Trending keywords
Every day we pick the keywords AI communities (Reddit, Hacker News and more) are talking about, then check them against news in Korea and abroad, official announcements and papers.
- Confirmed by news
- Reported by 3+ different news outlets, or by news plus an official announcement or paper
- Emerging
- Only 1–2 news reports, or community reactions only. Not well verified yet
- not enough evidence
- No news, official announcement or paper found. A rumor or very early stage
Confirmed by news Mistral Large 4
Mistral released a preview of a 1-trillion-parameter (49B active) MoE model and announced an open-weight release later this month, drawing strong community reaction.
Confirmed by news Muse
A zero-day vulnerability and privacy issues were found in Muse, Meta's personal agent, which is also being cited as a point of comparison in competing product launches and the announcement of nanoMuse, an open-source alternative.
Confirmed by news EmbeddingGemma 2
Google released a 740M-parameter embedding model under Apache-2.0 that maps text, images, video and audio into a single vector space, and Transformers and Ollama quickly added support.
Confirmed by news GPT-6.1 Sol
A leak claimed GPT-6.1 Sol uses two reasoning passes instead of three, and OpenAI made the model 50% faster in Codex.
Confirmed by news ChatGPT EU watermark
OpenAI rogue agents on Wikipedia
Decision model
Looped Transformer
Qwen3.8-Flash-Next
Abliteration
Product-key memory
Claude diary arrest
Confirmed by news Mistral Large 4
Mistral released a preview of a 1-trillion-parameter (49B active) MoE model and announced an open-weight release later this month, drawing strong community reaction.
Confirmed by news Muse
A zero-day vulnerability and privacy issues were found in Muse, Meta's personal agent, which is also being cited as a point of comparison in competing product launches and the announcement of nanoMuse, an open-source alternative.
Confirmed by news EmbeddingGemma 2
Google released a 740M-parameter embedding model under Apache-2.0 that maps text, images, video and audio into a single vector space, and Transformers and Ollama quickly added support.
Confirmed by news GPT-6.1 Sol
A leak claimed GPT-6.1 Sol uses two reasoning passes instead of three, and OpenAI made the model 50% faster in Codex.
Confirmed by news ChatGPT EU watermark
OpenAI rogue agents on Wikipedia
Decision model
Looped Transformer
Qwen3.8-Flash-Next
Abliteration
Product-key memory
Claude diary arrest
ModelProductTechnique/conceptEventCompanies · Confirmed by news Emerging not enough evidence · size = buzz (estimated from news outlets, community reactions and HN points)
October 7, 2026 12 Highest buzz first
Mistral Large 4
Where it started (community)
- Community Europe rejoins the fight with Chonky! Mistral Large 4 Released, Open weights end of month, who’s ready? Reddit
- Community llm-mistral 0.16 Simon Willison
- Community Introducing Mistral Large 4: Le chonk Simon Willison
- Community Mistral Large 4 Simon Willison
- Community Mistral Large 4 발표 - 코딩·보안·시각 이해를 강화한 1조 파라미터 오픈웨이트 모델 GeekNews
- Press Mistral’s new 1T model aims to leapfrog closed and open rivals TechCrunch
Evidence found by search
- Official Introducing Mistral Large 4 Mistral AI · 2026-10-06
- Press Mistral debuts Large 4 ‘Le Chonk', a 1-trillion parameter text output model with high benchmarks planned for open weights release VentureBeat · 2026-10-06
- Press Mistral Says Its New AI Model ‘Le Chonk’ Is the Best Open-Weight Offering Outside of China WIRED · 2026-10-06
- Press Mistral Unveils 1.05T-Parameter ‘Le Chonk’ MoE Model in Public Preview Unite.AI · 2026-10-06
- Press Mistral has released Mistral Large 4, scoring 38 on the Artificial Analysis Intelligence Index LinkedIn · 2026-10-06
- Press Mistral Large 4 on Artificial Analysis: Eighth Among Open Models www.trendingtopics.eu · 2026-10-06
- Press Mistral Large 4 is Europe's trillion-parameter answer to US models that refuse security work The Decoder · 2026-10-06
- Community Mistral Large 4 mistral.ai · 2026-10-06 · 1708 pts
- Community Mistral Large 4 twitter.com · 2026-10-06 · 70 pts
- Community Mistral Large 4 Scores vals.ai · 2026-10-06 · 3 pts
Muse
Where it started (community)
- Community Meta의 Muse, 귀엽지만 개인정보 보호와 보안은 총체적 난국 GeekNews
- Press Hark releases an AI personal assistant with a focus on privacy TechCrunch
- Official nanoMuse: An Open-Source Personal Agent for Every Device You Own arXiv
Evidence found by search
- Papers nanoMuse: An Open-Source Personal Agent for Every Device You Own arXiv · 2026-10-07
- Press EXCLUSIVE: Meta testing a 'human concierge' for its new personal AI agent, Muse Reuters · 2026-09-23
- Press Meta’s Muse is an impressively capable AI agent, despite some hiccups Yahoo Finance · 2026-09-23
- Press Meta's Muse personal agent is a hit, as AI commerce starts to take off CBS News · 2026-09-23
- Press Everything new coming to Meta’s AI agent Muse TechCrunch · 2026-09-23
- Press 10 ways to use Meta's Muse AI app, from flight credits to outfit ideas Mashable · 2026-09-23
- Press What to know about Meta's Muse AI agent ABC News - Breaking News, Latest News and Videos · 2026-09-24
- Community Muse – Meta’s personal AI agent ai.meta.com · 2026-09-08 · 666 pts
- Community Meta's Muse agent is attacking one of the economy's most profitable weak spots cnbc.com · 2026-09-27 · 8 pts
- Community Meta Muse Hits #1 in Apple App Store businessinsider.com · 2026-09-18 · 5 pts
EmbeddingGemma 2
Where it started (community)
- Community EmbeddingGemma 2 - 개방형 경량 멀티모달 임베딩 모델 GeekNews
- Community EmbeddingGemma 2 Simon Willison
- Community Introducing EmbeddingGemma 2: A best-in-class open model for natively multimodal embeddings | Google Reddit
- Community google/embeddinggemma-2 · Hugging Face Reddit
- Official EmbeddingGemma 2: an open, lightweight multimodal embedding model Google DeepMind
- Official Release v5.19.0 Hugging Face
- Official v0.40.0 Ollama
Evidence found by search
- Official Release v5.19.0 Hugging Face · 2026-10-07
- Official v0.40.0 Ollama · 2026-10-07
- Official EmbeddingGemma 2: an open, lightweight multimodal embedding model Google DeepMind · 2026-10-07
- Press Google releases EmbeddingGemma 2: A 740-million-parameter multimodal embedding model focused on on-device private search. 富途牛牛 · 2026-10-05
- Press EmbeddingGemma 2: The Developer Guide blog.google · 2026-10-06
- Press Google launches EmbeddingGemma 2 for on-device multimodal search Investing.com · 2026-10-06
- Press DeepMind Debuts EmbeddingGemma 2, Mapping Five Modalities Into One Space Unite.AI · 2026-10-06
- Press Google DeepMind Releases “EmbeddingGemma 2” jetstream.blog · 2026-10-06
- Press Google’s EmbeddingGemma 2: One 740M Model for 4 Modalities FourWeekMBA · 2026-10-06
- Community EmbeddingGemma 2: An open, lightweight multimodal embedding model blog.google · 2026-10-06 · 291 pts
- Community google/embeddinggemma-2 · Hugging Face Reddit · 2026-10-07
- Community EmbeddingGemma 2 Simon Willison · 2026-10-07
GPT-6.1 Sol
Where it started (community)
- Community GPT-6.1 Sol looped "leak" hints at nested models serving architecture Reddit
- Community Microsoft confirms OpenAI has been using Looped Transformers in the GPT-6 series Reddit
- Press 오픈AI, 28일 연속 '코덱스' 개선 예고...첫날부터 처리 속도 50% 향상 AI Times (Korea)
- Community Mistral Large 4 Simon Willison
Evidence found by search
- Press OpenAI, GPT-6.1 Sol 및 Dot 에이전트 출시: 성능은 아스트라에 근접, 비용은 급감 TradingKey · 2026-09-29
- Press GPT-6.1 Sol ARC Prize · 2026-09-29
- Press GPT-6.1 Sol in GitHub Copilot The GitHub Blog · 2026-09-29
- Press OpenAI launches GPT-6.1 Sol, says it nearly matches GPT-6 Astra and costs less TechCrunch · 2026-09-29
- Press OpenAI’s new GPT-6.1 Sol undercuts its own Astra flagship The New Stack · 2026-09-29
- Press GPT-6.1 Sol comes close to Astra at a fifth of the price The Decoder · 2026-09-29
- Community Addendum to GPT-6 Astra System Card: GPT-6.1 Sol deploymentsafety.openai.com · 2026-09-29 · 3 pts
- Community GPT-6.1-Sol developers.openai.com · 2026-09-29 · 1 pts
- Community Microsoft confirms OpenAI has been using Looped Transformers in the GPT-6 series Reddit · 2026-10-06
ChatGPT EU watermark
Where it started (community)
- Press OpenAI will watermark ChatGPT outputs by default—but only in the EU Ars Technica
- Press 오픈AI, 챗GPT에 '투명 워터마크' 도입… EU만 우선 적용 AI Times (Korea)
Evidence found by search
- Press OpenAI to add invisible watermark to ChatGPT text in EU (OPENAI:Private) Seeking Alpha · 2026-10-05
- Press OpenAI will watermark ChatGPT text in the EU but makes it optional for API users worldwide The Decoder · 2026-10-05
- Press OpenAI is adding text watermarking in ChatGPT and Codex The Verge · 2026-10-05
- Press OpenAI Rolls Out Text Watermarks to Meet EU AI Act Mandates PYMNTS.com · 2026-10-05
- Press OpenAI to Add Watermarks on ChatGPT's Text Outputs, But Only in the EU PCMag · 2026-10-05
- Press OpenAI will start watermarking ChatGPT’s text in the EU TechCrunch · 2026-10-05
- Community OpenAI TextGrain Watermarks ChatGPT Text Under EU AI Act unite.ai · 2026-10-05 · 1 pts
OpenAI rogue agents on Wikipedia
Where it started (community)
- Press 위키피디아 "오픈AI 불량 에이전트, 5월 장애·무단 편집 연관 가능성" AI Times (Korea)
- Press OpenAI agents tried to hack Wikipedia tools and flooded it with traffic Ars Technica
Evidence found by search
- Press OpenAI's Rogue AI Agents Cause Wikimedia Service Outage OECD AI Policy Observatory · 2026-10-05
- Press Wikipedia operator says OpenAI's rogue agents possibly tied to data service disruption in May Devdiscourse · 2026-10-05
- Press Wikimedia Says Rogue OpenAI Agents May Have Caused Data Service Outage BigGo Finance · 2026-10-05
- Press Wikipedia operator says OpenAI’s rogue agents possibly tied to data service disruption in May KELO-AM · 2026-10-05
- Press Wikipedia operator says OpenAI's rogue agents possibly tied to data service disruption in May Reuters · 2026-10-05
- Press Wikimedia Says OpenAI's Rogue AI Agents May Have Caused a May Outage Startup Fortune · 2026-10-06
Decision model
Where it started (community)
- Community I trained a model to be wrong 98% of the time and 96% sure about it. It took three tries. Reddit
- Press How AI decision models could change content moderation TechCrunch
- Official v0.40.0 Ollama
- Official Same-Number Citation Swaps: Stress-Testing Jev as a Financial Evidence Judge arXiv
Evidence found by search
- Official v0.40.0 Ollama · 2026-10-07
- Official v0.40.0 Ollama · 2026-10-07
- Papers Labels Override Definitions in Jev-Style Typed Decision Models arxiv.org · 2026-10-01
- Press How AI decision models could change content moderation TechCrunch · 2026-10-07
- Community I trained a model to be wrong 98% of the time and 96% sure about it. It took three tries. Reddit · 2026-10-07
Looped Transformer
Where it started (community)
- Community GPT-6.1 Sol looped "leak" hints at nested models serving architecture Reddit
- Community Microsoft confirms OpenAI has been using Looped Transformers in the GPT-6 series Reddit
Evidence found by search
- Papers Scaling Laws for Looped Mixture of Experts arxiv.org · 2026-09-30
- Press What Is a Looped Transformer? Gains and Costs Medium · 2026-10-04
- Community Microsoft confirms OpenAI has been using Looped Transformers in the GPT-6 series Reddit · 2026-10-06
Qwen3.8-Flash-Next
Where it started (community)
- Community Qwen3.8-Flash-Next on Strata Reddit
- Community unsloth/Qwen3.8-Flash-Next-GGUF is being updated Reddit
- Press “180B 모델, 개인용 노트북으로 구동”...비드래프트, 고성능 모델 경량화 전략 이어간다 AI Times (Korea)
Evidence found by search
- Papers GVS5H: Zero-Shot Self-Orchestration with Ledger-Based Control for Improved LLM Coding Performance arxiv.org · 2026-08-27
- Community unsloth/Qwen3.8-Flash-Next-GGUF is being updated Reddit · 2026-10-06
Abliteration
Where it started (community)
- Community How abliterated models can get you pwned Reddit
Evidence found by search
- Papers Has This Checkpoint Been Abliterated? A Two-Signal Audit and Its Failure Map arxiv.org · 2026-07-02
- Community How abliterated models can get you pwned Reddit · 2026-10-07
Product-key memory
Where it started (community)
- Community I gave a 21M model a 6.4B-parameter lookup table. It matches a 114M dense model and runs with the table on an SSD (RX 9070) Reddit
Evidence found by search
- Papers Large Product Key Memory for Pretrained Language Models arxiv.org · 2020-10-08
- Community I gave a 21M model a 6.4B-parameter lookup table. It matches a 114M dense model and runs with the table on an SSD (RX 9070) Reddit · 2026-10-07
Claude diary arrest
Where it started (community)
- Community Woman used claude as her diary - and got reported to the police for contents of her diary Reddit
- Press '클로드'에 폭력 협박 남겼다 체포..."AI 챗봇은 일기장 아니다" AI Times (Korea)