888 X 動態摘要|07/18
生成時間: 2026-07-18 05:03:12
總結
候選動態顯示了AI領域在模型效率、機器人長程學習、語音互動體驗上的重大進展,以及區域經濟體對新創與基礎建設的投資方向。
今日重點
1. Grok 4.5 極致效率
- 人物: Elon Musk (xAI 創始人)
- 時間: 3h
- 熱度: 👀 531,770
- 觀察: xAI的Grok 4.5在AI任務上展現極高成本效益,單次任務僅$0.31,遠低於GPT和Claude競品。
- 意義: 領先業界的成本效率是AI模型商用的關鍵競爭力,能大幅降低營運成本、加速市場滲透並擴大規模。
2. 機器人長程記憶突破
- 人物: Fei-Fei Li (ImageNet之母)
- 時間: Jul 15
- 熱度: 👀 242,788
- 觀察: ImageNet之母團隊發布RoboTTT,將機器人模型上下文擴展至8,000時間步,並保持固定推理成本。
- 意義: 大幅提升機器人持續學習與在線錯誤修正能力,有望實現更複雜的工業自動化與新應用。
3. ChatGPT語音互動躍進
- 人物: Sam Altman (OpenAI CEO)
- 時間: Jul 16
- 熱度: 👀 845,377
- 觀察: Sam Altman表示ChatGPT語音模型的品質已突破門檻,使得語音互動使用頻率超越文字輸入。
- 意義: 提升AI助理的用戶體驗與無障礙性,預示語音作為主流人機介面的加速普及,拓展商業應用潛力。
4. 印度建設新創城市
- 人物: Balaji Srinivasan (前Coinbase CTO)
- 時間: 2h
- 熱度: 👀 2,252,017
- 觀察: 前Coinbase CTO Balaji Srinivasan指出,印度正積極建設以新創為導向的城市。
- 意義: 顯示大規模的基礎設施投資和政策支持,有望催生新的科技生態系,吸引全球資本與人才進駐。
5. Grok獲用戶高偏好
- 人物: Elon Musk (xAI 創始人)
- 時間: 5h
- 熱度: 👀 739,856
- 觀察: Elon Musk表示Grok Build與Grok 4.5在用戶模型偏好中位居前列。
- 意義: 用戶的高度偏好是產品市場接受度的重要指標,預示Grok在AI市場中具備競爭力與商業化潛力。
原始動態
重點 | Elon Musk
- 時間: 3h
- 熱度: 👀 531,770
- 原文: "Grok 4.5’s efficiency is ridiculous On Artificial Analysis, it costs just $0.31 per Intelligence Index task while delivering frontier intelligence For comparison: • Claude Fable 5 (max): $2.75 • Claude Opus 4.8 (max): $1.80 • GPT-5.6 Sol (max): $1.04 • Kimi K3: $0.95 • Grok 4.5: just $0.31 That makes Grok 4.5 roughly: • Nearly 9× cheaper than Claude Fable 5 • Nearly 6× cheaper than Claude Opus 4.8 • 3× cheaper per task than Kimi K3 • 3.4× cheaper than GPT-5.6 Sol On FrontierSWE, Grok 4.5 also used nearly 6× fewer total tokens per task than Fable 5 while still ranking near the very top SpaceXAI has figured out how to deliver top-tier intelligence with extraordinary token efficiency At this level of performance, no other frontier model is operating in the same cost-efficiency league Grok 4.5 does not brute-force every problem with endless tokens and compute It makes every token and every dollar work harder"
其他 | Elon Musk
- 時間: 9h
- 熱度: 👀 1,584,447
- 原文: "How many of you have heard of Ilias Kasidiaris? The former Greek MP and politician remains imprisoned in Greece, due to one of the most controversial cases in the EU. He’s serving time from the Golden Dawn convictions, but a recent judicial decision has drawn heavy criticism: His June 2025 release request was denied 2-1, against the Prosecutor’s recommendation, partly because he maintains an active X account (" ), with help from supporters to comment on politics. “The applicant… maintains a personal account on… X… expressing his views on current political and social affairs,” the court said. They will therefore keep him imprisoned despite good conduct and significant time served. This sets a dangerous precedent for political speech. How does it differ from cases Elon Musk has criticized in the UK?
其他 | Nassim Taleb
- 時間: 8h
- 熱度: 👀 118,826
- 原文: The problem with medical papers is that there is no clear "norm". For a T de F cyclist, a 20 min walk is nothing. For a beer-guzzling, post-vegetarian, suburban soccer-fan (Argentina) --a retired banker w a 3rd cousin in Congress-- a twenty-minute walk is a major undertaking.
重點 | Elon Musk
- 時間: 5h
- 熱度: 👀 739,856
- 原文: Grok Build and Grok 4.5 top user model preference
其他 | Ray Dalio
- 時間: 2h
- 熱度: 👀 39,578
- 原文: I often hear people say, "Wouldn't it be good to do this or that?" It's likely they are being distracted from far more important things that need to be done well.
重點 | Balaji Srinivasan
- 時間: 2h
- 熱度: 👀 2,252,017
- 原文: India is building startup cities.
其他 | Elon Musk
- 時間: 1h
- 熱度: 👀 1,001,461
- 原文: Cool
重點 | Fei-Fei Li
- 時間: Jul 15
- 熱度: 👀 242,788
- 原文: "We scaled a robot model natively to 8,000 timesteps of context, 5 minutes worth of muscle memory, with constant inference cost. Robot policies used to live their lives a few frames at a time (< 0.1 sec), instantly forgetting what just happened. We pushed to 3 orders of magnitude beyond SOTA. Introducing RoboTTT. Test-Time Training (“TTT”) carries a tiny model inside the model. Every incoming sensor reading triggers one gradient step on that tiny core, so the history keeps getting compressed into its weights. The hidden state has a fixed size (literally a small neural net), so the robot can “grok” arbitrarily long experience with little overhead. Learning continues indefinitely after deployment. We can then put an entire video in context as prompt! RoboTTT enables one-shot in-context learning from human video: in circuit board assembly, a human demonstrates a never-seen configuration once, and the robot imitates it faithfully. Humans drop things all the time, but we pick them up so fast that we don’t even notice. That reflex to fix is half of our physical competence. RoboTTT shows self-improvement on the fly: the robot is skilled at recovering from its own errors mid-episode, and each fix enters its context to inform the next move. The TTT core distills a general-purpose, failure-to-correction mapping from the training data. One more thing. What excites me the most is a new Context Scaling Curve: from 128 to 8K timesteps, closed-loop performance hill-climbs steadily with no sign of saturation. 8K-context pretraining beats 1K by 62%. What LLM enjoys, robotics should too. Soon, even 1M context is not a fantasy. Deep dive in thread:"
其他 | Huanusa
- 時間: 21h
- 熱度: 👀 522,234
- 原文: 香港黑警对一个12岁的单薄小女孩下手的情景 很多事情真的不能遗忘
其他 | Nassim Taleb
- 時間: 19h
- 熱度: 👀 208,508
- 原文: History doesn't crawl; it leaps. (Forgot who said that).
其他 | Nassim Taleb
- 時間: 2h
- 熱度: 👀 103,286
- 原文: Great news. Your pizza has fewer calories than you think.
重點 | Sam Altman
- 時間: Jul 16
- 熱度: 👀 845,377
- 原文: i talk to chatgpt more than i type to it at this point new voice model really crossed a threshold
其他 | Ray Dalio
- 時間: Jul 16
- 熱度: 👀 100,448
- 原文: When you enter the third stage of your life, your joy changes. It’s no longer about being successful yourself anymore—the greatest thing you can do is help others be successful. That is why I wrote Principles in 2017. Over my life and career, every time I encountered a situation and thought about how to make a decision, I took the time to write down the criteria I would use. They became recipes—written principles that others could look at, challenge, and use to find clarity. Eventually, we even put many of these into algorithms so computers could make decisions with us. Writing them down changed everything for me. Now, I hope anyone who reads them can take these recipes and decide for themselves what works for them.
其他 | Greg Brockman
- 時間: 14h
- 熱度: 👀 90,311
- 原文: team is responding to feedback and iterating quickly. we ❤️ our users, thank you all!
其他 | Nassim Taleb
- 時間: 8h
- 熱度: 👀 87,969
- 原文: This non-cyclist friend of mine asked me to ask cyclist and non-cyclist experts in the subject to fact-check the list below.
其他 | Greg Brockman
- 時間: 22h
- 熱度: 👀 80,695
- 原文: benchmarks get saturated very quickly these days
其他 | Greg Brockman
- 時間: 14h
- 熱度: 👀 71,447
- 原文: "ChatGPT Work is great for tasks in your personal life:"
其他 | Huanusa
- 時間: 16h
- 熱度: 👀 52,643
- 原文: USC畢業典禮台灣同學手拿台灣國旗瀟灑出場、粉紅愛共者鼻子氣歪...
其他 | Balaji Srinivasan
- 時間: 7h
- 熱度: 👀 20,954
- 原文: "From another Malaysian, indeed a Johorian:"
其他 | Huanusa
- 時間: 5h
- 熱度: 👀 18,914
- 原文: 习近平最怕的事发生了! VPN法律被反杀,最高法院无奈驳回; 国家超算中心被骇背后,引发「翻墙清零」政权恐惧。
其他 | Balaji Srinivasan
- 時間: 2h
- 熱度: 👀 11,804
- 原文: The article below has an incorrect headline. We have multiple buildings. For one of them we need to change the text of a sign and the status of a license. We will do this, but the headline is just wrong. The should correct the record. thestar.com.my
其他 | Thomas Wolf
- 時間: 2h
- 熱度: 👀 3,493
- 原文: "- people are asking to release this benchmark but it’s way more significant that it’s an internal evaluation on which K3 scores so high - good writing in a controlled high taste voice so hard for LLMs in 2026… surprising given it was kinda the 1st capability unlocked in GPT 1&2"
其他 | Huanusa
- 時間: Jul 16
- 熱度: 👀 168,683
- 原文: 中國廣州一輛救護車跟多名大漢強行抓走一位年輕的女學生,女學生大喊救命我不想死。 她朋友在旁想阻止但徒勞無功...
其他 | Greg Brockman
- 時間: Jul 16
- 熱度: 👀 102,993
- 原文: "GPT-5.6 Sol Pro for resolving an important open question in statistics:"
其他 | Sam Altman
- 時間: Jul 16
- 熱度: 👀 237,859
- 原文: Article
其他 | Eric Topol
- 時間: 6h
- 熱度: 👀 41,493
- 原文: The ~25% increased risk of incident T2 diabetes with statins vs placebo. No sign that the effect is sex-specific.
其他 | Eric Topol
- 時間: 5h
- 熱度: 👀 16,331
- 原文: "This piqued my interest: \"Lymph Nodes Now Optional\" about the way dendritic cells form local lymphoid structures (acronym TLS) within cancer tissue (or tumors within TLS)"
其他 | Balaji Srinivasan
- 時間: 6h
- 熱度: 👀 100,519
- 原文: I don’t usually post DMs, but this is a rare exception as I actually asked him before doing so and he consented. Danial is a local Malaysian. Several months ago he asked “Would be cool if you could repost our current portfolio of projects as well, under KD Labs 👀”. I invited him to the Network School campus, as I do many Malaysian builders. He also said he wanted to build a virtual campus for Malaysians, a similar vision to me regarding a network state. Of course, I support him in that vision! And of course it shows that the concept of the network state, like the concept of cryptocurrency, has broad appeal across nationalities. Looking forward to Danial’s Malaysian network state.
其他 | Demis Hassabis
- 時間: Jul 16
- 熱度: 👀 771,812
- 原文: "3 years ago we started as a tiny experiment with the goal of helping you learn faster. Since then, we grew to bring audio, video, and interactivity to your sources, transitioning from a passive workspace to your true research companion. And now, notebooks have even become an entire ecosystem: you can already access them in the" "and soon in Google Search So, with these advancements, it’s time for us to evolve once again: NotebookLM is now Gemini Notebook ✨📓 The same app you know and love isn’t going anywhere, we just have an updated name that reflects our role in Google's AI portfolio. And our mission stays exactly the same: helping you learn, faster. Thank you for believing in us— this wouldn't have been possible without your passion (and feature requests...) Big things to come (yes, even folders📂!) so stay tuned. Sincerely, The Project Tailwind team"
其他 | Peter Steinberger
- 時間: Jul 15
- 熱度: 👀 1,631,338
- 原文: Introducing GPT-Red An internal automated red teamer on a mission to find our models’ prompt injection vulnerabilities at scale, helping us build stronger defenses before wider deployment.
其他 | Peter Steinberger
- 時間: Jul 16
- 熱度: 👀 214,704
- 原文: Mostly true, 5.6 is relentless.
其他 | Eric Topol
- 時間: Jul 16
- 熱度: 👀 166,990
- 原文: First pill that blocks PCSK9 approved by FDA, achieves LDL cholesterol lowering comparable to injectables (Repatha, Praluent) nytimes.com
其他 | Eric Topol
- 時間: Jul 16
- 熱度: 👀 117,172
- 原文: This is big. A vaccine to prevent pancreatic cancer in high-risk individuals
其他 | Greg Brockman
- 時間: Jul 15
- 熱度: 👀 124,205
- 原文: the new pelican test
其他 | Sean Kelly
- 時間: 4h
- 熱度: 👀 15,584
- 原文: spent the last hour shovelling water out of the basement, hope your day is going better
其他 | Sean Kelly
- 時間: 8h
- 熱度: 👀 12,654
- 原文: ok, this is actually reasonable
其他 | Peter Steinberger
- 時間: 3h
- 熱度: 👀 20,688
- 原文: We need evals on irony.
其他 | Sean Kelly
- 時間: Jul 16
- 熱度: 👀 18,787
- 原文: Aliens Can Talk To Us Through Psychedelics, Scientists Claim youtube.com