I'm done with Bonsai 2 27B. This model has taken enough of my time and patience. After the whole argument, I ran 12 OMP sessions on my 4x RTX 3090 rig using PrismML's launcher settings and the reasoning budget nisten asked for. 1 session per GPU, 128K context, no follow-up prompts. Same ternary model, 2 weight packings, 3 KV caches, 2 prompts. Here are the results, the exact prompts, the memory guide and the settings. Do whatever you want with them. Big thread 🧵 First, the extended prompt. 3 of the 6 runs produced usable scenes. PQ2_0 with Q8 KV made the garden I liked most: 42.0 minutes, 69,240 output tokens. PQ2_0 with Q4 KV + bias spent 4h54m and 418,141 output tokens producing an empty sky. I stopped it myself. FP16 KV also failed to produce a usable scene after 151.7 minutes and 210,512 output tokens. The videos show every configuration, including the failures. Output tokens include reasoning; time is the full agent session. Exact extended prompt: Design and create a very creative, elaborate, and detailed voxel art scene of a pagoda in a beautiful garden with trees, including some cherry blossoms. Make the scene impressive and varied and use colorful voxels. Use whatever libraries to get this done but make sure I can paste it all into a single HTML file and open it in Chrome. I'd call the model tolerable for its size. PrismML's marketing still pisses me off. Simple results and the memory guide below.

Day 151 of My Founder Journey. Today I registered a domain for one of my new app ideas: postpilot(.)app 🚀 Over the last few days, I started posting my daily videos not only on X and Telegram, but also on Threads, YouTube, and Instagram. And I quickly realized how much time it takes to publish the same content everywhere. So I’m building a tool to solve my own problem. The idea is simple: → Generate YouTube thumbnails automatically. → Crop videos for Instagram Reels, YouTube Shorts, and other platforms. → Schedule and publish everything from one place. I’m keeping it mobile-first and plan to share the first version through TestFlight. My goal is to build it in 3-4 days. I’ll show you as soon as it’s ready. See you tomorrow.

Someone trained Qwen3.8-27B to stop chatting like an AI assistant respond like a normal person. The local quant is only 15.1GB. Qwen3.8-27B-Humanlike-Chat 2.0 In the creator’s blind test which tests what response was actually human 🤖 Official Qwen3.8-27B → 15.1% ❤️ Humanlike 2.0 → 23.5% It was trained on real + synthetic conversations, then on-policy distilled so it learned the behavior without needing a “talk like a human” system prompt. It didn’t forget how to reason accurately either 🛠️ tool calling 💻 code 📝 instructions 🧠 reasoning 👁️ optional vision 📚 262K context Local versions 💾 IQ4_XS → 15.10GB 💾 Q4_K_M → 16.56GB 🦙 llama.cpp 🖥️ LM Studio / Ollama 🔓 Apache 2.0 What I'm saying is it isn't another uncensored model. It makes a Local AI model feel less like an assistant and more like somebody you're texting. 🔗 Link in ALT.

American decided to go back to college. She says it’s all liberal propaganda being taught “So at 43, I'm back in school trying to get my bachelor's degree. And I just have a question. Is it every college that spews their liberal bullsh*t in your face and tries to brainwash you? Or is it just my school? I really want to know.” She’s right it’s all liberal propaganda - Harvard faculty is just 2% conservative - A study of 40 university found a 15:1 ratio of liberals to conservative professors - Another National Association of Scholars study found 78%+ of departments at top liberal arts colleges contained zero republicans It’s mass indoctrination

Discovered a wallet address that bought $431.74 worth of $ZEVIO, acquiring a total of 15.1M tokens. It has already sold for a profit of $6.17K, with the remaining position worth $18.5K, showing an unrealized profit of +$24.19K. More info: Win Rate: 38.46% Total PnL: +$32.3K (+145.57%) Bal: 1.77 ETH ($4.58K) Wallet address:

Qwen3.8-Flash API is live on QwenCloud. 262K native context, extensible to 1M, and priced to scale. About Price⬇️ Input: $0.15 / 1M tokens Output: $0.47 / 1M tokens Cache hit: $0.016 / 1M tokens Get your API key and start building： #QwenCloud# #Qwen#

Finally secured all 151/151 pokemon from the 1997 bandai carddass series All strictly PSA 10s Took me an entire year to catch them all Feels like completing the Pokedex I already told you why this set is the purest, most valuable, iconic, curated ever created See you at MoMa

🚀 Codex CLI 0.151.0 is out! ⏱ Grace period for optional MCP server tool discovery 🔌 Extension hooks for MCP tool result interception 📦 Per-repo plugin catalogs with marketplace checks Changelog: