Data Scientists & Computational Researchers

OpenAI CEO Sam Altman confirmed that GPT-4's training cost exceeded $100 million, while Google's Gemini Ultra reached an estimated $191 million -- representing a 287,000x increase from the cost of training a Transformer model in 2017 ($670). These costs concentrate frontier AI development in a handful of well-resourced corporations. Academic researchers, as noted by cognitive scientist Sean Trott, find it "hard to run (and even harder to train) state-of-the-art LLMs on an academic budget, which limits the kinds of questions we can ask," creating a structural divide where publicly funded researchers cannot compete with or verify the claims of private AI labs.

Despite requiring advanced degrees (often PhDs) and expertise spanning biology, statistics, and computer science, bioinformatics professionals face severe pay disparity. The bottom 10% of biological scientists earn just $54,500 annually, while average bioinformatics scientist salaries range from $85,012 to $116,054 depending on the source -- substantially below the $141,000-$250,000 range for ML engineers with comparable technical skills but no domain science requirement. Many bioinformaticians "know they're underpaid but don't know where to start" negotiating, and geographic concentration means those outside San Francisco, Boston, and San Diego earn 20-30% less than national averages.

The ninth annual State of Open Data report from Digital Science, Figshare, and Springer Nature found that while open data is "on the edge of becoming a recognized global standard," critical equity gaps persist. Average repository sharing rates hover around 25% in wealthy nations (US, UK, Germany, France) but remain "significantly below a quarter" in Brazil, Ethiopia, and India. Researchers face unfunded mandates to share data openly while corporations commercialize those same datasets. The report warns of "a potential divide where open science becomes the preserve of better-resourced research environments, potentially marginalizing researchers in low- and middle-income countries."

A comprehensive survey found that 90-95% of researchers in the US and UK rely on research software, and more than 63% reported they could not continue their work if such software stopped functioning. Yet Research Software Engineers (RSEs) who build and maintain this critical infrastructure remain largely invisible in academia -- unable to earn authorship credit on papers, excluded from traditional promotion criteria, and lacking formal career paths. The US-RSE community has grown to 2,800 members advocating for recognition, but as a Princeton University study documented, "lack of a clear career path" remains RSEs' top concern, with 194 different job titles fragmenting the profession.

Princeton University's RSE Group documented the systemic absence of career structures for research software engineers in academia. When polled, RSE Group members identified "lack of a clear career path" as their top concern. Rapid expansion of RSE programs combined with retention challenges and limited promotion paths were "amplified by growing demand for RSEs in the private sector, which added risk of turnover." Traditional academic evaluation metrics -- publications, citations, grants -- fail to capture RSE contributions, and universities struggle to align RSE work with existing HR frameworks, forcing these professionals into ill-fitting job classifications that undervalue their expertise.

A white paper submitted to the Heliophysics Decadal Survey documented that Research Software Engineers "receive unequal treatment compared to their science counterparts, including lack of credit for their contributions and insufficient training." Despite being essential to 63% of US researchers who cannot continue their work without software, RSEs are systematically excluded from authorship, grant PI eligibility, and academic recognition systems. The paper advocates for RSEs to receive "equality of contribution" -- equivalent credit, career advancement opportunities, and institutional support as domain scientists -- arguing that the current system exploits technical labor while reserving prestige and funding for traditional research roles.

A Proof News investigation revealed that Nvidia scraped YouTube videos to train its Cosmos AI model, with VP of Research Ming-Yu Liu writing in an internal email about building "a video data factory that can yield a human lifetime visual experience worth of training data per day." Nvidia used dozens of virtual machines with rotating IP addresses to evade YouTube's detection systems. Companies including Meta, Microsoft, and Nvidia extracted over 15.8 million videos from more than 2 million YouTube channels without creator consent, prompting class-action lawsuits alleging unjust enrichment and unfair competition.

MIT Technology Review documented how AI companies have systematically scraped training data without consent or compensation, prompting a wave of lawsuits and licensing deals. The New York Times alleges millions of copyrighted articles were used to train AI models without consent. LinkedIn faces a class-action lawsuit for allegedly harvesting private messages for AI training. Reddit sued Perplexity AI for obtaining data "using false identities, proxies and other antisecurity techniques." Meanwhile, data scientists and researchers whose public datasets, code, and analyses were scraped for training have no mechanism for attribution, opt-out, or compensation -- their computational work treated as raw material for corporate AI products.