Scale AI

The article warns about the dangers of granting excessive permissions to sensitive files, noting how setting a document to “anyone with the link” can lead to data exposure, as seen when a Pageloot contractor inadvertently exposed internal credentials. Other instances mentioned include Ateam leaving personal data open for years and Scale AI having training material accessible publicly on Google Drive.

AI development is shifting away from chatbot training on internet text toward using reinforcement learning environments (RL) to teach agents how humans work. Tech companies are creating realistic digital workplaces where AI can practice tasks like coding and running business software. Major players include Meta, which tracks employee activity, and Google, which is considering a large investment in Mechanize, a startup that builds virtual training environments for AI agents.

Scale AI has appointed Francis deSouza as its new CEO. He previously served as COO and president of security products at Google Cloud, and his role begins on August 10.

Alexandr Wang, former Scale AI founder and current Meta leader, argues AI enables startups to compete with established giants by reducing resource barriers. He and others like Sam Altman and Vinod Khosla suggest AI agents and tools now allow small companies to outperform larger incumbents in tasks like coding and customer support.

Boris Cherny, creator of Claude Code, argues companies should measure AI success by human hours saved rather than token usage. He outlines a framework for tracking AI returns in enterprise workflows and highlights the shift from token-centric metrics as businesses reassess AI costs. Executives from Coinbase, Vercel, JPMorgan, and OpenAI have also discussed strategies for optimizing AI spending and value.

Instagram CEO Adam Mosseri emphasizes three key traits in employees: being driven, self-aware, and quick learners. He highlights adaptability as crucial in the age of AI, noting that these traits remain central despite technological changes. Other tech leaders, including those from Scale AI and Lovable, share similar priorities for employee qualities.

Anthropic's Boris Cherny argues companies should prioritize AI's return on investment (ROI) while allowing employees to experiment with AI tools. He emphasizes that controlled experimentation fosters innovation and that cost management should occur after identifying viable use cases.

Mode Inc, an AI training startup, has acquired Trimbox and QR Code Reader, expanding its user base to 100 million monthly users. The company pays consumers for everyday data like emails, photos, and receipts to supply consented datasets for AI labs, distinguishing itself from gig-focused competitors.

Coinbase, Block, and Crypto.com have cited AI as a reason to cut staff, but Scale AI's Jason L. Droege believes companies are using AI as an excuse. The companies are using AI to justify recent staff cuts. This trend is observed in the crypto industry.

Scale AI CEO Jason Droege emphasizes the importance of reliability in AI systems, stating that current AI technology is often too unreliable for mission-critical use by businesses, military, and government. Droege unveiled his mantra, 'The Reliability Race', to focus on making AI reliable enough for customers to see a return on investment. Scale AI aims to bridge the gap between expectations and reality in AI adoption.

Meta has launched the Muse Spark model, its first release under the Superintelligence Labs led by Alexandr Wang. The model represents a major restructuring of Meta's AI development efforts.

Scale AI, a company 49%-owned by Meta, is employing gig workers to train AI by scraping Instagram accounts, harvesting copyrighted content, and transcribing pornographic material. The platform Outlier is used to recruit experts in fields like medicine and physics for AI development, raising ethical concerns about data collection practices.