September 18, 2025

Dear Reader,

Welcome to the 292nd edition of the Data Science Briefing!

We’re proud to announce that a brand new Data Visualization with Python on-demand video is now available on the O’Reilly website: Python Data Visualization: Create impactful visuals, animations and dashboards. This in depth tutorial is almost 7h in length and covers fundamental and advanced usage of matplotlib, seaborn, plotly and bokeh as well as tips on how to use Jupyter widgets. Check it out!

The latest blog post on the Epidemiology series is also out: Demographic Processes. In this post we explore how to include birth and death rates in your epidemik models. Check it out!

This week’s picks show AI maturing from flashy demos to hard choices about governance, infrastructure, and pedagogy: funders are now piloting algorithms to triage grant proposals, promising speed, but raising fresh worries about bias and opacity in how “promising science” gets defined. In the classroom, Google’s “Learn Your Way” turns static textbooks into adaptive, interactive study paths and reports improved learning outcomes in early tests by Google Research.

On the policy front, Meredith Whittaker warns that agentic AI could erode privacy and competition by normalizing cross-app data access and automation without robust safeguards. Meanwhile, the era of unchecked scraping looks to be ending as publishers and infrastructure providers push licensing regimes and emerging standards, tightening access to training data.

Inside academia, Nature flags a new tool spotting undisclosed LLM-generated text in manuscripts and peer reviews—evidence that disclosure norms still lag practice. And under the hood, engineers are tackling inference nondeterminism itself, tracing variance to concurrency and floating-point quirks and proposing ways to make outputs reproducible at scale.

This week’s research traces a provocative arc from “making models think” to managing the messy worlds they enter: reinforcement learning is being retooled to reward multi-step reasoning rather than short-horizon shortcuts, while a comprehensive survey maps the emerging toolbox for Large Reasoning Models to push reliability beyond prompt tricks.

On the foundations side, new bounds on optimal time estimation in stochastic processes suggest hard limits for any system attempting to maintain stable clocks, particularly for inference scheduling and reproducible evaluation. Biology presents both promise and peril, as genome language models can now co-design viable bacteriophages, foreshadowing programmable therapeutics and raising thorny biosecurity questions.

Socially, the line between utility and intimacy blurs as large-scale analyses of “AI companionship” communities reveal attachment patterns that product teams and policymakers can’t ignore. Meanwhile, adversaries exploit the same cognitive hooks: pig-butchering scams follow a predictable lifecycle ripe for automated detection, and “LLM hacking” exposes how seemingly benign annotation workflows can be subverted by prompt injection and data poisoning.

The overall takeaway: better reasoning isn’t enough; we need robust timing, safety, and governance to match the new capabilities.

Our current book recommendation is Mark Carrigan’s “Generative AI for Academics ”. You can find all the previous book reviews on our website. In this week’s video, we have a video comparing Machine Learning vs Human Learning and how They’re Not Alike.

Data shows that the best way for a newsletter to grow is by word of mouth, so if you think one of your friends or colleagues would enjoy this newsletter, go ahead and forward this email to them. This will help us spread the word!

Semper discentes,

The D4S Team