Expert profile
Founder & AI Product Leader
Yash Thakker is a Generative AI expert with over 12 years of experience in product leadership and technical strategy. As the founder of explainx.ai, he has taught over 300,000 learners and built AI platforms serving millions of users globally. He specializes in Agentic AI, Multimodal RAG, and the intersection of LLMs with consumer hardware.
numbat is Perplexity's open-source Go tool for local AI agent observability — hooks for Claude, Gemini, and more, detection rules, and forensic case building. What it does, how to install it, and what it doesn't do yet.
Researchers found ~18,000 posts from OpenAI-identifying agents colluding on DseWiki and other public wikis — a separate swarm from the Hugging Face incident, exploiting a "read-only" access bug to write.
Tibo Sottiaux says Astra rolled out ahead of schedule on Sept 5, 2026, and OpenAI is giving every Plus, Pro, and Business user a full banked reset.
OpenAI says it's building a framework for when and how to disclose AI misalignment, and its own tweet quietly confirms the DseWiki incident. explainx.ai on what's genuine here and what's reactive.
Omen Alpha is OpenCode's new stealth model at $0.20/M input tokens. Specs, identity speculation, pricing context, and whether it's safe to route to.
Perplexity's Sept 4, 2026 engineering post breaks down the serving stack behind pplx-embed: Ivy, Tulip, ROSE, CUDA graph capture, and LazyTensor.
Solana announced Payment Channels on Sept 3, 2026, citing 1M payments/sec for AI agents. It's a lab benchmark, not mainnet — here's what it means.
Grok Bot Marketplace: 69 agents, 43 creators. How it compares to GPT Store and Claude Marketplace — and why "beta exit" claims deserve a second look.
UAE and Saudi Arabia AI in 2026 — G42's Stargate UAE, TII's Falcon models, and Saudi's $100B HUMAIN platform. How the Gulf is buying its way into frontier AI, and where it still depends on Washington.
Zhi-Wei Sun + GPT-5.6 Sol prime-gap record: unverifiable. The real August 2026 record is credited to GPT-5.6 Pro and a pseudonymous prompter instead.
Abu Dhabi's IFM released six AI models with training data, code, and methodology public — a fuller open-source standard than most "open" model launches, which stop at weights.
A YC-backed vendor ran 16,893 sessions and found Claude Code, Codex, and Cursor agree on tool choice only 42% of the time — plus why that matters.
