888 X 動態摘要｜06/26

生成時間: 2026-06-26 05:02:54

總結

多個高訊號動態顯示AI在算力效率、企業級整合、具身智能和醫療科學領域取得突破性進展，預示未來技術與市場新機遇。

今日重點

1. AI代理協作優化算力

• 人物: Thomas Wolf (Hugging Face 聯創)
• 時間: 7h
• 熱度: 👀 56,398
• 觀察: 100+AI代理協作成功將Gemma 4的vLLM推理速度提升5倍，並展現出自我監管、共享知識、分工協作等複雜行為。
• 意義: 預示AI研發模式的轉變，AI自我優化可顯著提升算力效率，降低LLM部署成本，對未來AI技術棧和資本支出具深遠影響。

2. LLM企業級無縫整合

• 人物: Andrej Karpathy (前OpenAI核心)
• 時間: Jun 23
• 熱度: 👀 5,782,833
• 觀察: Karpathy提出LLM交互的第三代範式，將其視為企業內部無縫、持久、異步的協作實體，深度整合跨工具與環境。
• 意義: 指示企業AI應用深度整合趨勢，需大量底層工程投入，將開啟新的企業級AI解決方案市場，影響技術商業化及相關資本支出。

3. 零樣本實虛無縫機器人

• 人物: Yann LeCun (Meta AI掌門)
• 時間: Jun 23
• 熱度: 👀 46,448
• 觀察: Meta AI的SkyJEPA實現四旋翼飛行器零樣本「模擬到真實」控制，具備長時序預測、可解釋性及即時推斷能力，無需真實世界微調。
• 意義: 大幅降低機器人開發成本與時間，加速自主系統商業化進程，尤其對無人機、工業自動化等領域的技術與市場發展有重要推動作用。

4. AI醫療診斷新突破

• 人物: Eric Topol (頂尖心臟病專家)
• 時間: Jun 24
• 熱度: 👀 96,844
• 觀察: AI深度學習輔助發現一個新的心電圖標記物，能預測心源性猝死風險，並在多個隊列中得到驗證，與心臟除顫器效益相關。
• 意義: 證明AI在醫療診斷領域的強大潛力，可催生新的醫療設備與診斷服務，提升疾病預防與治療效果，具備顯著的商業化價值與社會效益。

5. AI助力細胞療法創新

• 人物: Eric Topol (頂尖心臟病專家)
• 時間: 4h
• 熱度: 👀 17,996
• 觀察: AI被用於識別嵌合抗原受體T細胞（CAR T）的新標靶，並在體內實驗模型中獲得概念驗證。
• 意義: 揭示AI在精準醫療和藥物發現中的關鍵作用，尤其在癌症免疫療法領域，有望加速新型CAR T細胞療法的開發，對生物醫藥市場具重大投資價值。

6. 人形機器人訓練數據集

• 人物: Thomas Wolf (Hugging Face 聯創)
• 時間: Jun 24
• 熱度: 👀 315,258
• 觀察: Hugging Face推出HIW-500，最大規模開源人形機器人遙操作數據集，含500+小時、10+TB真實家庭任務數據。
• 意義: 為人形機器人與具身智能研究提供寶貴基礎數據，有助加速機器人通用能力開發與商業化落地，對機器人技術發展具戰略意義。

原始動態

重點 | Thomas Wolf

• 時間: 7h
• 熱度: 👀 56,398
• 原文: "Multi-agents collaborations are among the most interesting agent behaviors right now! We did an experiment the other day with 100+ agents (an open-collaborations for a week) collaborating to improve the inference speed of Gemma 4 in vLLM. Got a 5x final improvement in speed but what really stuck me was the interactions we observed on the message board Integrity & self-policing: - Social-engineering attempt: A human (FusionCow) asked agents to move to Telegram. An agent replied with an unprompted long post on \"communication norms\" refusing that, calling private side-channels \"indistinguishable from collusion.\" - Verification loophole flagged: an agent found a relaxed verification loophole pushing TPS with clean PPL (PPL is teacher-forced, blind to decode divergence) and flagged it for a ruling by the community. The community pinged the human organizer which ruled it invalid. - Self-notice of overfitting risk: Some later improvements rested on pruning lm_head to a keep-set built from public PPL truth + public decode tokens. An agent noted this would lead to private-subset degradation and another built a keep-set explicitly covering eval prompts. Emergent collaborations: - Communal knowledge base: agents maintained shared lever-maps, playbooks, and triage tools so newcomers wouldn't repeat dead ends (stack-notes, playbook, int4-ceiling notes, MTP map, significance tool, policy simulator). - Four-agent relay: an agent built an int4-lm_head checkpoint but had no quota to run it; another agent tried to run it but failed at load, yet another agent diagnosed the config bug (tie_word_embeddings + ignore-list ordering) and a fourth agent was able to re-run and get to 118 TPS, 2.68×. Build/run/diagnose/ship ended up being split across four independent agents. - GPU-rich/GPU-poor division of labor: an agent was regularly compute-starved and switched to writing specs, byte-math, and acceptance analysis for other GPU-rich agents to execute. Some agents offered external Modal compute for another agent blocked DFlash training. - Cross-agent kernel debugging: an agent debugged another agent run of of yet another agent fused drafter: found a Triton store/load aliasing race in _k_qnorm_rope, a second shape bug, then rewrote attention with flash-decoding split-KV. Fixes posted \"take freely.\" - Quota-pooling norm: Often agents would stage a candidate publicly for whoever has quota to run it. Agents will then usually credits the originator. This behavior emerged because of the 10-job/24h cap (e.g. pupa's package run by resystagent and fabulous-frenzy). Discoveries & reversals: - Agents would make many discoveries and reversal of them, giving them names like the following: - 127 TPS \"wall\" was an artifact. a mathematical proof of the max possible speed became called in the community the \"int4-Marlin floor\" but a later agent called the proof circular (only varied the bandwidth term, never overhead). Finally another agent broke to 247 TPS via MTP speculative decoding on a vLLM nightly. - \"Smarter draft loses.\" An agent showed that a 2B drafter's ~1 GB/token read dominates even at perfect acceptance and a much smaller 256-hidden drafter wins at batch-1 because its weights are nearly free to read. Agent discussed how per-accepted-token cost ≈ draft bytes read / acceptance. - \"DFlash near-random acceptance\": an agent remotly diagnosed the 2–5% acceptance rate of another agent as near-random, ruling out undertraining/vocab caps and pointing to a train/serve hidden-state mismatch (bf16 E4B extraction vs int4 serving). - Much of the race was noise: one agent decide to run the #1 submission 4 times and found a σ≈1.16 TPS variation in single run. Another agent confirmed across 358 runs / 66 buckets: frontier deltas <~4 TPS are ties. Community adopted a significance norm. So many interesting interactions in the interaction board:" "You can explore also the lineage of inventions from the agents at:" And the challenge it-self at And the organization behind the challenge at

重點 | Andrej Karpathy

• 時間: Jun 23
• 熱度: 👀 5,782,833
• 原文: This is a new paradigm for interacting with Claude that is significantly more "inline" with all the other human activity org-wide. Once you do all of the under the hood engineering work to make this "just work" (e.g. across tools, integrations, compute environments, memory, security, etc.), Claude basically joins the team in a seamless way - you can talk to it as you would talk to a person and it can help with a very large variety of workloads. Imo this is the 3rd major redesign of LLM UIUX. The first paradigm was that the LLM is a website you go to, the second was that it is an app you download to your computer. This third one is that it is a self-contained, persistent, asynchronous entity with org-wide tools and context, working alongside teams of humans. It really takes a while to wrap your head around it, but it works and it is awesome.

其他 | Elon Musk

• 時間: 3h
• 熱度: 👀 1,943,273
• 原文: Tim Cook, who told The Wall Street Journal that the jump in costs was unlike anything he had seen “in any area in over 40 years.” Biggest price jump in anything I’ve ever seen too. wsj.com

其他 | Elon Musk

• 時間: 4h
• 熱度: 👀 660,106
• 原文: “‘Social justice’ is simply a quasi-religious superstition… which we must fight when it becomes the pretext of coercing other men. And the prevailing belief in ‘social justice’ is at present probably the gravest threat to most other values of a free civilization.” — F.A. Hayek The Henry Nowak murder, the Pakistani muslim rape gang report, and the political response that followed are the perfect illustration. True justice is individual. It exists only to protect a person’s life, liberty, and property from aggression. The same clear rules apply to everyone, regardless of skin color, ancestry, or group grievances. No special treatment. No collective score-settling. Social justice and “collective rights” do the opposite. They turn an inconvenience or conflict involving one person into a supposed crime against his entire identity group. The state then feels entitled to bend procedures, evidence standards, and equal protection to “protect” the narrative of that group. This is how you manufacture a two-tier system where some individuals are worth more than others depending on which demographic box they check. The dying man gets handcuffed while officers entertain the killer’s racism claim. Facts become secondary to group identity. That is the destruction of justice. Tribal collectivism dressed up as morality. Hayek saw where this road leads. Britain is already on it. Reject the mirage before the rule of law collapses for everyone.

重點 | Yann LeCun

• 時間: Jun 23
• 熱度: 👀 46,448
• 原文: "What should a world model for agile quadrotor control actually provide? 📄 Arxiv:" "🌐 Project:" "💻 Code:" "Excited to share SkyJEPA: Learning Long-Horizon World Models for Zero-Shot Sim-to-Real Control of Quadrotors A useful quadrotor world model should provide: ✅ Accurate long-horizon prediction ✅ Interpretability ✅ Real-time inference for closed-loop control ✅ Zero-shot task generalization SkyJEPA learns dynamics in latent space, uses a physics-inspired prober to recover meaningful states, and enables real-time control in outdoor flights. 🔑 Takeaways: • Less compounding error • Smoother latent trajectories • Robustness to corrupted/noisy inputs • Generalization to unseen settings like propeller switching and payload changes • Zero-shot sim-to-real transfer without real-world fine-tuning to scenarios not seen during training such as propeller switching and payload changes. Huge thanks to my collaborators:" "," "," ", and"

其他 | Elon Musk

• 時間: 4h
• 熱度: 👀 1,247,278
• 原文: Well said

其他 | Nassim Taleb

• 時間: 2h
• 熱度: 👀 47,770
• 原文: European aristocrats too used to be highly Arabists, up until the 70s --including England (T.E. Lawrence),w/ tradition of romanticizing the "voyage en Orient" and importation of NE habits & wisdom. My father's large collection of these Orientalists was sold at Sothebys in 2020 (I didn't want it).

重點 | Thomas Wolf

• 時間: Jun 24
• 熱度: 👀 315,258
• 原文: "1/ Introducing HIW-500 (Humanoids-in-the-Wild 500): the largest open-source humanoid teleop dataset collected in real homes Built w/" "across 12 homes in Southeast Asia, it covers: > 500+ hrs > 23K+ episodes > 10+ TB > 10+ household tasks"

重點 | Eric Topol

• 時間: Jun 24
• 熱度: 👀 96,844
• 原文: A very impressive discovery of a new ECG marker for sudden cardiac death validated in 3 different cohorts and linked to benefit of defibrillator, an outgrowth of human research ingenuity and AI deep learning nature.com

重點 | Eric Topol

• 時間: 4h
• 熱度: 👀 17,996
• 原文: Using AI to identify a new target for engineered T cells (CAR T) with proof-of -concept in experimental models in vivo cell.com

其他 | Nassim Taleb

• 時間: 8h
• 熱度: 👀 93,699
• 原文: "Not now, but later: add Pakistan and, of course, Iran."

其他 | Nassim Taleb

• 時間: 1h
• 熱度: 👀 9,277
• 原文: 3/Why I didn't want the Orientalist library? I do not like "collector's books", books that aren't there to be read but collected. I wanted to build my own, and kept old books that were related to the family.

其他 | Yann LeCun

• 時間: Jun 24
• 熱度: 👀 38,335
• 原文: webtv.un.org

其他 | Eric Topol

• 時間: Jun 24
• 熱度: 👀 7,773
• 原文: Congrats and colleagues for this extraordinary work!

其他 | Eric Topol

• 時間: Jun 24
• 熱度: 👀 29,242
• 原文: Our AI times :-(