The Hallway Track

No. 024

Rosalind Workbench, Claude for scientists, LLM edits alter meaning

Google released Co-Scientist last month. Since then, Anthropic opened 10,000 subsidized seats for researchers and OpenAI launched a life-sciences workbench with its own domain model. All three frontier labs now sell a scientist-branded product line, making this a product category rather than a series of one-offs. What none of the product pages address is that grammar-only LLM edits shift content and meaning. A team from Princeton and UCSD reported a nearly 70 percent increase in neutral essay responses when they applied AI edits marketed as purely mechanical.

Scientist Products

  • Meet Rosalind Workbench: Empowering every scientist to be their own research team

    OpenAI, August 28 2026

    A dedicated life-sciences environment inside ChatGPT, built on GPT-Rosalind (a domain-specific model for biochemistry, genomics, and chemistry), with molecular viewers, sequence alignment tools, and provenance tracking attached to each result, though the entire system runs within a single closed vendor.

  • Claude Team plan for scientists

    Anthropic, August 27 2026

    Free standard seats and $15/month premium seats for up to 10,000 PIs at nonprofit research institutions, bundled with Claude Science (launched June 30), Code, and connectors to PubMed, bioRxiv, and ChEMBL, with eligibility restricted to academic and nonprofit labs and excluding industry R&D.

  • freephdlabor: open-source multiagent framework for autonomous scientific research

    GitHub (ltjed), August 2026

    An MIT-licensed multiagent framework that orchestrates specialized agents (ideation, writeup, experiments) to produce papers end-to-end, positioning itself as the open-source counterpart to the frontier-lab bundles now forming a category.

Research Practice

  • How do LLMs affect writing? Talks by visitors to the AI STORIES project

    Center for Digital Narrative, University of Bergen, September 4 2026

    Five talks on measured effects of LLMs on research writing, with the key finding from Abdulhai (Princeton) and White (UCSD/Microsoft Research) that grammar-only LLM edits cause a nearly 70 percent increase in neutral essay responses, meaning what looks like a light copyedit is actually a content change.

  • 98 out of 100: humans solved one of the two math problems AI could not, at a Leipzig benchmark

    Czech Technical University in Prague, August 2026

    At a Leipzig advanced mathematics benchmark, AI models solved 98 of 100 problems, and a team from CTU Prague and Universität Leipzig cracked one of the two remaining (a combinatorics case involving breaking triples with permutations), a useful contrast to the 30 percent pass rate on open-ended research tasks from Terminal-Bench Science.

  • AI-Assisted Tools in the CHI 2027 Papers Review Process

    ACM CHI 2027 organizing committee, August 29 2026

    CHI 2027 will deploy AI for submission completeness checks, reviewer matching, and desk-reject screening, but keeps humans responsible for every acceptance decision and deliberately shelved a rubric tool that would have scored paper content, drawing a clear line between automation for logistics and automation for judgment.