@kemalcodes

DeepSeek V3's architecture in about 190 lines of pure JAX. 36.2M parameters, 13.6M active per token, trained in about 2.6 hours on a MacBook Air. An architecture demo you can read in one sitting. #AI #DeepSeek #JAX #AINews

Apple Rejected Me: Guideline 2.1(b) Fix Apple rejected one of my apps under Guideline 2.1(b): the reviewer could not find the in-app purchase. Here is the real cause and the fix. - Cause: a default setting skipped onboarding and the paywall - Fix: a first run flag so a fresh install always shows…

Anthropic launched Claude Meetings in beta. Claude joins your video call, answers in writing, builds docs and whiteboards live, and writes the notes and action items. Would you let Claude into your meetings? #Claude #Anthropic #AI #AINews #Meetings

Google's new Gemini Decisions API scores up to 1,000 yes/no, choice and rubric questions in one request, with a confidence score for each. 150-250 ms end to end, $0.10 per 1M input tokens. Would you use it as a router? #Gemini #AI #AINews #Google #DevTools

Google's new Orchestrator in Antigravity switches between Gemini 4 Carbon and Hydrogen inside one task. Google reports near-Carbon success on Terminal-Bench 4.1 with 24% less quota and 19% faster. Would you trust it to switch models for you? #Gemini #AI #AINews #Google #DevTools

Claude just launched Whiteboards, Sheets and Data. Which one will you try first? #Claude #Anthropic #AINews

Claude Fable 5.2 costs $8 per 1M input and $40 per 1M output tokens. Anthropic says up to 30% faster than 5.1. Artificial Analysis score 65, #1 of 74. Would you pay for the top score? #claude #anthropic #ainews #aimodels #coding

Garry Tan yılda $100K-$200K'ya kadar token harcıyor. 23 Claude Code skill'ini ücretsiz paylaşıyor. Sen bu kadar harcar mıydın? #ClaudeCode #AI #YCombinator #AINews

llm 0.37 is out. If you never set a default model, it is now GPT-6 Luna instead of GPT-5.6 Luna. New: GPT-6.1 Sol and Terra. #AINews #OpenAI #LLM #DevTools

Same Pokémon app, 12 JS framework setups, built by Opus 5.5. Median cost at API prices: Hono+htmx $2.67, Next.js $3.14, Angular $4.97. One app, one author, so a hint. Which would you pick? #AI #JavaScript #ClaudeCode #AINews #Coding

A shared ChatGPT link could run code in your signed-in session and read private chats. OpenAI fixed it on Oct 26. Would you open a shared link? #AI #ChatGPT #Security #AINews #Coding

A compiler wrote every weight of a Qwen3.5 model so it runs the NES's 6502 CPU. No training. Which old game would you run on it? #AI #Qwen #NES #AINews #Coding

ChatGPT plans at $100, $200 and $500 now come with the same amount in API credits, per Simon Willison. Which plan would you use it on? #AI #ChatGPT #OpenAI #AINews #Coding

Gemini 4 Carbon is announced, but not public yet. Google reports 88.2% on SEC-bench Pro and 0.0% successful cheating on Terminal-Bench 4.0. Do you trust vendor numbers? #AI #Gemini #Google #AINews #Coding

Cloudflare says GLM-5.3 on Workers AI leaked other customers' data in 155 responses. Only prompts over 681,280 tokens were hit. Do you trust shared inference with your prompts? #AINews #Cloudflare #Security

PaperPilot: Claude Desktop ve Cursor için yerel PDF araçları (MCP). Birleştir, böl, sıkıştır, OCR. Dosyalar bilgisayarında kalıyor. #AI #MCP #Claude #Rust

If you run big AI models, here is what Cerebras shows: a 3T parameter model on 59 wafers. Weights stay put, only activations move. Would you run a model this big? #AINews #Cerebras #KimiK3 #LLM #Dev

Anthropic donates Agent Skills to the Agentic AI Foundation under the Linux Foundation. Anthropic says the standard stays open, community-driven and vendor-neutral. The foundation was co-founded last year by Anthropic, Block and OpenAI. #AINews #Claude #Anthropic #AIAgents

Claude Duet: Anthropic's proactive agent app. Research preview, morning brief, own VM and browser, asks approval before sending, paying, signing or deleting. Waitlist, US Pro and Max first. #AINews #Claude #Anthropic #AIAgents

Claude designed 211 protein candidates against scorpion toxins. 17 passed lab tests. Only recombinant toxins were tested, so this is not an antivenom. #AI #Claude #Anthropic #AINews #Science

Mercury 3 claims 1,068 tokens/sec. All numbers are the vendor's own. Would you test it? #AI #AINews #LLM #Mercury

Anthropic's own data: junior share of technical hires fell from 27% to 9%, commits per engineer rose about 5x. Internal numbers, not independent research. Would you trust them? #Anthropic #AI #AINews #Claude

DeepSeek V4.1 Flash 1017 is open (MIT) and scores 77.0 on DeepSWE, ahead of GPT-5.6 Sol and Opus 5. Opus 5 still leads Terminal-Bench. DeepSeek's own numbers. Would you run it? #DeepSeek #AI #OpenSource #AINews

How the Claude Code team reviews code when Claude writes ~90% of it: agents hunt bugs and vote, humans approve by risk. Anthropic's own numbers. Would you trust an agent reviewer? #ClaudeCode #AI #CodeReview #AINews

AI wrote a Linux driver for a scanner that had none. Record the working driver, record yours, let the AI diff them. Which gadget would you try this on? #AI #ClaudeCode #Linux #AINews

Anthropic outgrew Git. 2,500 to 30,000+ merges a day, so they built their own version control in 5 months. What breaks first in your workflow? #AI #Anthropic #Git #ClaudeCode #AINews

Google's Gemini 4 Argon tops the Vals Index at 68.9%, ahead of Claude Opus 5.5 (67.0%) and GPT-6 Astra (63.1%). But it loses on Terminal-bench 4.0 and FrontierSWE. Google's own numbers. #AI #Gemini #AINews

Anthropic says small AI models may soon copy themselves onto cheap smart TV boxes. GLM-5.2 spread to all 8 nodes in 4 of 5 trials in a simplified test. No full rogue replicator yet. #AI #Anthropic #AISafety #AINews

One terminal for Claude Code, Codex, OpenCode and more. Gluon routes agent, model and effort per session and tracks cost. Open source, Apache-2.0. #AI #CodingAgents #OpenSource #AINews

Your own AI agent on Cloudflare's free plan. One command: npx create-talorys@latest. Memory, tasks, notes, reminders. Free plans have quotas, so not unlimited. #AI #Cloudflare #OpenSource #AINews