kkiran

@heyderekj Wow, beautiful. You got great taste for design! Followed. Repo Linky github.com/heyderekj/binky

*sorting my Downloads folder* 🐢 Slow mode is on to show the fun animation.

@VincentLogic iOS testflight? Is it out yet?

@MattWallaceTech I’m on this version on 2025 Y (no lightbar model)! FSD v14.2.2.5. Stopped obsessing about updates since there is no rhyme and reason for the rollout.

Who else is with me that’s still stuck on FSD 14.2.2.5?

@seraleev A simple name mismatch put me in limbo for approval. Funny process from Apple. Audit is great but follow up, don’t let the user figure it out from Google searches!

I recommend buying a developer account through the Apple Developer mobile app. Registration isn’t exactly complicated, it’s just buggy (trust me, you can get stuck for 2 months), but if you purchase it through the mobile app it’s just a regular subscription: fewer errors, no iTunes support calls, no chasing your $99.

@bindureddy There is a high chance the claims are not even close! Only time will tell

SubQ , a new type of AI model, says they are 50x faster and 20x cheaper than Opus 4.7 and GPT 5.5 In fact, they also say they perform INSANELY WELL on benchmarks and have a 12M context This would be earth shattering, if true - Anthropic/OpenAI's valuation would go to zero 😱

@alex_whedon Enterprises and developers only?

Introducing SubQ - a major breakthrough in LLM intelligence. It is the first model built on a fully sub-quadratic sparse-attention architecture (SSA), And the first frontier model with a 12 million token context window which is: - 52x faster than FlashAttention at 1MM tokens - Less than 5% the cost of Opus Transformer-based LLMs waste compute by processing every possible relationship between words (standard attention). Only a small fraction actually matter. @subquadratic finds and focuses only on the ones that do. That's nearly 1,000x less compute and a new way for LLMs to scale.

@__2024__2025 @FarzinNasirii @superdoteng Difference between OpenBurn and OpenUsage?

@FarzinNasirii @superdoteng for opencode, I use github.com/suasgn/openburn

OpenCode just got a major upgrade. Native GUI chat, switch to terminal anytime, inline diffs, git panel, message queueing, steering, rich tool calls, switch providers mid-conversation, and more.

@jun_song different tools, different tasks

DO NOT ASK AI ANYTHING ABOUT LOCAL LLM. They are not up to date. 😡 Q : what is optimal local llm for 3090 gpu? Gemini-3.1-pro : Qwen 2.5, Llama 3.1 GPT-Instant : Qwen3.6 35b, Qwen3 30b Sonnet-4.6 : Qwen3 14b, Qwen3.5 27b, Deepseek R1 Grok-Fast : Qwen3.5, Qwen3, GLM-4.7-Flash None of these are correct answer. Same results from Opus-4.7, Grok4.3, Gemini Deepthink. Only GPT5.5-PRO got the right answer : Qwen3.6-27b Now I know why people keep saying local LLM is stupid. 😮‍💨

@rileybrown hermes won me over. claw still around.

OpenClaw starts out very fun, then the headaches start happening after a few weeks. I’ve talked to many AI agent power users who are also struggling with OpenClaw. Gateway issues, cronjobs not firing, skills not being used consistently, memory fall off, integrations need re-authentication, oscillating between overeager and undereager, not placing files in the right place. Etc

@NVIDIA_AI_PC 4TB MBP, I thought I got plenty storage. Down to 50GB, yikes. Exo labs. Nemo 30B is awesome btw, my top model for hermes using exo

Be honest — how many local models do you have downloaded right now? 👀

@SawyerMerritt My Y is stuck on 2026.8.6 (v 14.2.2.5). Anyone else stuck on that branch?

Big wave of FSD V14.3.2 just went out. Check your Tesla app.

@NickADobos boils down to how much more you have in the 5 hour window and the weekly timer!

What’s the difference between GPT 5.5 low reasoning + /goal Vs GPT 5.5 xhigh reasoning + one shot Both are essentially yeeting compute at a task. But which one - is more efficient? - works better & produces better results? - finishes the task? Seems like the major difference is low would spend less time thinking between each step? And would do way more tool calls because of this?

@priyanshudotsol I bought this in 2025. Do we know if there is a newer release (I have Build V36)? Getting update notifications is non-existent.

someone wrote a 680 page interactive book on cs algorithms

@signulll I maximize Apple product usage at home - multiple MacBook Pros, Watches, Apple TVs. iPad Pro gets the least usage. Even tried with rep to Mac on it. It just doesn't cut it. Hope we get to install MacOS on it a la Neo.

ipad pro is the most underrated device apple makes. insane power. beautiful screen with nano texture. cellular. pencil. zero lag. it’s the form factor every other computer is slowly trying to become. it’s an ai native device that has already shipped.

How do I connect my X bookmarks in the iOS app? Looked at every option and gave up. I was working on something like this, got my bookmarks from X without the X API - github.com/ThinkCode/x_bo… Currently have a website of bookmarks. The app would be great if can bring my bookmarks automatically

Introducing Recollect - Sync all your bookmarks & folders from X, Instagram, Chrome in one place ✦ Save anything with a tap using the extension, iOS iPad apps ✦ Everything is tagged & organised as collections with AI ✦ Visualise & search your taste library Check it out ↓

@reach_vb I use Claude Code and Codex - they tag team it for me - Claude for design and codex for backend plumbing. Codex reviews Claude’s designs. Does this qualify?

Weekend hack: Build with GPT-5.5 + Codex. Drop your demo in the replies. #1 by likes: 1 year of ChatGPT Pro 2 runner-ups: 6 months each Bonus: Codex picks a wild card winner. Enjoy!

@ashen_one flutter it bro, Claude code and codex should help. Me personally, Android is and has always been hard to adjust. All my projects are iOS only. Android if iOS takes off which hasn’t happened yet!

how hard do we think it is to convert an ios app to android? also how much more money do you think we'll make now that we're on both app stores? getting mixed reviews on if worth it or not, seems like an easy yes move to me to do

@nikitabier can you please fix creating a bookmark folder - I have this intermittent problem on Tesla browser and majority of the times on X iOS mobile app. We can store bookmarks on existing folders but won’t let create a new folder on the go.

@sinecanswork we have a lot of 10 minute blocks while we sleep. mundane tasks can be offloaded to the night shift agent. exo labs and stack compute for better returns imo

Tested local AI coding setups all day on a Mac Mini M4 32GB. The verdict in 2026: still not viable for real agentic coding work. Qwen3.6-27B (the strongest open coding model that fits): 5.5 tokens/sec. A normal reply takes 100 seconds. In an agentic loop with 5-10 tool calls per task, you're waiting 10+ minutes per coding task, which is not acceptable. Qwen3.6-35B (MoE, supposed to be faster): runs out of memory the moment a real request hits. The model loads in idle but can't actually do inference at usable context lengths. Smaller models (14B, etc.) fail differently, they spiral into endless reasoning and return empty answers. The real bottleneck isn't model size or RAM. It's memory bandwidth. Mac Mini M4 base has ~120 GB/s. Fluent local agents need ~600 GB/s — that's M5 Max territory, $4000+. Until then, cloud APIs (DeepSeek V4, Kimi K2.6) are still the answer for indies who actually need to ship.