Bounded knowledge map
Recent connections
The 90 newest entries plus the curated synthesis cluster that connects ideas across the wiki.
120 nodes · 22 links. The server bounds this neighbourhood, so the browser never downloads or simulates the complete wiki graph.
120 nodes · 22 links shown. Hover or focus to isolate neighbours; select a node to open its source entry.
Linked entries
22 shownI ran Qwen 3.8 Flash-Next on my AMD 7900 XTX at 500k context. All local and what a shift it has been.→What setup do you use, which harness? and how do you customize it to run qwen 3.8 flash next swift with a high context length and a reasonable amount of tok/sec?
What to expect during SEMICON West: Join theCUBE Oct. 14→What to expect during the AI Data Pipeline Forum: Join theCUBE Oct. 13
Introducing Multi-Agent Self-Supervision (MASS): Recursive self-improvement through collective intelligence Blog: https://pub.sakana.ai/mass…→New paper from the team at Sakana AI: Recursive Self-Improvement through Multi-Agent Self-Supervision (MASS) How do we keep improving AI whe…
Basalt: Flash-Next at 665 tok/s structured, 354 prose on a 5090 + 5060 Ti (2.6x Strata)→Qwen Flash Next @ 137 tok/s & 3,497 tok/s Prefill w/ 512k context on a 5090, 192gb ram, Windows Build, comparing Strata and Infernix
Syntheses
30 shownWiki Lint Report — 2026-04-196Wiki Lint Report — 2026-04-126Wiki Lint Report — 2026-04-265Synthesis: Arxiv-Cs-Lg3Synthesis: Arxiv-Cs-Cl3Synthesis: Arxiv-Cs-Ai3Synthesis: Harrison-Chase--X2Synthesis: Swyx--X2Synthesis: Elon-Musk--X1Synthesis: Arxiv-Cs-Cv1Synthesis: R-StablediffusionUS family SUVs: Tesla FSD, Super Cruise, BlueCruise and Rivian Autonomy+Synthesis: Arxiv-Cs-RoSynthesis: Gary-Marcus--XSynthesis: TechmemeWiki Lint Report — 2026-10-04Synthesis: R-ChatgptWiki Lint Report — 2026-09-27Wiki Lint Report — 2026-09-20Inbox triage 2026-09-19Wiki Lint Report — 2026-09-13Wiki Lint Report — 2026-09-09Wiki Lint Report — 2026-07-19Wiki Lint Report — 2026-07-15Wiki Lint Report — 2026-07-05Wiki Lint Report — 2026-06-28Wiki Lint Report — 2026-06-22Wiki Lint Report — 2026-06-07Wiki Lint Report — 2026-05-03How Vercel AI Gateway zero data retention protects data
Model Releases
51 shownBasalt: Flash-Next at 665 tok/s structured, 354 prose on a 5090 + 5060 Ti (2.6x Strata)2Qwen Flash Next @ 137 tok/s & 3,497 tok/s Prefill w/ 512k context on a 5090, 192gb ram, Windows Build, comparing Strata and Infernix2What setup do you use, which harness? and how do you customize it to run qwen 3.8 flash next swift with a high context length and a reasonable amount of tok/sec?1I ran Qwen 3.8 Flash-Next on my AMD 7900 XTX at 500k context. All local and what a shift it has been.1b115591b115581microsoft/AesCode 8B and 32BConverting dense models into Mixture-of-Expertsvllm-ascend updates (GLM5.3-flash) on dual 310p Ascend cardsI trained a 102M recursive BitNet-v2 model from scratch: 64K context, trained on less than 5B tokens[vllm] Is it possible to run Qwen3.8-27B-FP8 on a 5090?UPDATE: Qwen 3.8 27B 140 tok/s on single RTX 3090 Megakernel: KL divergence 0.0009 vs llama.cppb11555Impressive by @river_ai_inc to beat @deepseek_ai rates - cache in particular is tough! Talented team there, full RL + inference stack should…b11553OMG! If you have a Mac with 64GB, try Qwen3.8-Flash-Next-oQ4e-mtp with oMLX!Final revision of zzpanic vllm-radiance for 1 x R9700 GPU, qwen3.8-27bRunning Next Flash IQ3_XXS at ~70 tok/s with 100k context or 2 instances of Qwen 3.6 35B A3B IQ4 at ~145 tok/s with 256k all on $500 of ex mining BC-250 boardsb11556Building a 4x R9700 setup for a 10 person startupQualcomm CEO Cristiano Amon says AI companies are 'starting to do phones' and asking Qualcomm to support running 100B-parameter local LLMs continuously by 2028 (Alex Heath/Sources)b11557: rpc : turn GGML_RPC_DEBUG into a verbosity level and add logs (#29544)I tested different Qwen 3.8 27B quantsQwen3.8 Flash Next fixed my GNOME extensionEvery Model That Can Be Run On 10-16GB VRAM Rankedlocal semantic file search for Linux (Rust, llama.cpp, EmbeddingGemma 2)v0.40.3'Strata' for GLM5.3 Flash is here for some! Project MayaAnatomy skill for building wireframe designs and interactive prototype machines is now an officially integrated optional skill in Hermes Age…Qwen 3.8 27B Q5 vs Qwen 3.8 Next Q3_S for document analysisGemma4 E4B EXL3 - Alternative to Llama.cpp on Jetson Orinb11554The NFC East winning 0 games on a week in which two teams played each other feels like a pretty good encapsulation of the division at the mo…We were able to fine tune an open weight model (Qwen3.5-9B) on proprietary internal data at @coinbase, and it out performed frontier models,…OpenMed 3.0 is out: Apache-2.0 clinical AI that runs fully local and never falls back to the cloud. 422 open issues if you want in on 3.1Now you can grow Bonsai on your potatoThis is an incredible exchange.I have no insight into the facts of this situation, but “CLAIM: (outline a claim)/FACT: (outline facts that don’t address the claim just lai…Benefits of using bigger models than Qwen 3.8 flash next?honestly pretty cool to have this on my new tab page just merged in some PRs by others too!I asked 3 Qwen 3.8 models to solve recent AtCoder problems at 4 thinking effortsBanger paper from Meta on AI research agents. (bookmark it) If you build AI-scientist systems, this one is worth your time. The result: A 27…How to make gemma 4 works with SWEBench?3.66B formal logic model built on Granite 4.2 3B · Hugging Faceb11552b11540A Little Help Please?Improve token per second without touching quantThe ultimate LLM benchmark: Chilean Spanish. 80% of AI models fail to recognize Chilean Spanish and insist on speaking like an ArgentineCreated laya : Now Introducing a new 800 Million Param physics-based typed decision model with 73k context and image supportb11550
Agents
8 shownNew paper from the team at Sakana AI: Recursive Self-Improvement through Multi-Agent Self-Supervision (MASS) How do we keep improving AI whe…1Dusting off this tweet from April because this gap in shared understanding of LLM capability is *widening*. It's now less 'two groups of peo…The Top AI Papers of the Week (October 5 - October 11): - VERA - HERMES - CorpusMap - VeriHarness - Sharpening Tax - Harness Learning - Mult…One of the biggest historical bottlenecks for training specialized models has been curating ground-truth. This is typically a very labor int…Why AI agents might not be right for you“The field of artificial life is a very long-term bet on intelligence.” 🐙 Our former intern Akarsh Kumar recently joined @MLStreetTalk to d…AI agent makers are promising privacy — will they deliver?good framing from @ch3nweiii - a lot of agent steps are yes/no calls, not generation Jev answers those as typed questions with calibrated pr…
Industry
14 shownWhat to expect during SEMICON West: Join theCUBE Oct. 141What to expect during the AI Data Pipeline Forum: Join theCUBE Oct. 131🚨 Elon Musk agrees with 'some of the things Anthropic has said' and warns against 'torturing superintelligence' after reading Grok's intern…Report: Amazon in talks with AI startup Decart over possible $7B acquisitionPitchBook: robotics and physical AI companies have raised ~$48B YTD, as they gather training data from people completing tasks in factories, offices, and homes (Rafe Rosner-Uddin/Financial Times)Decision 3.0 now on Core ML 🍎 d3-lite triaging 1,000 pull requests into vllm-sr workgroups, live on an M5 Pro: 0.12 s per PR, 8.3 PRs/s, no…Sources: Amazon is in talks to buy AI startup Decart in a deal that could value it at ~7B; Anthropic had decided against acquiring Decart for ~6B in September (Wall Street Journal)Introducing Decision 3.0: towards open multimodal foundation decision models ⚡ Open and state-of-the-art from 0.6B to 27B: #1 on both Jev De…Q&A with Palantir CTO Shyam Sankar on the Iran school strike and AI targeting, the DOD dropping Anthropic, AI liability, effective altruism, and more (CBS News)Microsoft unveils Microsoft-Decision-1, a fast decision-scoring model trained on Qwen3.5-9B, and says it will soon rebase it on MAI, OpenAI, and other models (Achint Srivastava/Command Line)We turned @OpenAI's math problems into open-source RL environments on @huggingface! Early and rough (the verifier only accepts the exact ori…AI reshapes professional services around trust and business outcomesAnthropic is cutting off its internal evaluations from the internetI need you guys to understand just how insanely good Meta Muse is at idea generation. Look at your “ideas” tab. The AI labs - OpenAI, Anthro…
Safety
2 shownApplications
2 shownHardware
6 shownNvidia reportedly in early talks to buy, acquihire or invest more in Reflection AIyesterday we took SF @Techweek_ to the water ✈️ The Science of Speed brought together founders, researchers, and builders front row for Flee…Silicon Valley is falling in line with Trump’s attempt to rename AIhere’s some of what I went to friday night sf tech week > agent economy gala great jobb @imaanxsultan @llama_index > silicon valley idol so …What does the world look like inside an AI model? Here are 3 world model architectures based on how they represent the world (+ the subtypes…SAM 3.1 + DINOv3 from @AIatMeta is a crazy combo 🔥 it can even hold onto surgical tools while a forearm sweeps over them. one leaves for 15…