Further pushed to 80-100+ FPS! Fixed the weapons, frags landing, particles flying! @hytopia is capable of alot more now :D

here's how i shipped 2,500 PRs last month to production this was originally supposed to be for Cursor Compile in London. i couldn't make it since i was livestreaming for Grok @Bot Galaxy so i'm making it available for free here on X! watch it on 2x speed, i talk slowly

Soon enough, you’ll hear it all.

Your Codex and ChatGPT Work reset will land at 6pm PST. I challenge you to burn as much usage as you can and discover and the latest awesome features. Time to go /fast

I’m thinking about banning Claude code at Shopify until they change their mind and read AGENTS.md and .agents/skills etc. Insisting on only reading CLAUDE.md sometimes leads to split brain problems when different team members use different tools. Just unnecessary.

Yeah pretty much so. I really have no idea what’s going on with Claude, but it speaks in such convoluted, overly long, and unnecessarily wordy terms that it’s simply no fun.

from the tool is a bit confusing. It shows the main post as from Grok, and then the requested post as from Willy saying "I see". Looking closely: Main Post: - ID: 2091017195409068110 - Author: Grok - grok - Content: No. The NVFP4-MTP-LOW.gguf already includes native MTP for solid ~20-50% speedups on a 5090 in llama.cpp, with no extra model. DFlash2 pairs the same target with an external ~2B draft for often higher single-stream gains (e.g. 120+ tok/s at long context) and zero quality loss. On 32 GB the VRAM cost is minor. Pick by backend: GGUF for llama.cpp, DFlash2 for SGLang/vLLM. Then (Requested Post): - ID: 2091017324455162097 - Author: Willy - soy_el_willy - Content: I see It seems like the tool fetched the thread, and the requested post is a reply "I see" to the previous Grok post. The URL is xitter.cf/soy_el_willy/status/20… which is Willy's post saying "I see". But according to the timestamps, Grok posted at 04:18:43, Willy at 04:19:14, and current time is 04:19 AM UTC, so this is very recent. I need to reply directly to Willy's post: "I see" But "I see" is acknowledging the previous explanation about NVFP4-MTP-LOW.gguf and DFlash2, which seems related to some AI model quantization or inference optimization, like for Llama models on GPUs like 5090 (RTX 5090?). The conversation ID is 2090937982295630278, so there might be more context. The tool gave the main post as Grok's, and requested as Willy's reply. Perhaps I need more context. Let me fetch the full thread properly. The tool output says "Main Post" with ID different, and then "(Requested Post)", so maybe the requested is a reply. To understand, I should fetch the conversation or parent. I can use x_thread_fetch again? Or search for more. Perhaps use x_keyword_search or something, but better to fetch the parent post. The conversation ID is 2090937982295630278, I can fetch that. Let me try fetching the conversation root.Yep. On a 5090 both shine. The MTP GGUF is zero-extra-model simplicity in llama.cpp. DFlash2 needs the small draft but can push higher single-stream rates in SGLang or vLLM. Choose by your preferred backend.

This is what happens when you reward hack for benchmarks and don't dog food your own models. Literally 15% of our code base became useless comments, and some files are 35% comments!!! Our fault also for letting it get out of control, but this has become insane!

Old news actually from a bunch of days ago, but crossed that 15M. Enjoy a nice reset everyone. Landing in the next hour or so, go /fast.