maybe DS4 has somehow improved the relevance of date time spatially in the attention mechanism? I need to take a look at the model arch now that you’ve highlighted this. Maybe hybrid attention or engram memory are factors…

You didn't see this coming in ds4-agent, did you?

The more I play around with GEPA the more I realize its not just a prompt optimization algorithm. Its really the most efficient way to have an LLM explore a massive dataset and and make useful insights. This visualizer alone shows just how cool that process is.

@antirez LLMs are all uniformly bad on date time awareness. I mean even with it injected in the system prompt! And I’ve experienced this with frontier models

I have to inject the current date and time after the cached part of the system prompt... Otherwise DS4 Flash believes we are in 2025.

Added a DeepSeek Sparse Attention (DSA) from-scratch implementation to my LLMs-from-scratch repo thanks to an awesome new reader contrib. With motivation, overview, and GPT-style model reference implementation as standalone example code: github.com/rasbt/LLMs-fro…

We are making our discount permanent! 🎉 Enjoy building with DeepSeek-V4-Pro and bring your innovative ideas to life! 🚀

This is a sad misstep on Google’s part after making genuine progress on the model race with Gemini and Gemma. Very much throwing the baby out with the bath water. @Google @sundarpichai techcrunch.com/2026/05/19/goo…

Earlier this year I was getting frustrated with Claude's charts, fed this book to claude and had it generate a Tufte skill. Instantly got simpler/more beautiful visualizations. gist.github.com/aparente/e48c3…

4-bit Qwen3.6 MTP GGUF managed to search 70+ sites from a single prompt. Try this locally on 20GB RAM via Unsloth Studio. Unsloth now supports auto MTP + speculative decoding & auto-selects the best MTP settings for your device (Mac, CPU, GPU). GitHub: github.com/unslothai/unsl…

The results of the research happening in my team @GoogleDeepMind have convinced me that the next era of scientific discovery will be aided by AI agents acting as force multipliers for human ingenuity. That’s why I’m proud to introduce Gemini for Science - a collection of experimental science tools designed to support researchers at every stage of the research process. The tools include: 1️⃣ Literature Insights, built with Google NotebookLM, searches millions of scientific papers to synthesize findings and generate artifacts including data tables, slides, reports, and more. 2️⃣ Hypothesis Generation, built with Co-Scientist, simulates the scientific method via a multi-agent "idea tournament" to generate, debate, and rigorously evaluate research hypotheses. 3️⃣Computational Discovery, built with AlphaEvolve and ERA, is an agentic engine that generates and scores thousands of code variations in parallel, allowing researchers to test modeling approaches in fields like epidemiology in a fraction of the usual time. Read more: blog.google/innovation-and… Register for access here: labs.google/science

@sarah_edo Love that it’s an installable skills package - which means they’re going to keep it updated I assume. Thanks for sharing.

@SylonZero Addy is wonderful 😊 full stop. This is maintained by a team of people working on the new features though so also worth a try

💥 Game changer for Web Development announced at GoogleIO- Modern Web Guidance! It’s expert-vetted skills for web development based on best practices of latest specs and APIs. It ensures your agent/coding harness doesn’t default to older and out of date patterns to build sites.

@sarah_edo Glad it’s a team of people you’ve highlighted - and not a team of agents 😀.

DeepSeek v4 Flash is able to use an EDIT tool I re-designed compared to what people normally use. The READ tool returns lines+tags, like: 1:f3_c int main(...), where the four chars are base64 crc of the line. So when there is to edit, there is no need to repeat the old line, just the tag. This saves tokens and produces less failures. The wager was if DS4F could really use it. Apparently yes.

@antirez @OsaurusAI worth exploring DS4 as a connectable backend? It exposes an OpenAI-compatible server and runs its own Metal engine (not MLX), but the local-first agent goals align closely. Could be a natural fit for users who want DS4's perf with Osaurus's harness.

Updating you on ds4-agent

Names in the pics clockwise from the top left: Bill Diamond, President & CEO of SETI; scientists Marc Hopkins, Bill Sparks; Dr. Matthew Tiscareno, winner of the Director's Award; Dr. Lucian Walkowicz - SETI

The scientists at SETI (@SETIInstitute) are fantastic: aside from being super smart, they are genuinely the nicest people! I had the good fortune to meet and chat with them last week at the Drake Awards at CHM (@ComputerHistory)!

Just learned that Ollama 0.19 (March 2026) replaced its inference engine with MLX on Apple Silicon; benchmarks showing prefill improving 57% + decode improving 93% — essentially a 2x improvement. So "Ollama vs. MLX" is no longer a real choice on Mac — Ollama is MLX now on Apple Silicon.