All articles tagged with #training data
Handshake AI and other data-labeling firms are recruiting improv performers to help train leading AI labs, aiming to teach models to recognize and express human emotion in unscripted scenes for multimodal AI. The gigs pay around $74 per hour and are pitched as flexible, but workers warn that pay can dwindle and schedules can be unstable, raising concerns about the impact on performers’ careers as labs push toward more humanlike AI.
More than 100 researchers back a framework to treat certain biological data like sensitive health records, arguing most data should remain open while a narrow subset that could enable misuse—such as linking viral genetics to real-world traits—needs protection. They warn that training AI models on such data could lower the barrier to designing dangerous pathogens, and while legitimate researchers should have access, it shouldn’t be uploaded anonymously or browsable on the open web. The aim is to balance scientific progress with biosecurity, advocating regular reassessment of restrictions as science evolves to prevent worst-case scenarios.
Stanford and Yale researchers tested four major LLMs—OpenAI’s GPT-4.1, Google’s Gemini 2.5 Pro, xAI’s Grok 3, and Anthropic’s Claude 3.7 Sonnet—and found they can reproduce lengthy, copyrighted passages with high accuracy (Claude 3.7 Sonnet near-verbatim ~95.8%; Gemini 2.5 Pro ~76.8% on Harry Potter; Claude 3.7 Sonnet >94% on Orwell’s 1984), suggesting these models may store or copy training data rather than simply learning patterns. Some reproductions required jailbreak-style prompts (Best-of-N), underscoring potential legal liabilities as copyright lawsuits proceed and the industry debates what counts as “learning.”
Research by Anthropic and partners shows that injecting just 250 carefully crafted poison samples into training data can compromise large language models, causing them to produce gibberish or potentially dangerous outputs, highlighting vulnerabilities in AI training processes.
A new study suggests that training AI on low-quality, clickbaity content causes lasting cognitive decline in models, similar to human effects of brain rot, and cannot be easily fixed, highlighting risks of unregulated data use.
Anthropic agreed to a $1.5 billion settlement for authors whose books were used to train its AI model, Claude, with a minimum of $3,000 per book, leading the author to reconsider the value of compensation for such use.
Anthropic has agreed to pay at least $1.5 billion to authors whose pirated works it used to train its AI models, creating a precedent for AI companies compensating for illegal content use. The settlement involves establishing a fund valuing each pirated book at $3,000 and destroying the pirated materials, highlighting increasing legal risks for AI firms.
Anthropic agrees to pay at least $1.5 billion in the largest U.S. copyright settlement, compensating authors for the use of their works in AI training, highlighting legal and ethical issues in sourcing training data for AI models.
Researchers found that filtering risky content from AI training data, such as bioweapons instructions, can create safer models without harming performance, highlighting the importance of pre-training safeguards over post-training tweaks. The study emphasizes transparency and proactive safety measures in AI development, contrasting with industry secrecy.
xAI requested over 200 employees to record conversations for facial recognition training, aiming to help Grok analyze facial expressions, but faced internal resistance due to privacy concerns and the sensitive nature of biometric data, amid broader privacy law challenges.
A US judge ruled that AI training using copyrighted books by Anthropic is 'transformative' and not a copyright violation, but the case will proceed to trial over the use of pirated copies, highlighting ongoing legal debates about AI training practices.
The music industry is developing infrastructure to detect and trace AI-generated songs early in the production and distribution process, focusing on licensing and control rather than enforcement, with tools that analyze tracks for synthetic elements, attribute creative influence, and regulate training data to prevent misuse.
Adobe's Firefly AI, touted for its practicality and ethical training data, has been revealed to have been trained on Midjourney images in addition to public domain material and Adobe Stock, casting doubt on its commercial safety claims and ethical standing.
AI companies are facing a shortage of training data as they continue to build larger models, leading to the exploration of alternative sources such as publicly-available video transcripts and synthetic data. Some companies are considering controversial methods like training on transcriptions from public YouTube videos, while others are working on creating higher-quality synthetic data. Concerns about AI running out of data have been raised, but researchers believe that breakthroughs could address the issue. However, the solution may also involve reevaluating the pursuit of larger models due to environmental and resource concerns.
OpenAI continues to tease its upcoming AI video generator, Sora, with text-to-video clips that are impressing viewers, and plans to release it later this year with sound and metadata. The company is giving early access to some individuals in the film industry for testing. However, there is secrecy surrounding the training data used for Sora, with OpenAI's CTO Mira Murati being vague about its sources in a recent interview with The Wall Street Journal.