kfant

CT imaging results of the alien / E.T. mummies from Mexico / Peru #UAP #ufotwitter #Aliens [vid talk @QuanticaScience ]

@Usoppu 🚀

Ex-CEO of Singapore-based crypto platform Hodlnaut Faces Up to 20 Years in Prison After Fraud Charges > Zhu Juntao, 36, former CEO of crypto platform Hodlnaut, was charged with six counts of fraud today on 26 May 2026. > Launched in April 2019, Hodlnaut allowed users to deposit BTC, ETH and stablecoins to earn interest. > Zhu graduated from SMU, worked at Credit Suisse, before co-founding Hodlnaut in 2019 with Simon Lee. > At its peak, the platform served over 30,000 users worldwide, offered up to 10% APY, and managed $750 million. > Hodlnaut lost $189.7 million when the TerraUSD (UST) stablecoin crashed in May 2022. > The company had secretly put $317 million of user funds into Terra's Anchor Protocol. > No official statements before the crash disclosed this massive exposure to UST. > Zhu allegedly ordered staff to lie on Telegram, email, and X, falsely claiming Hodlnaut had no UST exposure. > The platform froze withdrawals in June 2022 and later collapsed into judicial management. > It owed users $281 million but held just $88 million in assets - a $193 million shortfall. > Zhu faces up to 20 years in prison, a fine, or both if convicted.

@theblackvault applause for your appeal

When the Pentagon refused to search for emails about "Immaculate Constellation" in the records of Maj. Gen. Derek J. O’Malley, I appealed. Today, I won that appeal, and a search will take place. Always Appeal.

@Osmg900 @s8box shaje taht a$$ baby

Dev Log — "You're not alone" P1, Music: Annihilation OST - The Alien @s8box #sbox #Sandbox #GameDev #IndieGame #IndieDev #Cockpit

@KyleHessling1 @arlogilbert what are good params for llama server ? with mine its dog s..t

@arlogilbert Here's an MTP version to try! huggingface.co/Jackrong/Qwopu…

BREAKING! Qwopus 3.6 27B is LIVE! Thank you for your patience on this one, but I believe you'll find the wait was worth it! We've benchmarked this thing up and down, verified that it holds at least a 75.25% (152/202) in the initial 202 SWE bench solves. Not a full run of 500, but it shows the agentic coding quality from the original 27B is retained while adding all of the additional Qwopus benefits across many domains. As always, Jackrong is absolutely cooking here! COT quality has improved significantly through the inversion techniques from our Negentropy proof of concept. It also went through thorough curriculum training. You can check out the MMLU pro benchmarks on the model card, but it improved a whopping 10 points over the base model in physics, as well as meaningful jumps in Chemistry, business, and computer science. However, the best part is that I was able to build an entire survival shooter game using this local model entirely. I genuinely was blown away by the results, which you can play right now on my HF space (link in comments below). "Qwopus Commander" was completed in 9 turns of Qwopus 3.6! To test the new long context training, I made it re-output the entire 3000+ line program each turn, and it would make fixes and add features that I requested in large prompts, while perfectly replicating the entire rest of the game from context. What's more is that I did it all at Q8 KV cache quantization, and never had an issue over the entire 303k token run! IMPORTANT: Run it at --temp 0.75 to 1. Mess with it in that range for your use case. Higher temp actually lets the fine-tune shine and be exploratory and is also more stable. Swe Bench was run at temp 1, the game was built mostly at 0.8! We're so blessed to have all of you here and using the models! The support means so much! Please let me know what you build with it in the comments! Or if you have any issues getting it up and running, I will try my best to get back to you! Looking forward to seeing what you legends produce with it this weekend! huggingface.co/Jackrong/Qwopu…

@multimodalart @StabilityAI it passed loud wet fart test

Stable Audio 3 by @StabilityAI is just out It mainly comes with 3 open source variants: - Stable Audio 3 Medium (2B) - Stable Audio 3 Small (0.6B) - Music - Stable Audio 3 Small (0.6B) - VFX (and a "large" closed variant) The open models are really fast and high quality

@SkinwalkerRyan ;)

@6___0 did Trump just post this on Truth Social? 😁

I was unable to make the trip to Sedona, however I had my good friend take these pictures of Ross Coulhart's radiation "Portal Site", aka the maintinance shed.....? 34°54'54.8"N 111°54'00.9"W

@HowToAI_ link to the actual thang bohrium.com/en/sciencepedia

China has open-sourced a Wikipedia's replacement and it exposes the fatal flaw in how we store ALL human knowledge. Right now, every textbook, research paper, and Wikipedia article suffers from a massive problem. Researchers call it "reasoning compression." When humans write down knowledge, we write down the conclusions. We skip the messy, step-by-step logic it took to get there. We record the "what." We completely delete the "how" and the "why." This creates the "dark matter" of science, the invisible pathways that actually connect different concepts together. Because of this, we can't verify knowledge easily. We just have to blindly trust the authority who wrote it. Now, researchers just dropped a paper that completely fixes this. They built SciencePedia. An emergent encyclopedia built entirely by AI. But they didn't just ask an LLM to write articles. That would just replicate the same compressed, hallucinatory garbage. Instead, they did something genius. They gave an AI agent a curriculum of 200 university courses and told it to generate 3 million first-principles questions. Then, they had multiple independent AI models solve every single question step-by-step, showing their work in Long Chains-of-Thought. They threw out anything that couldn't be mathematically or logically verified at the endpoint. What they got was a massive, verified knowledge base of pure, unadulterated reasoning. Then, they built an "inverse search engine." When you search for a concept, it doesn't just look up the definition. It retrieves every single logical derivation and causal pathway that leads to that concept across different disciplines. Finally, an AI synthesizes those raw, verified chains into a human-readable article. The results are staggering. The initial version already has 200,000 entries spanning math, physics, chemistry, biology, and computation. Factual errors plummeted by 50% compared to standard AI. Knowledge density skyrocketed. We have spent decades building search engines to look up facts. But facts without the reasoning behind them are just blind memorization. The future of knowledge isn't about looking up answers. It's about uncompressing the logic that makes the answers true.

@SenhorZiborro been there done that

Quando você não tem pendrive ou rede e precisa transferir os arquivos de qualquer modo 🤯🤯🤯

@bridgemindai whats the plot from?

Kimi K2.6 is 6x cheaper per token than Claude Opus 4.7. But per task? It's only 39% cheaper. $0.76 per task for Kimi K2.6. $1.24 for Claude Opus 4.7. Kimi burns so many tokens to complete a task that the 6x pricing advantage nearly disappears. Cheaper per token does not mean cheaper to use. If a model takes 2x the tokens and 7x longer to finish, the savings are an illusion. Stop comparing token prices. Compare cost per task.

Qwen-35B-A3BにDCAを適用した技術的な解説とベンチマーク結果に関するRedditのスレッド reddit.com/r/LocalLLaMA/c…

【Qwen-35Bが超進化！最新手法DCAで最高峰モデルに肉薄】 ローカルLLMのQwen-35Bに「Dynamic Compute Budget Allocation（DCA）」を適用することで、GPT-5級の驚異的な性能を実現しました！✨ 注目の技術ポイント： ・テスト時計算量の最適化で「AI自ら考える深さ」を調整 ・BFS（幅優先探索）で多様な回答案の種を生成 ・難問や有望な解法へ動的にリソースを集中配分 ・DFS（深さ優先探索）で回答の精度を極限まで深化 ベンチマークスコアは21.4%から39.9%へ倍増！ローカル環境でも、複雑なコード生成や論理的タスクを低コストで実行できる可能性が大きく広がりました。🔥 #AI #ローカルLLM

@rumgewieselt try --spec-draft-n-max 6 should be better for longer contexts (16k+) i hear a lot that q4_0 is not good for quality and q8 is min

This is ridiculous: MTP with llama.cpp now runs ... 3× GTX 1080 Ti from 2017 No Tensor Cores No NVLink 33GB total VRAM Qwen 3.6 35B A3B MoE 229K context 71.43 tok/s benchy best 78 tok/s peak Qwen 3.6 27B Dense 196K context 29.68 tok/s best No TurboQuant fork. Key flags: --cache-type-k q4_0 --cache-type-v q4_0 --spec-type draft-mtp --spec-draft-n-max 2 Pascal is not dead 😄

@predictdigest not looking good

POLYMARKET DIGEST — 17 MAY 2026 ⚠️ Bugs & Issues Performance Crisis: Platform experiencing severe latency issues with v2 API — order execution jumping from Authentication Problems: Widespread login failures including fake logout bugs, 2FA codes failing, Google authentication issues. Accounts created before May 4, 2026 experiencing signature errors with v2 API. API Disruptions: Historical orderbook data endpoint potentially discontinued. "Maker address not allowed" errors affecting multiple users. "Order queue full for maker" rejections appearing frequently. Market Display Issues: Bitcoin/Ethereum Saturday markets not appearing (recurring bug). Split/merge token ID mismatches preventing sell orders. Profile pages not loading. ⚡ Disputes Weather Market Manipulation: Jakarta temperature markets under suspected insider manipulation by new accounts, causing $400-30k+ losses. Users report weeks of ignored complaints about removing these markets. Seoul temperature markets missing resolution tabs. Trump "Dumbocrat" Market: Active dispute clarified at 4:00 PM ET May 16. Trump's 20:23 timestamp mention qualifies for resolution, orderbook cleared. 📋 Clarifications System Upgrades: Team acknowledging performance issues, working on legacy contract/backend upgrades with promised daily improvements. KYC-approved users still experiencing poor latency, with Dublin servers outperforming London for some traders. Are you staying away from Polymarket until these API and authentication issues get fixed, or do you think the daily improvements will be fast enough to justify sticking around?

@Snixtp @leftcurvedev_ how about just using the standard MTP (previous mtp)? because --spec-draft-p-min 0.75 consumes VRAM unless --spec-draft-n-max 6 alone is comparable or better than standard

Small update and more data! Reran the MTP tests with these three flags (thanks to @leftcurvedev_ for sharing): --spec-draft-n-max 2 --spec-draft-n-max 2 --spec-draft-p-min 0.75 --spec-draft-n-max 6 --spec-draft-p-min 0.75 Same setup: - Single RTX 3090 - Qwen3.6 27B Q4_K_S - q8_0 KV Short context (4k) can lose if the draft path is too aggressive or rejected too often, thats why it's slower than longer context. n-mac 6 + p-min gives nice speed up in 16k, but even slower in 4k.

@DailyDoseOfDS_ link?

Turn any Autoregressive LLM into a Diffusion LM. dLLM is a Python library that unifies the training & evaluation of diffusion language models. You can also use it to turn ANY autoregressive LM into a diffusion LM with minimal compute. 100% open-source.

@ExpLang_Cn 😗

发个福利，每天1000美金额度消耗，残血Claude 支持外接第三方工具，例如OpenClaw、Hermes等 可用模型： claude-opus-4-6-thinking claude-opus-4-7-high claude-opus-4-7-low claude-opus-4-7-max claude-opus-4-7-medium claude-opus-4-7-xhigh claude-sonnet-4-6-thinking 跟合伙人好说歹说申请下来了的API Key 不够用的话可以找我要邀请码自行注册哈～ Key：sk-ef9185c35f9d3aa405e73554aa40ea44590f6505d67773b071a3f81711327416 国内BaseUrl：ccapi.scydao.com 海外BaseUrl：api.ccode.dev

@ExpLang_Cn gimme 给我嘛 🫡

@populartourist u tried different vals --spec-draft-p-min 0.75 ? i hear many people taht it has no impact

Unsloth Qwen3.6 27B Q6_K doing over 100 t/s with MTP on RTX 5090. That's coming up from 45-50 t/s without MTP. That's insane. --spec-draft-n-max 3 --spec-draft-p-min 0.75

@Tiny_Fish i am failing in setting up the mcp search in LMSTUDIO @lmstudio "tinyfish": { "url": "agent.tinyfish.ai/mcp" } i also tried with api key provided

@leopardracer who is this guy ? credit/reference

THIS AMERICAN DEVELOPER SPENT WEEKS DEBUGGING TIMEOUT ERRORS IN OLLAMA. THEN HE LOOKED UNDER THE HOOD LM Studio is just llama.cpp Ollama is just llama.cpp so he cloned llama.cpp from source, pulled Qwen 3.6 35B off Hugging Face, set up asymmetric KV quantization and got a local server running on 127.0.0.1:8080 plugged it into VS Code, connected it to OpenClaw, 53 tok/s on an M1 Max with 262K context zero wrappers, zero timeout errors, zero API fees bookmark & like this before your next timeout error hits​​​​​​​​​​​​​​​​ full breakdown of my raw llama.cpp setup ↓