Scale AI
AI overview
Sign in and the AI will write an overview from our coverage.
Headlines · 19
- Outlier and AI Data Work Face Labor-Practice Scrutiny
Rest of World reports that Scale AI's Outlier advertises up to $180 an hour for U.S. physicians designing cardiac scenarios and reviewing AI-generated reports, while similar work in the Philippines and Tanzania is advertised at much lower rates. Interviews with more than 30 workers describe a recurring pattern of rising demands, shorter target times, disappearing tasks and falling pay as projects progress. The report also says a proposed settlement involving Scale AI workers received preliminary approval in San Francisco, although the company denies wrongdoing.
Hacker News · AI(100+ 分) · 🔥 0 - Menlo Ventures partner ranks 30 private AI companies above $10B
Menlo Ventures partner Deedy Das published a ranking of 30 unlisted AI companies with valuations above $10 billion and publicly disclosed annualized revenue of at least $100 million, according to 36Kr. OpenAI ranked first by valuation at about $1.25 trillion, ahead of Anthropic at about $965 billion, while Anthropic ranked ahead of OpenAI by annualized revenue. DeepSeek ranked fifth and Moonshot AI seventh, with both reporting annualized revenue of roughly $1 billion in the ranking.
36氪 人工智能 · 🔥 0 - China's Intelligent Knowledge targets expert data for AI post-training
PINGWEST reports that 智能知识 was founded in April 2025, raising seed funding from FiveYuan Capital and MiraclePlus and angel funding jointly from Jingqiu Fund and Yaotu Capital. The startup focuses on post-training data, organizing expert knowledge around reinforcement-learning environments, tasks and validators, and argues that human experience remains necessary despite advances in RSI. Its leadership includes former ByteDance Seed model-data engineering head Wang Sheyang and CTO Li Kuanye, with other team members coming from companies including Scale AI.
品玩 实时要闻 · 🔥 0 - PowerContext v1.5.0 targets token overhead in long-running agents
OceanBase released PowerContext v1.5.0 with reusable prepared context, reversible shared-JSON encoding, evaluation tools, and an alpha Rust Gateway for long-running agent workflows. In the publisher’s paired SWE-bench Pro comparison, the integration reportedly reduced cumulative input tokens by 25.4% without lowering task-completion performance. The roughly 75% reduction cited in the headline applies to a single-task input peak, not the overall benchmark total.
掘金 · 人工智能 · 🔥 0 - An HRT alumni network draws attention in Silicon Valley
A 36Kr analysis examines a network of former Hudson River Trading interns, including Alexandr Wang, Scott Wu, Jeff Yan and Jesse Zhang, who went on to AI, software, customer-service and crypto businesses. On October 19, Wang, who leads Meta Superintelligence Labs, posted a reunion photo with Wu and Yan. The report also identifies talent moving from Scale AI to Cognition and shared backing from Founders Fund, but says the group remains less commercially integrated than the PayPal Mafia. The “Silicon Valley new mafia” label is the report’s framing rather than an established description.
36氪 人工智能 · 🔥 0 - Maxinsights builds a business around human egocentric video for robotics
Maxinsights collects first-person video of everyday activities with head-mounted devices and processes it into training data for robotics models. The report says the company has accumulated 2 million hours of human egocentric data and built a global network of about 4,000 collectors. Founder Yuguang Yuan said the company has more than 30 customers and projects 2026 annual recurring revenue of about $250 million, compared with roughly $1 million the previous year. These figures were not independently verified. Yuan said scaling the collection network while maintaining data quality and diversity would require stronger infrastructure and direct management. The interview also discussed Maxinsights’ ambition to extend beyond data production into model assessment and deployment support.
钛媒体 · 🔥 0 - Wang Tao’s unconventional bets and Meta’s Muse gamble
TMTPost profiles Wang Tao’s path from data annotation and Scale AI to leading Meta’s AI organization after Meta acquired a 49% stake in Scale AI for $14.3 billion. The account says his team developed Muse as a personal AI assistant rather than following the industry’s focus on coding agents, and that the product launched on September 8. Sensor Tower’s estimate that Muse reached five million US downloads in roughly 22 days makes the strategy commercially notable, while reports of a potential Zuckerberg succession remain speculative.
钛媒体 · 🔥 0 - Leidos wins $95M DLA contract to add AI automation to Pentagon's ELMS logistics system
Leidos said Thursday it has won a $95 million, five-year contract from the Defense Logistics Agency to add AI automation and cloud tools to the Pentagon's Enterprise Logistics Management System, known as ELMS, which tracks more than 80 million items of government property worth over $140 billion across more than 50 agencies. The contract runs one year plus four optional years, through August 2031, and Leidos said the AI and cloud work is meant to let it ship updates faster and keep the system secure and running. The Virginia-based contractor, which reported about $17.2 billion in annual revenue, has supported ELMS and its predecessor system for more than 21 years.
The Next Web · 🔥 0 - Fast Company feature: elite experts are training the AI that may replace them
A Fast Company Chinese edition feature, republished by Huxiu AI, profiles the fast-growing white-collar gig workforce that probes frontier models for failures and teaches them to do better. It follows game-theory researcher Lazaros-Antonios Chatzilazarou, recruited by San Francisco AI training startup Mercor via LinkedIn, who now writes prompts, tests how models apply game theory and scores their answers as a side job. Mercor CEO Brendan Foody says the company pays up to $5 million a day to experts on its roster of more than 100,000, and that demand runs three to four times its supply capacity. The piece also covers Surge AI, founded by Edwin Chen in 2020 and bootstrapped past $1 billion in revenue with reported clients including OpenAI, Google, Anthropic and Meta, plus Scale AI's Outlier network, Ethos, Wirestock and Shift; it notes concerns about unpaid assessments and deactivated accounts, and an IMF estimate that generative AI will touch about 40% of jobs globally.
虎嗅 AI · 🔥 0 - Scale AI's HSS benchmark tests AI 'sixth sense'; top model hits 53.6% vs humans' 93.1%
Scale AI and Elorian released Humanity's Sixth Sense (HSS), a benchmark that tests whether models can infer what is not explicitly stated in an image or clip — whether another car fits in a gap, why a woman chasing a bus suddenly slows down, or what number belongs on a blurred beaker label. The set holds 522 open-ended tasks drawn from 288 images and 234 videos totaling 17.6 hours, split into temporal/causal dynamics, physical and spatial reasoning, social understanding, and abstract/contextual inference. Twenty human participants averaged 93.1%, while GPT-6 Astra topped the model leaderboard at 53.6% even at maximum reasoning effort, followed by GPT-6.1 Sol (46.6%), Claude Opus 5.5 (44.6%), Gemini 3.8 Flash (41.6%) and Claude Fable 5.1 (40.8%); the median across 25 models was 30.9%. Attributing 8,573 failures, the team says 94% stemmed from perception or latent-inference errors rather than flawed logic, and wrapping models in Claude Code or Codex agents improved scores on a 388-task subset but barely dented 2D-occlusion-read-as-3D-alignment mistakes. The benchmark, scores and human baseline come from the releasing organizations and have not been independently reproduced.
新智元 · 🔥 0 - Nvidia releases 120B Nemotron 3.5 Super VL and two Bolt edge models
Nvidia released Nemotron 3.5 Super VL, an open-weight multimodal mixture-of-experts model with 120 billion total and 12 billion active parameters, aimed at always-on agents and combining reasoning, instruction following and tool use with text, image, document and video understanding. Weights are available on Hugging Face in BF16 and NVFP4, organizations can post-train it with Nvidia NeMo on their own data and workflows, and both vLLM and SGLang announced day-0 serving support. Nvidia also introduced the Nemotron 3.5 Bolt line: the 74B-A9B 24GB, a 74-billion-parameter, 9-billion-active model compressed with 2-bit quantization to fit 24GB of memory, is available now, while a roughly 4GB Bolt E4B is planned for October 23. Nvidia claims up to 5x the throughput of comparable open models on agentic tasks, an 80.8% CharXiv chart-reasoning score (84% with Python code execution), and that Dynamo can serve up to 8x as many concurrent users with 85% lower time to first token under specified conditions; these figures are vendor-reported and not independently verified.
vLLM Blog · 🔥 0 - Physical AI funding tops $48B in 2026, beating all of last year
Citing PitchBook data, the Financial Times reports that robotics and physical AI companies had raised nearly $48 billion globally by October 8, 2026 — roughly 1.7 times the whole of last year. Capital is concentrating in leaders: Anduril raised $5 billion in May at a $61 billion valuation, NEURA Robotics announced a C round of up to $1.4 billion in June, Skild AI raised $1.4 billion led by SoftBank in January, and Physical Intelligence raised $1 billion in the second quarter. Robot training data is also becoming its own industry: Figure's Index program has collected over 16 million videos and paid out $15 million, Dyna Robotics says its Dyna-2 model, pretrained on more than 1 million hours of human video, lifted average performance across 14 real tasks from 20% to 53%, and Skild AI reported annual recurring revenue above $100 million about 10 months after its first commercial deployment.
量子位(原生 RSS) · 🔥 0 - Alexandr Wang on startup management and recruiting AI talent at Meta
Scale AI founder and current Meta chief AI officer Alexandr Wang, speaking on David Senra's podcast on October 11 local time, advised first-time founders to personally take on every task and master the details before gradually delegating. Using Scale AI as an example, he said direct founder control over execution was one factor in the company's success, but the cost was that every decision ran through him, making him a bottleneck. At Meta he has changed approach: because Meta Superintelligence Labs must complete a large amount of work in a very short time and he describes himself as neither an AI researcher nor a product designer, he has shifted his focus to recruiting top talent across specialist areas and building an environment where they can apply their expertise. The article notes Wang became Meta's chief AI officer in June 2025, heading the newly formed Meta Superintelligence Labs, while Meta invested $14.3 billion in Scale AI for a 49% non-voting stake.
IT之家 AI · 🔥 0 - Salesforce renames AIForce to SIForce, following Trump's 'super intelligence' push
Salesforce CEO Marc Benioff said Friday that the company is renaming its AIForce agent platform to SIForce; the change covers the platform only, not the company name, Business Insider reported. It follows Elon Musk saying last Sunday that SpaceXAI would become SpaceXSI, and both moves track Donald Trump's push to replace the term "artificial intelligence" with "super intelligence" — a 29 September executive order told federal agencies to make the switch in official documents and websites, though it does not apply to private firms. Business Insider noted that renaming a product does not show it meets the superintelligence bar, and firms with AI in their names, such as OpenAI and Scale AI, did not respond.
The Next Web · 🔥 0 - Why Google is betting on RSI, starting with Gemini 4 Argon
On Sept 30 Google released Gemini 4 Argon, the first model in its Gemini 4 series, the same day a Google DeepMind researcher posted that the model had "surpassed" him. The piece traces Google's September RSI steps — Gemini 3.8 Flash, Dream-RSI and RRSI — plus AlphaEvolve's use inside data centres, chips and AI training, arguing the company wants AI to help build its next models. Co-founder Sergey Brin is reported to be steering more resources toward RSI, as Alphabet raised its 2026 capex guidance to $195–205 billion.
36氪 人工智能 · 🔥 0 - CSIS: U.S. federal AI contracts reach $4.1B, over half lack governance clauses
A CSIS review of SAM.gov solicitations and USAspending contract records found that U.S. federal agencies signed 2,255 identifiable AI-related contracts in fiscal years 2019–2025, with cumulative obligated amounts of $4.1 billion; the Department of Defense accounted for about 82.8% of obligations and 63.8% of contracts, and annual obligations reached $1.162 billion in FY2025. Among the 75 AI-related solicitations sampled, 41 (54.7%) contained no confirmed governance clause; where clauses existed they centred on performance measurement — benchmarking appeared in 37% and data rights in 35% — while audit access, red-teaming and incident reporting appeared in just 9%, 5% and 4%, and monitoring and human-oversight clauses each appeared in 5%. On the supply side, 1,250 vendors held contracts, 66.2% of them only one, the top three took 30.4% of net obligations, and Palantir and Scale AI together held under 1% of contracts but about $597 million, roughly 15% of obligations.
虎嗅 AI · 🔥 0 - Inside the AI data boom: from labeling to RL environments
In this Silicon Valley 101 podcast episode, He Yunzhong, who works on post-training and evaluation research at Scale AI, and Sun Yiyou, a UC Berkeley postdoc on the Agents’ Last Exam (ALE) project, unpack the fast-growing but opaque market for AI training data. They note AfterQuery was valued at $300 million in its April A round and, per media reports, $3.2 billion by September, while Bespoke Labs announced $40 million in combined seed and A funding in July, and trace the shift from crowd labeling and pre-training toward rubrics, verifiable signals and RL environments bundling tasks, tools and verification. They also discuss procurement bottlenecks in vertical domains, the blurred line between benchmarks and data sales — including a claim that some data firms sold evaluation data — expert fabrication and contamination, and OpenAI's February decision to stop reporting SWE-bench Verified scores. Guest views are personal.
36氪 人工智能 · 🔥 0 - China’s AI Data Firms Draw Funding as Post-Training Demand Surges
Chinese AI training-data and evaluation companies are attracting renewed investor interest as model developers seek more post-training data. 品玩 reported that Alibaba may lead a roughly $300 million round for UniPat at a valuation near $2.5 billion, with Tencent and Sequoia China also expressing interest, but the financing and final terms remain unconfirmed. The reports also detail rising demand for expert-generated tasks, interaction environments, and scoring criteria, alongside concerns about data quality, privacy, and buyers’ return on investment.
品玩 实时要闻 · 🔥 0 - Enterprise AI Deals Depend on Deployment, Not Just Models
An analysis of a resurfaced 2024 interview with Scale AI founder Alexandr Wang argues that Palantir’s enterprise AI growth depends on customer perception, on-site engineering and trust-building as much as on software. It says OpenAI and Anthropic are developing similar deployment capabilities, including frontline deployment engineers and partnerships intended to embed AI systems into regulated enterprises.
钛媒体 · 🔥 0
Experience and discussion from the community
Share my Scale AI experienceAsk about Scale AI
Nobody has shared their experience with Scale AI yet.