(Bloomberg) -- Chinese AI chipmaker Cambricon Technologies Corp. soared its 20% daily limit on Monday, leading a sector rally after Bloomberg News reported Beijing was stepping up pressure on domestic companies to ditch Nvidia Corp. processors for local alternatives.Most Read from BloombergA ...

China AI Chip Leader Soars 20% Limit as Beijing Warns on Nvidia

Limit

limit

‘Let It Wag!’ and the Limits of Machine Learning on Rare Concepts

Authors:
(1) Vishaal Udandarao, Tubingen AI Center, University of Tubingen, University of Cambridge, and equal contribution;
(2) Ameya Prabhu, Tubingen AI Center, University of Tubingen, University of Oxford, and equal contribution;
(3) Adhiraj Ghosh, Tubingen AI Center, University of Tubingen;
(4) Yash Sharma, Tubingen AI Center, University of Tubingen;
(5) Philip H.S. Torr, University of Oxford;
(6) Adel Bibi, University of Oxford;
(7) Samuel Albanie, University of Cambridge and equal advising, order decided by a coin flip;
(8) Matthias Bethge, Tubingen AI Center, University of Tubingen and equal advising, order decided by a coin flip. Authors: Authors: (1) Vishaal Udandarao, Tubingen AI Center, University of Tubingen, University of Cambridge, and equal contribution; (2) Ameya Prabhu, Tubingen AI Center, University of Tubingen, University of Oxford, and equal contribution; (3) Adhiraj Ghosh, Tubingen AI Center, University of Tubingen; (4) Yash Sharma, Tubingen AI Center, University of Tubingen; (5) Philip H.S. Torr, University of Oxford; (6) Adel Bibi, University of Oxford; (7) Samuel Albanie, University of Cambridge and equal advising, order decided by a coin flip; (8) Matthias Bethge, Tubingen AI Center, University of Tubingen and equal advising, order decided by a coin flip. Table of Links Abstract and 1. Introduction Abstract and 1. Introduction 2 Concepts in Pretraining Data and Quantifying Frequency 2 Concepts in Pretraining Data and Quantifying Frequency 3 Comparing Pretraining Frequency & “Zero-Shot” Performance and 3.1 Experimental Setup 3 Comparing Pretraining Frequency & “Zero-Shot” Performance and 3.1 Experimental Setup 3.2 Result: Pretraining Frequency is Predictive of “Zero-Shot” Performance 3.2 Result: Pretraining Frequency is Predictive of “Zero-Shot” Performance 4 Stress-Testing the Concept Frequency-Performance Scaling Trend and 4.1 Controlling for Similar Samples in Pretraining and Downstream Data 4 Stress-Testing the Concept Frequency-Performance Scaling Trend and 4.1 Controlling for Similar Samples in Pretraining and Downstream Data 4.2 Testing Generalization to Purely Synthetic Concept and Data Distributions 4.2 Testing Generalization to Purely Synthetic Concept and Data Distributions 5 Additional Insights from Pretraining Concept Frequencies 5 Additional Insights from Pretraining Concept Frequencies 6 Testing the Tail: Let It Wag! 6 Testing the Tail: Let It Wag! 7 Related Work 7 Related Work 8 Conclusions and Open Problems, Acknowledgements, and References 8 Conclusions and Open Problems, Acknowledgements, and References Part I Part I Appendix Appendix A. Concept Frequency is Predictive of Performance Across Prompting Strategies A. Concept Frequency is Predictive of Performance Across Prompting Strategies B. Concept Frequency is Predictive of Performance Across Retrieval Metrics B. Concept Frequency is Predictive of Performance Across Retrieval Metrics C. Concept Frequency is Predictive of Performance for T2I Models C. Concept Frequency is Predictive of Performance for T2I Models D. Concept Frequency is Predictive of Performance across Concepts only from Image and Text Domains D. Concept Frequency is Predictive of Performance across Concepts only from Image and Text Domains E. Experimental Details E. Experimental Details F. Why and How Do We Use RAM++? F. Why and How Do We Use RAM++? G. Details about Misalignment Degree Results G. Details about Misalignment Degree Results H. T2I Models: Evaluation H. T2I Models: Evaluation I. Classification Results: Let It Wag! I. Classification Results: Let It Wag! Abstract Web-crawled pretraining datasets underlie the impressive “zero-shot” evaluation performance of multimodal models, such as CLIP for classification/retrieval and Stable-Diffusion for image generation. However, it is unclear how meaningful the notion of “zero-shot” generalization is for such multimodal models, as it is not known to what extent their pretraining datasets encompass the downstream concepts targeted for during “zero-shot” evaluation. In this work, we ask: How is the performance of multimodal models on downstream concepts influenced by the frequency of these concepts in their pretraining datasets? generalization How is the performance of multimodal models on downstream concepts influenced by the frequency of these concepts in their pretraining datasets? We comprehensively investigate this question across 34 models and five standard pretraining datasets (CC-3M, CC-12M, YFCC-15M, LAION-400M, LAION-Aesthetics), generating over 300GB of data artifacts. We consistently find that, far from exhibiting “zero-shot” generalization, multimodal models require exponentially more data to achieve linear improvements in downstream “zero-shot” performance, following a sample inefficient log-linear scaling trend. This trend persists even when controlling for sample-level similarity between pretraining and downstream datasets [79], and testing on purely synthetic data distributions [51]. Furthermore, upon benchmarking models on long-tailed data sampled based on our analysis, we demonstrate that multimodal models across the board perform poorly. We contribute this long-tail test set as the Let it Wag! benchmark to further research in this direction. Taken together, our study reveals an exponential need for training data which implies that the key to “zero-shot” generalization capabilities under large-scale training paradigms remains to be found. Let it Wag! 1 Introduction Multimodal models like CLIP [91] and Stable Diffusion [96] have revolutionized performance on downstream tasks—CLIP is now the de-facto standard for “zero-shot” image recognition [133, 72, 126, 48, 132] and imagetext retrieval [46, 64, 24, 117, 129], while Stable Diffusion is now the de-facto standard for “zero-shot” text-to-image (T2I) generation [93, 17, 96, 41]. In this work, we investigate this empirical success through the lens of zero-shot generalization [69], which refers to the ability of the model to apply its learned knowledge to new unseen concepts. Accordingly, we ask: Are current multimodal models truly capable of “zero-shot” generalization? Are current multimodal models truly capable of “zero-shot” generalization? To address this, we conducted a comparative analysis involving two main factors: (1) the performance of models across various downstream tasks and (2) the frequency of test concepts within their pretraining datasets. We compiled a comprehensive list of 4, 029 concepts[1] from 27 downstream tasks spanning classification, retrieval, and image generation, assessing the performance against these concepts. Our analysis spanned five large-scale pretraining datasets with different scales, data curation methods and sources (CC-3M [107], CC-12M [27], YFCC-15M [113], LAION-Aesthetics [103], LAION-400M [102]), and evaluated the performance of 10 CLIP models and 24 T2I models, spanning different architectures and parameter scales. We consistently find across all our experiments that, across concepts, the frequency of a concept in the pretraining dataset is a strong predictor of the model’s performance on test examples containing that concept. Notably, model performance scales linearly as the concept frequency in pretraining data grows exponentially i.e., we observe a consistent log-linear scaling trend. We find that this log-linear trend is robust to controlling for correlated factors (similar samples in pretraining and test data [79]) and testing across different concept distributions along with samples generated entirely synthetically [51]. model performance scales linearly as the concept frequency in pretraining data grows exponentially model performance scales linearly as the concept frequency in pretraining data grows exponentially i.e., Our findings indicate that the impressive empirical performance of multimodal models like CLIP and Stable Diffusion can be largely attributed to the presence of test concepts within their vast pretraining datasets, thus their reported empirical performance does not constitute “zero-shot” generalization. Quite the contrary, these models require exponentially more data on a concept to linearly improve their performance on tasks pertaining to that concept, highlighting extreme sample inefficiency. In our analysis, we additionally document the distribution of concepts encountered in pretraining data and find that: • Concept Distribution: Across all pretraining datasets, the distribution of concepts is long-tailed (see Fig. 5 in Sec. 5), which indicates that a large fraction of concepts are rare. However, given the extreme sample inefficiency observed, what is rare is not properly learned during multimodal pretraining. • Concept Distribution: • Concept Correlation across Pretraining Datasets: The distribution of concepts across different pretraining datasets are strongly correlated (see Tab. 4 in Sec. 5), which suggests web crawls yield surprisingly similar concept distributions across different pretraining data curation strategies, necessitating explicit rebalancing efforts [11, 125]. • Concept Correlation across Pretraining Datasets: • Image-Text Misalignment between Concepts in Pretraining Data: Concepts often appear in one modality but not the other, which implies significant misalignment (see Tab. 3 in Sec. 5). Our released data artifacts can help image-text alignment efforts at scale by precisely indicating the examples in which modalities misalign. Note that the log-linear trend across both modalities is robust to this misalignment. • Image-Text Misalignment between Concepts in Pretraining Data: To provide a simple benchmark for generalization performance for multimodal models, which controls for the concept frequency in the training set, we introduce a new long-tailed test dataset called “Let It Wag!”. Current models trained on both openly available datasets (e.g., LAION-2B [103], DataComp-1B [46]) and closed-source datasets (e.g., OpenAI-WIT [91], WebLI [29]) have significant drops in performance, providing evidence that our observations may also transfer to closed-source datasets. We publicly release all our data artifacts (over 300GB), amortising the cost of analyzing the pretraining datasets of multimodal foundation models for a more data-centric understanding of the properties of multimodal models in the future. “Let It Wag!”. Several prior works [91, 46, 82, 42, 83, 74] have investigated the role of pretraining data in affecting performance. Mayilvahanan et al. [79] showed that CLIP’s performance is correlated with the similarity between training and test datasets. In other studies on specific areas like question-answering [62] and numerical reasoning [94] in large language models, high train-test set similarity did not fully account for observed performance levels [127]. Our comprehensive analysis of several pretraining image-text datasets significantly adds to this line of work, by (1) showing that concept frequency determines zero-shot performance and (2) pinpointing the exponential need for training data as a fundamental issue for current large-scale multimodal models. We conclude that the key to “zero-shot” generalization capabilities under large-scale training paradigms remains to be found. This paper is available on arxiv under CC BY 4.0 DEED license. This paper is available on arxiv under CC BY 4.0 DEED license. available on arxiv available on arxiv [1] class categories for classification tasks, objects in the text captions for retrieval tasks, and objects in the text prompts for generation tasks, see Sec. 2 for more details on how we define concepts.

What 300GB of AI Research Reveals About the True Limits of “Zero-Shot” Intelligence

Wat 300GB van AI-navorsing onthul oor die ware limiete van "Zero-Shot" intelligensie

Lo que 300GB de investigación de IA revela sobre los verdaderos límites de la inteligencia “zero-shot”

O que 300 GB de investigación de IA revela sobre os verdadeiros límites da intelixencia "Zero-Shot"

Mit árul el a 300 GB-os AI-kutatás a „zero-shot” intelligencia valódi határairól

Այս 300GB AI- ի հետազոտությունների մասին բացահայտվում է «Zero-Shot» ինտեգրման ճշգրտությունը

Apa 300GB Penelitian AI Mengungkapkan Tentang Batas Kebenaran Kecerdasan “Zero-Shot”

300GBのAI研究が「ゼロショット」インテリジェンスの本当の限界を明らかに

អ្វីដែល 300GB នៃការស្រាវជ្រាវ AI បានបង្ហាញអំពីតម្រូវការពិតប្រាកដនៃការសាកល្បង "Zero-Shot"

Inona 300GB ny fikarohana AI dia maneho momba ny fetran'ny "Zero-Shot" Intelligence

Ce 300GB de cercetare AI dezvăluie despre adevăratele limite ale inteligenței "zero-shot"

300GB of AI Research Reveals About the True Limits of “Zero-Shot” Intelligence – அறிவியல் அறிவியல்

Yini i-300GB ye-AI Research ibonise mayelana nemimfuneko efanelekayo ye-"Zero-Shot" Intelligence

The Limits of Automated Selfish Mining Detection

**New York, NY, USA, July 1st, 2025/Chainwire/--**The largest prediction market on Base, limitless.exchange, today announces the closure of $4M in fresh funding in a strategic round and welcomes Arthur Hayes as an advisor alongside an investment from his family office, Maelstrom. limitless.exchange limitless.exchange The funding follows a prior $3M pre-seed round led by 1confirmation, and comes after the team found breakout demand for short term prices markets on assets like BTC, which are similar in nature to 0DTE options but a much easier way to trade and feature even shorter dated expiries such as hourly, racking up over $250M in volume soon after launch. This brings Limitless' total funding raised to $7M, backed by Coinbase Ventures, 1confirmation, Maelstrom, Collider, Node Capital, Paper Ventures, Public Works, Punk DAO, and WAGMI Ventures, as well as individual investors via the Base Ecosystem Fund group on Echo. In preparation for an upcoming TGE, the team today launched a points program targeted at prediction market enthusiasts who can get skin in the game by using the product, providing liquidity, and referring their friends to join the platform. Limitless seems likely to become the first major prediction market platform to launch a token and distribute an airdrop to its early customers, marking a notable opportunity for retail traders. The team also just introduced a new mobile-first trading experience that enables people around the world to seamlessly wager on their favorite assets' performance in the next hour or day. “The future of trading is easy, fast, and powered by an army of token holders. We’re excited to bring this vision to reality,” said CJ Hetherington, CEO at Limitless Labs. About Limitless Limitless is the largest prediction market on Base with over $250M bet on unique contracts that allow users to wager on the performance of their favorite assets in the next minutes, hour or day - a net new, easy way to trade for casual users. Limitless Limitless Contact CEO CJ Hetherington Limitless Labs cj@limitless.network This story was published as a press release by Chainwire under HackerNoon’s Business Blogging Program. This story was published as a press release by Chainwire under HackerNoon’s Business Blogging Program. Program Program

Limitless Raise $4m Strategic Funding, Launch Points Ahead Of TGE

Onbeperkte verhoging van $ 4 miljoen strategiese finansiering, lanseer punte voor TGE

Incremento ilimitado de 4 millones de dólares en financiación estratégica, puntos de lanzamiento por delante de TGE

Un financement stratégique illimité de 4 millions de dollars, des points de lancement en avance sur TGE

העלאה בלתי מוגבלת של 4 מיליון דולר במימון אסטרטגי, נקודות שיגור לפני TGE

Retounen $ 4 milyon dola finansman estratejik, lanse pwen anvan TGE

Aumento illimitato di 4 milioni di dollari di finanziamento strategico, punti di lancio prima di TGE

無制限に400万ドルの戦略的資金調達、TGEの先駆け

무제한적으로 4 백만 달러의 전략적 자금 조달, TGE 앞에 출시 포인트

Onbeperkte verhoging van $ 4 miljoen strategische financiering, lanceringspunten voor TGE

Neobmedzené zvýšenie strategického financovania vo výške 4 milióny dolárov, spúšťacie body pred TGE

لامحدود $ 4 ملین استراتجی فنڈنگ، TGE سے پہلے اسٹوریج پوائنٹس

4 milyon dollar strateji financing, TGE-nin o‘zingizini bo‘ladi

Anthropic Copyright Lawsuit Tests Fair Use Limits in AI Model Training

Limits of Fair Rebate Mechanisms in Blockchain Transactions

The Black Rural South Has Limited Access to the Internet: What It's Like

LoRA's Limitations in Code and Math Tasks

Limited Gains: Multi-Token Training on Natural Language Choice Tasks

Victoria, Seychelles, June 5th, 2025/Chainwire/--Following the successful debut of the limited-edition BYDFi x Ledger Nano X hardware wallet at TOKEN2049 Dubai, global crypto exchange BYDFi and hardware wallet leader Ledger have launched a global campaign to engage crypto KOLs and content creators. limited-edition BYDFi x Ledger Nano X hardware wallet at TOKEN2049 Dubai limited-edition BYDFi x Ledger Nano X hardware wallet at TOKEN2049 Dubai BYDFi BYDFi With only 500 units available worldwide, the initiative invites Web3 participants to promote secure self-custody and mobile-first asset protection by applying to join this exclusive collaboration. Campaign Overview: Limited 500-Unit Cold Wallet Collaboration The BYDFi x Ledger campaign invites creators and KOLs to participate in a simple, open-entry initiative to receive a limited-edition wallet. To join, participants are encouraged to post on X using the hashtag #BYDFixLedger, share their thoughts on the collaboration, and submit a brief application form via the official campaign page. Selected applicants will receive the exclusive BYDFi x Ledger Nano X, along with special perks as part of the partnership program. Users can learn more and apply at: https://www.bydfi.com/en/ledger https://www.bydfi.com/en/ledger https://www.bydfi.com/en/ledger Product Highlight: BYDFi x Ledger Nano X The BYDFi x Ledger Nano X is a customized version of Ledger’s flagship cold wallet, equipped with Bluetooth, USB-C, and support for over 15,000 cryptocurrencies and NFTs. Key features include: Mobile-first design compatible with iOS, Android, macOS, and Windows
Military-grade security via CC EAL5+ certified secure element
Ledger Live integration for seamless asset tracking, staking, and transactions
Support for multi-chain and multi-asset management, including Bitcoin, Ethereum, Solana, and ERC20 tokens Mobile-first design compatible with iOS, Android, macOS, and Windows Military-grade security via CC EAL5+ certified secure element Ledger Live integration for seamless asset tracking, staking, and transactions Support for multi-chain and multi-asset management, including Bitcoin, Ethereum, Solana, and ERC20 tokens Engineered for users who prioritize portability and protection, the Nano X ensures private keys stay offline—eliminating the remote risks of custodial wallets or centralized platforms. Vision: Empowering Web3 Ownership Through Self-Custody In today’s decentralized world, where digital assets are becoming integral to both individual and institutional portfolios, self-custody has emerged as a foundational principle. BYDFi and Ledger share a vision that emphasizes secure, user-controlled asset management as the cornerstone of the Web3 movement. This collaboration promotes a future where participants in the crypto ecosystem—whether traders, builders, or educators—can confidently manage their assets with mobility, transparency, and control. Michael, Co-founder of BYDFi, commented: “True ownership starts with self-custody. Our collaboration with Ledger aims to equip users with secure, intuitive tools to manage their digital assets with confidence—anytime, anywhere.” Michael, Co-founder of BYDFi, commented: “True ownership starts with self-custody. Our collaboration with Ledger aims to equip users with secure, intuitive tools to manage their digital assets with confidence—anytime, anywhere.” About Ledger Ledger is a global leader in digital asset security, known for its hardware wallet innovations since 2015. With over 6 million devices sold and zero hacks, Ledger’s products—including the Nano series and Ledger Live app—enable millions of users and enterprises to manage their crypto, NFTs, and data securely in the Web3 age. About BYDFi Founded in 2020, BYDFi now serves over 1 million users across more than 190 countries and regions. The platform has been featured by Forbes as one of the Best Crypto Exchanges & Apps for Beginners of 2025. Best Crypto Exchanges & Apps for Beginners of 2025 Best Crypto Exchanges & Apps for Beginners of 2025 With a comprehensive product suite—including spot trading, perpetual contracts, copy trading, automated bots, and on-chain tools—BYDFi supports both novice and professional traders in navigating the digital asset market with confidence. BYDFi is committed to providing a world-class crypto trading experience for every user. BUIDL Your Dream Finance. Website: https://www.bydfi.com
Support email: cs@bydfi.com
Business partnerships: bd@bydfi.com
Media inquiries: media@bydfi.com Website: https://www.bydfi.com https://www.bydfi.com https://www.bydfi.com Support email: cs@bydfi.com Business partnerships: bd@bydfi.com Media inquiries: media@bydfi.com Twitter( X ) | LinkedIn | Telegram | YouTube | How to Buy on BYDFi Twitter( X ) Twitter( X ) LinkedIn LinkedIn Telegram Telegram YouTube YouTube How to Buy on BYDFi How to Buy on BYDFi Contact Media Manager Anna BYDFi Fintech LTD annachloe@bydfi.com This story was published as a press release by Chainwire under HackerNoon’s Business Blogging Program. This story was published as a press release by Chainwire under HackerNoon’s Business Blogging Program. Program Program

BYDFi And Ledger Launch Global Campaign with Limited BYDFi x Ledger Nano X

BYDFi og Ledger lancerer global kampagne med Limited BYDFi x Ledger Nano X

BYDFi y Ledger lanzan una campaña global con Limited BYDFi x Ledger Nano X

BYDFi ak Ledger lanse yon kampanje mondyal ak Limited BYDFi x Ledger Nano X

BYDFi と Ledger が Limited BYDFi x Ledger Nano X のグローバルキャンペーンを開始

Алдыңғы мақалаЧТО НАДО ДЕЛАТЬ ДЛЯ РАЗВИТИЯ ПРОМЫШЛЕННОСТИ?

BYDFi i Ledger uruchamiają globalną kampanię z Limited BYDFi x Ledger Nano X

BYDFi iyo Ledger la soo bandhigay warshadaha ah oo leh Limited BYDFi x Ledger Nano X

BYDFi och Ledger lanserar global kampanj med Limited BYDFi x Ledger Nano X

BYDFi ve Ledger Limited BYDFi x Ledger Nano X ile Küresel Kampanyayı Başlattı

BYDFi اور Ledger Limited BYDFi x Ledger Nano X کے ساتھ عالمی مہم کا آغاز

BYDFi và Ledger ra mắt chiến dịch toàn cầu với Limited BYDFi x Ledger Nano X

I-BYDFi kunye ne-Ledger isethula i-Global Campaign nge-Limited BYDFi x I-Ledger Nano X

Stories, News & Mentions about LIMIT

Trending Topics

Classic

Neon Noir

Minty

Newspaper

HN StartUps