Lighthouse AdvisorySLED AI Adoption Intelligence

Higher Education & ResearchIssue 204 records

Research Universities

Four newly archived sources juxtapose same-day European academic AI/HPC capacity (KIT HoreKa 2) with measured autonomous-lab gains from evidence grounding (Andromeda 2), cautionary agent claim-reproduction rates (RECLAIM at UIUC/NCSA), and university research-admin AI controls (Georgia Tech interim policy). Distinguishes accelerator inaugurations and hit-rate benchmarks from reproducible scientific progress and institutional governance readiness.

Evidence records
4
Cross-source patterns
1
Evidence classes
2 academic research1 independent reporting1 standards or public-body guidance
Outcomes
2 emerging1 mixed1 cautionary
Source freshness
3 new this fortnight1 older, newly relevant
Research completed
2026-09-27

Choose a role to see its takeaway beside every record in the ledger.

Patterns across the evidence

Synthesis · Lighthouse Advisory interpretation

1 pattern, each supported by at least two sources

  1. AI/HPC capacity and research agents need evidence grounding, claim-reproduction audits, and approved-tool governance

    Same-day KIT HoreKa 2 expands shared academic AI/HPC (300 AI accelerators, ~33 PFLOPS planned, operator-stated doubling of sim/AI capacity). Same-week RECLAIM shows best agents recover only 41%/27%/15% of NeurIPS’25 claims by artifact tier under independent execution audit. Andromeda 2 shows evidence-grounded agentic labs can lift matched-budget hit rates (50% vs 17%/2%) when historical experimental evidence is available. Georgia Tech’s interim research AI policy supplies the institutional control layer (approved tools, protected-data limits, human oversight, disclosure). Operating implication: fund capacity together with evidence stores, sealed evaluation, and research-admin AI governance—not GPUs or agent demos alone.

    Operating questionWhen expanding campus AI/HPC or deploying research agents, which artifact-release tiers, execution-based audits, evidence-store requirements, and approved-tool/disclosure controls must be funded before treating capacity or agent outputs as scientific progress?

    Supporting evidenceKarlsruhe Institute of Technology (KIT) / NHR@KIT Scientific Computing CenterIntrepid Labs (Toronto) — arXiv preprintUniversity of Illinois Urbana-Champaign / National Center for Supercomputing Applications (NCSA)Georgia Institute of Technology (Office of the Provost / EVPR)

Evidence ledger

Full record · every source keeps its link and limitations

4 records
  1. Independent reportingEmergingNew this fortnight

    KIT inaugurates HoreKa 2: >42k cores, 300 AI accelerators, ~33 PFLOPS; doubles sim/AI compute (€17M)

    On 25 September 2026 KIT inaugurated HoreKa 2, a €17 million NHR-accessible supercomputer with more than 42,000 processor cores, 300 AI accelerators, and planned peak performance of about 33 PetaFLOPS. SCC director Martin Frank states the system doubles compute capacity for simulations and AI applications. Hot-water cooling and 100% renewable Campus Nord power are cited for efficiency; GPU partition supports AI workloads across earth system, engineering, materials, particle physics, and other disciplines. Phased rollout: CPU Onyx began earlier; GPU Jade (NVIDIA Blackwell) expands AI capacity.

    Karlsruhe Institute of Technology (KIT) / NHR@KIT Scientific Computing CenterGermany (Karlsruhe)September 25, 2026

    Why it matters, evidence and limitations
    Why it matters
    Campus and national research-computing leaders evaluating AI/HPC upgrades need capacity, efficiency, and shared NHR-style access claims with clear limits—this is an inauguration with architecture specs, not measured queue SLOs or scientific discovery outcomes.
    Evidence and measured results
    KIT press release dated 25 September 2026 opened via HTTPS; core/accelerator/PFLOPS, €17M cost, doubling claim, and sustainability statements taken from the opened German page.
    Limitations and uncertainty
    Institutional self-report; PFLOPS is planned peak; 'doubling' is operator claim without published baseline table; no queue-wait or energy-efficiency ranking metrics in this page; German national context limits direct U.S. transfer.
    University press release (KIT) Role takeaways & implementation
  2. Academic researchMixedNew this fortnight

    Andromeda 2: evidence-grounded agentic lab hits 50% high-AUC formulations vs 17% / 2% baselines (paclitaxel)

    Andromeda 2 is an agentic formulation system that reasons over structured in-house experimental evidence and drives a miniaturized autonomous SEDDS lab. At a matched budget of 96 paclitaxel formulations, it achieved a 50% high-performance hit rate versus 17% for probabilistic Andromeda 1 and 2% for wet-lab DoE; 12 vs 6 vs 0 formulations met all four TPP objectives. Median AUC10–240 was 70.1 / 12.0 / 3.5 mg·min/mL. Ablation: withholding historical in-house evidence cut mean AUC by 34% (62.0→46.2). Authors note FaSSIF is not bioavailability; single campaign limits inference; industry lab setting.

    Intrepid Labs (Toronto) — arXiv preprintCanada (Toronto); transferable methods with limitsSeptember 16, 2026 (arXiv)

    Why it matters, evidence and limitations
    Why it matters
    University and national-lab AI-for-science programs evaluating autonomous labs need measured hit-rate gains from evidence grounding—not agent demos alone—and must separate in-vitro assay gains from clinical claims.
    Evidence and measured results
    arXiv abs 2609.19099 opened via HTTPS; hit rates, TPP counts, median AUCs, and 34% ablation taken from abstract/abs page text (paper dated 16 Sep 2026).
    Limitations and uncertainty
    Industry (Intrepid Labs) not a university lab; FaSSIF ≠ oral bioavailability; single campaign per arm; preprint not peer-reviewed journal version; paclitaxel-specific design space.
    Academic preprint (arXiv) Role takeaways & implementation
  3. Academic researchCautionaryNew this fortnight

    RECLAIM: best agent reproduces 41% / 27% / 15% of NeurIPS’25 claims by artifact tier (Run/Retrain/Reimplement)

    RECLAIM benchmarks whether agents can reproduce pinned central claims from 100 NeurIPS 2025 papers using author-released artifacts, graded by a separate LLM auditor on execution evidence (not agent self-reports). Best agent per tier reproduces only 41% of Run-tier (code+data+weights), 27% Retrain (no weights), and 15% Reimplement (no code). Failed attempts use on average 29% of GPU budget. Common failure: reimplement without checking against paper numbers (63/400 runs). Authors release harness, rubric, and graded transcripts; human–auditor F1 0.82 on reproduced verdict.

    University of Illinois Urbana-Champaign / National Center for Supercomputing Applications (NCSA)United States (Illinois)September 23, 2026 (arXiv)

    Why it matters, evidence and limitations
    Why it matters
    Research computing and research-integrity leaders deploying AI research agents need independent, execution-based reproduction rates by artifact availability—not leaderboard demos—before treating agent 'papers' as scientific progress.
    Evidence and measured results
    arXiv abs 2609.28850 opened via HTTPS; tiered reproduction rates, budget underuse, and failure-mode counts taken from abstract text (submitted 23 Sep 2026).
    Limitations and uncertainty
    Preprint; one venue/year; at most one run per cell; no human reproduction on same cluster for misses; possible contamination on released-code tiers; closed frontier models omitted for cost.
    Academic preprint (arXiv) Role takeaways & implementation
  4. Standards or public-body guidanceEmergingNewly relevant · Jun 2026

    Georgia Tech interim AI Academic & Research policy: approved tools, protected-data limits, human oversight, disclosure

    Georgia Tech’s interim policy governing AI in Academic and Research Contexts (effective June 2026) requires community members to use Institutionally Approved or Local-POC Limited Use AI tools for academic/research work; prohibits submitting Protected Data or PII into tools outside approved scope; mandates human oversight/verification of AI outputs before reliance; and requires appropriate disclosure of AI use in scholarly communications. An AI Governance Officer and Committee coordinate tool approval and points of contact. Why now: as campuses expand AI/HPC capacity and research agents show low claim-reproduction rates, research-admin control layers become operationally binding—not optional guidance.

    Georgia Institute of Technology (Office of the Provost / EVPR)United States (Atlanta, GA)June 2026 (effective); review December 2026

    Why it matters, evidence and limitations
    Why it matters
    Research administration adopting AI agents and shared AI tools needs enforceable approval, data-handling, oversight, and disclosure controls that transfer to other R1 governance designs.
    Evidence and measured results
    Georgia Tech Policy Library page opened via HTTPS; effective date June 2026, tool-approval, protected-data, human oversight, and disclosure provisions taken from the opened policy text.
    Limitations and uncertainty
    Interim policy without published compliance metrics; effective June 2026 (older than last edition) included for newly relevant control-layer why-now; does not evaluate scientific agent performance.
    University policy (Georgia Tech) Role takeaways & implementation

How to read this edition

Source findings, measured results and limitations come from the cited publications. Patterns, operating questions, role takeaways and implementation considerations are Lighthouse Advisory interpretation, stated as questions to validate locally rather than guaranteed outcomes. Vendor and operator claims are labeled as claims. Full research method.

Academic research
Research produced through an academic institution or peer-reviewed venue.
Independent reporting
Independent reporting with attributable sources but without a formal evaluation design.
Standards or public-body guidance
Normative or advisory guidance from a standards body or public institution.