Weekly Recap (January 2, 2026)

NSF reorg, Science in 2050, 2025 LLM recap, R+Python, R Data Scientist, virtual cells, genomics in 2026, Claude Code course, AI and labor, how uv got so fast, Anthropic/biotech, papers+preprints

Science: The National Science Foundation just had a big reorganization. Here are five things to know.

Nature: Science in 2050: the future breakthroughs that will shape our world.

Simon Willison: 2025: The year in LLMs.

Boaz Barak at Less Wrong: You will be OK. Barak argues that while society should take low-probability AI catastrophes seriously, individuals are better served by focusing on what they can control, expecting AI-driven change to be fast but uneven and gradual rather than a single “AGI moment,” and staying engaged, optimistic, and psychologically steady without becoming complacent.

Emily Riederer: R + Python: From polyglot to crosspolination.

Andrew Carroll: The Virtual Cell Will Be More Like Gwas Than Alphafold.

From Andrew’s post: The “virtual cell” as currently discussed centers on the transcriptome, with several large public datasets of a few cell lines with genetic or drug perturbations, with some datasets overlaying spatial data (spatial transcriptomics) and some including other imaging.

Genomics in 2026: Trends and Predictions from the Fulcrum Genomics Team: AI in Genomics: Powerful, but Not a Magic Wand; Standards Cross the Academia–Industry Divide; The Genomics Stack Grows Up: Tools, Workflows, and Rust; From Short Reads to Structural Truth: Long Reads, SVs, and Error-Corrected Sequencing; Beyond the Genome: Proteins Take Center Stage; Looking Ahead: Making Genomics Work in the Real World.

Anthropic launches a free short course on using Claude Code. In the course you’ll:

• Understand the underlying architecture of Claude Code, the tools it uses to navigate your codebase, and how it stores memory across sessions.

• Explore and understand the codebase of a RAG chatbot and how information flows between the frontend and the backend.

• Initiate a CLAUDE.md file inside your project directory containing information and guidelines about your codebase that Claude Code can remember across sessions.

• Get context into Claude Code by mentioning the relevant files and providing screenshots or images, and control the context using escape, clear, and compact commands.

• Add features to the frontend and backend of the RAG chatbot: ask Claude Code to plan first to improve its performance, use thinking mode for harder tasks, and brainstorm ideas using Claude Code’s subagents.

• Write tests to evaluate the RAG chatbot functionalities, and refactor parts of the chatbot.

• Use git worktrees to run multiple Claude sessions simultaneously, each focused on adding an independent feature to the chatbot.

• Fix Github issues, and create, review and merge Github pull requests using Claude Code’s Github integration.

• Execute code before and after using tools through Claude Code hooks.

• Refactor a Jupyter notebook for e-commerce data analysis and transform it into a dashboard.

• Connect Claude to the Figma MCP server to import a design mockup to Claude Code, and develop a web interface showing economic data from the Federal Reserve Economic Data.

• Use Playwright MCP server to automatically open a web browser, take screenshots, and guide Claude Code to improve the UI design of an application.

Sal Khan: A.I. Will Displace Workers at a Scale Many Don’t Realize.

The R Data Scientist 2025-12-30: R ecosystem radar, R package engineering, hands-on R analysis, stats & inference, academic research.

Andrew Nesbitt: How uv got so fast. (See also my series on uv, part 1, 2, 3, 4).

When A.I. Took My Job, I Bought a Chain Saw.

Is Anthropic becoming a biotech company? They’re advertising life sciences positions.

Finally, a few other papers and preprints that caught my attention this week:

• A SNP panel for co-analysis of capture and shotgun ancient DNA data

• R2G2: A Python-R Framework for Seamless Integration of R/Bioconductor Tools into Galaxy

• (Nucleotide Transformer v3) A foundational model for joint sequence-function multi-species modeling at scale for long-range genomic prediction (pdf)

• ColabPCR: A validated Google Colaboratory Notebook for Reproducible and Precise Primer Design

• CRISPR-HAWK: Haplotype- and Variant-aware guide design toolkit for CRISPR-Cas

• Clair3-RNA: a deep learning-based small variant caller for long-read RNA sequencing data

• Creating Enforceable Biosecurity Standards for Nucleic Acid Providers

• Artificial intelligence and synthetic biology: biosecurity risks, dual-use concerns, and governance pathways

• Paleogenomic insight into the collapse, recovery, and management of American bison

Related: my post from yesterday on how I find blogs, news stories, and papers like what’s in this post.