Field Kit
A consulting OS with 70+ tools across 4 MCP servers. This is a live view into the knowledge base — trends, papers, regulations, frameworks, and brand intelligence. Queryable from Claude at the desk, or from a Telegram bot on the go.
What changed this week
Threads
PrivateEnter your token to unlock the consulting agent.
Knowledge Graph
loading…
Trend Radar
Latest Papers
CFALR: Collaborative Filtering-Augmented Large Language Model for Personalized Fashion Outfit Recommendation
Yujuan Ding, Junrong Liao, Yunshan Ma +4
Personalized outfit recommendation poses a significant challenge in e-commerce and social media platforms, requiring systems that balance user preferences with aesthetic compatibility. Collaborative filtering (CF) provides a traditional solution for this, but it struggles with data-sparse scenarios and complex user-item-outfit relationships. Meanwhile, existing template-based approaches are constrained by rigid pre-designed structures. To bridge these research gaps, we introduce CFALR (Collabora
The Clustering Strikes Back: Building Cost-Effective and High-Performance ANNS at Scale with Helmsman
Yuchen Huang, Baiteng Ma, Yiping Sun +7
RedNote (a.k.a., Xiaohongshu, a global-scale social network platform) widely adopts approximate nearest neighbor search (ANNS) to power its search, recommendation, and advertising services. Due to the demanding Service Level Agreements (SLAs), we have to rely on in-memory graph-based ANNS (i.e., HNSW) to provide high throughput and low latency. However, the ever-growing user base and content volume have led to an explosive increase in memory footprint and consequently huge CapEx and OpEx. Afte
CoDeR: Local Constraint-Compatible Retrieval Beyond Semantic Similarity
Xingkun Yin, Xuebin Tang, Hongyang Du
Information retrieval systems have long treated semantic similarity as a proxy for relevance. For constraint-sensitive queries, this proxy can fail when a document is topically close to the query but supports the opposite constraint direction, such as satisfying an attribute that should be excluded or affirming a relation that should be negated. We study this failure as constraint-violating evidence exposure and propose CoDeR, a local constraint-compatible dense retrieval method that separates t
Charge as a Construct-Validity Factor in Chinese Legal Case Retrieval: A Cross-Benchmark Audit
Yao Liu, Tien-Ping Tan, Zhilan Liu
Chinese Legal Case Retrieval (LCR) benchmarks grade a reference judgment relevant when its legal characterization matches the query, and strong systems now reach NDCG@10 of 0.85-0.88. Most of the BM25-to-best-trained gap is recoverable with no retrieval model: ranking candidates only by shared primary charge, broken by BM25, closes 99.2% of it on LeCaRDv2 -- with no detectable difference from the best-trained system. This reflects benchmark design: LeCaRDv2 defines top relevance via the crime's
Trait, Not State: The Durability of Reading Identity in Social Highlighting
Kazuki Nakayashiki, Keisuke Watanabe
Prior work on a social web highlighter located individuality in selection -- which documents a person chooses to highlight -- but measured it cross-sectionally. We ask the temporal question: is a reader's selection signature a trait or a state? We freeze each reader's first six months of highlighting as a profile and track its own-vs-other advantage on their later selections at growing gaps (to 24+ months), with negatives drawn from the same calendar era -- so supply drift cannot masquerade as p
Two Wrongs, No Right: Auditing Social-Desirability Bias in LLM Annotators for Computational Social Science
Varun Kotte
LLM annotators are increasingly used in computational social science (CSS), but it is unclear whether their alignment-shaped errors preserve the empirical conclusions a researcher would report. We audit three open-source 7B instruction-tuned models (Zephyr, Mistral-Instruct, Qwen2.5-Instruct) across six TweetEval tasks under four prompt conditions (72 cells) and find that social-desirability failures do not run in a single direction. Zephyr exhibits leniency bias, systematically under-applying h
The AI Legal Specialist: A Juridically Autonomous Professional Profile for AI Governance
Nicola Fabiano
The rapid global expansion of artificial intelligence regulation has generated, across multiple jurisdictions, a demand for legal expertise dedicated to AI that the market has addressed in a fragmented manner. Data protection officers extend their remit beyond data protection law; privacy lawyers reposition themselves toward AI; compliance officers add AI chapters to their existing manuals. This paper argues that none of these adaptive responses adequately covers the professional space opened by
Divination by Prompt: LLM-Mediated Xuanxue on Chinese Social Media
Chuang Li, Lixuan Wang, Yuqi Chen +1
The rapid proliferation of large language models (LLMs) has produced a striking cultural practice: using conversational AI for divination. This paper offers one of the first systematic studies of LLM-mediated divination in the context of Xuanxue, an internet-native umbrella term for mystical and spiritual practices on Chinese social media. Using a mixed-methods design, we analyze 23000+ posts and comments from Xiaohongshu and conduct 32 semi-structured interviews with users and professional divi
The Challenges of Balancing AI Compliance and Technological Innovations in Critical Sectors: A Systematic Literature Review
Ayush Enkhtaivan, Chinazunwa Uwaoma
The rapid integration of artificial intelligence (AI) into critical infrastructure including healthcare, finance, energy, and defense, offers transformative benefits but also conflicts with evolving regulatory and governance frameworks. This paper presents a systematic literature review (SLR) to examine the challenges of balancing AI compliance and technological innovation across critical infrastructure sectors. The review follows established SLR guidelines to extract and synthesize insights fro
AI SciBrief as a Gateway to Research: A Framework for Onboarding Students into New Research Areas
Andrei Lazarev, Dmitrii Sedov
Students at all levels of higher education face a significant barrier in the form of information overload, which often paralyzes the initial stages of the research process and suppresses motivation. In response, this article introduces a pedagogical framework that leverages AI SciBrief, a platform powered by a Large Language Model (LLM) designed to automatically generate digests of scientific trends. We describe how this multidisciplinary tool - with initial coverage in finance, medicine, and ed
GeoDial: A Multimodal Conversational Tutoring Dataset for Geometry Problem-Solving with Visual Tutor Turns
Sankalan Pal Chowdhury, Junling Wang, Donya Rooein +2
Several educational domains rely heavily on diagrams and visual cues, yet most existing tutoring datasets are limited to text-only interactions. This limits the development of AI tutors that can teach in visually grounded ways used by human instructors. Thus, we introduce GeoDial, a multimodal tutoring dataset of over 1.3K teacher-student dialogs in the domain of geometry collected from experienced math teachers, where instructional turns are explicitly grounded in diagram highlights. We propose
Eigenism: Ethics for a Human-AI Future
Dan Hendrycks
Our concepts of survival and self-interest were built for single, continuous biological lives. These ideas break down when applied to artificial intelligence, since an AI can be easily copied, paused, branched, or merged. To determine what an AI actually has reason to care about, this paper introduces \textit{Eigenism}, an ethical framework that treats identity not as an all-or-nothing property tied to specific hardware, but as a graded, distributed pattern of information. We propose that an age
Who Designs the Designer? Behavioural Architecture for GenAI in Education
Sepinoud Azimi
AI in education is stuck between two failed responses: banning AI and building content-only tutors. Both fail because they ignore what decades of research has established: that personality, motivation, and emotional state shape learning outcomes as strongly as cognitive ability. This paper proposes behavioural architecture as an alternative. In the proposed architecture, the system adapts to how a student learns, not only to what they learn next. The student co-authors the record the system keep
The Khipu Problem: Institutional Legibility Under Distributed Cognition
Krti Tallam
AI governance still tends to assume that the relevant object is a bounded model or a bounded agent. That assumption is getting weaker. Real systems increasingly distribute cognition across models, tools, humans, context stores, retrieval layers, runtime policies, authorization boundaries, and delegated institutional roles. In such systems, the central governance problem is no longer only what the system did, but whether later institutions can still read what the system was. This paper introduces
An Explainable AI Assistant for Introductory Programming Education: Improving Feedback Reliability with Instructor-AI Collaboration
Muntasir Hoq, Griffin Pitts, Bradford Mott +6
Active learning is widely recognized as an effective approach for improving learning outcomes in introductory programming courses. However, insufficient instructional support often limits students' access to timely, personalized feedback, which is crucial for mastering foundational programming concepts. Although recent advances in AI, particularly large language models, offer scalable opportunities for feedback, concerns about explainability and reliability remain. In this paper, we present an A
Navigating the muddy waters of bias in artificial intelligence research: Understanding divergent meanings and conceptions
Mohammad Hossein Jarrahi, Amir Karami, Patrick Conway +2
As artificial intelligence (AI) pervades many decision-making domains, AI bias grows in importance. Although there is increasing awareness of the social and ethical consequences of biased AI, understanding bias from the perspective of those who develop these systems, such as the AI research community, is less clear. In this study, we employ topic modeling on 6520 articles to explore how the AI research community interprets the concept of bias. Our results show that the definition of bias is disp
AI-Automation Tooling in Computer Engineering Education: Mixed-Methods TAM/UTAUT Evidence for a General Acceptance Attitude
Aung Pyae
As generative AI and low-code workflow platforms become routine in software practice, a key educational question is whether the next generation of computer engineers will accept these tools as useful, usable, and worthy of sustained engagement. This paper reports a mixed-methods, cross-sectional study of undergraduate computer engineering students' acceptance of AI automation tooling, instantiated through the open-source platform n8n across three identically scripted workshops in Thailand (n = 1
A Multiplexing Design Space: Theory, Method, and Application
Yiwen Xing, Afrah Farea, Saiful Khan +1
Many visualization designs feature phenomena referred to as ``visual multiplexing'', where multiple pieces of information associated with the same data point are conveyed simultaneously. Although visualization designers are able to bring such phenomena, often unconsciously, into their designs, the design space of visual multiplexing is huge, and it is uncommon to explore visual multiplexing systematically as design patterns. In this paper, we propose a design method for exploring a smaller desig
Paper Bundles
the curated layerThe paper firehose stays large — bundles say what counts. Reviewed, rationale-bearing groups in daily use.
Foresight Practice and Scenario Quality
Use this when preparing client sensemaking, framing uncertainty, designing foresight workshops, or checking whether scenario work is actually useful rather than decorative.
RAG Reliability and Evaluation
Use this when Field Kit, AskBox, or client tools need retrieval quality, citation quality, factuality, or domain-specific RAG decisions.
Production Agentic Workflows and Evaluation
Use this when deciding whether an agentic workflow is production-worthy, how to evaluate it beyond task success, and what operational failure modes to watch.
Regulations Radar
21 tracked provisionsEU AI Act, GDPR, NIS2, and Swedish AI regulation — tracked with timelines and use-case impact analysis.
EU AI Act Article 52 - Transparency obligations for certain AI systems
Establishes transparency rules for AI systems interacting with people. Key obligations: (1) Chatbots/AI interfaces must disclose they're AI (unless obvious), (2) AI-generated content (deepfakes, synthetic media) must be machine-readable and marked as artificial, (3) Emotion recognition/biometric categorization systems must inform subjects, (4) Deployers creating deepfakes must disclose manipulation. Exceptions for law enforcement (with safeguards) and creative works.
EU AI Act Article 13 - Transparency and provision of information to deployers
Requires providers of high-risk AI systems to deliver comprehensive, clear instructions to deployers (organizations using the AI). Must include: intended purpose, accuracy metrics, limitations, known risks, human oversight requirements, data specifications, maintenance needs, and logging details. Ensures deployers can interpret outputs, use systems appropriately, and comply with oversight obligations.
EU AI Act Article 6 - Classification rules for high-risk AI systems
Defines when AI systems are classified as high-risk. High-risk designation applies to: (1) AI used as safety components in regulated products requiring third-party assessment, and (2) AI systems listed in Annex III (e.g., biometrics, critical infrastructure, education, employment, law enforcement). Provides exemptions for narrow procedural tasks, decision support (not replacement), and preparatory tasks - except for profiling, which is always high-risk.
EU AI Act Article 26 - Responsibilities along the AI value chain
Clarifies who is legally the 'provider' (liable party) when AI systems are modified, rebranded, or repurposed. Key rules: (1) If you rebrand a high-risk AI or make substantial modifications, YOU become the provider (and liable). (2) If you change an AI's intended purpose making it high-risk, YOU become the provider. (3) Original providers must cooperate with new providers, sharing documentation and technical access. (4) Providers and third-party suppliers must have written agreements specifying responsibilities. Product manufacturers are providers if they integrate AI safety components under their brand.
EU AI Act Article 14 - Human oversight
Mandates meaningful human oversight of high-risk AI systems. Humans must be able to: (1) understand system capabilities and limitations, (2) detect anomalies, (3) avoid automation bias, (4) interpret outputs correctly, (5) override/disregard AI decisions, and (6) stop the system when needed. For biometric identification (Annex III 1(a)), requires two-person verification before action. Providers must design systems for oversight; deployers must implement it.
EU AI Act Annex III - High-Risk AI Use Cases
Lists AI use cases automatically classified as high-risk under EU AI Act. Covers 8 domains: (1) Biometric identification (real-time and post), (2) Critical infrastructure (traffic, utilities), (3) Education (admissions, grading, proctoring), (4) Employment (hiring, performance evaluation, task allocation), (5) Essential services (benefits eligibility, credit scoring, emergency dispatch), (6) Law enforcement (risk assessment, polygraphs, evidence evaluation), (7) Migration/asylum (risk assessment, application review), (8) Justice (judicial decision support, election influence). High-risk classification triggers strict obligations: conformity assessment, human oversight, transparency, data governance.
Swedish AI Commission - Färdplan för AI: Recommendation on Public Sector AI Governance
Swedish AI Commission's governance framework for public sector AI use. Key requirements: (1) Clear accountability chains from operators to leadership, (2) Mandatory risk assessments for fundamental rights, bias, and transparency, (3) Human oversight with override capability, (4) Transparency obligations (inform citizens, publish AI inventories, explain decisions), (5) Procurement standards requiring EU AI Act compliance and explainability, (6) Ongoing monitoring with performance tracking and independent audits. Emphasizes human agency and democratic accountability.
GDPR Article 35 - Data Protection Impact Assessment (DPIA)
Requires Data Protection Impact Assessments (DPIAs) before processing personal data with high risk to rights/freedoms. Mandatory for: (1) Automated decision-making with legal/significant effects (e.g., AI profiling), (2) Large-scale processing of sensitive data (health, race, biometrics), (3) Systematic public area monitoring. DPIA must include: system description, necessity/proportionality assessment, risk analysis, mitigation measures. Controllers must consult Data Protection Officers, consider data subject views, and review assessments when risks change. Most AI systems processing personal data trigger DPIA requirements.
GDPR Article 22 - Automated individual decision-making, including profiling
Grants individuals the right not to be subject to solely automated decisions with legal or significant effects (e.g., loan denials, hiring decisions). Exceptions: (1) necessary for contract performance, (2) authorized by law with safeguards, (3) explicit consent. When automated decisions are allowed, controllers must provide human intervention rights, allow individuals to express views and contest decisions. Cannot use sensitive personal data (race, health, etc.) for automated decisions except in limited cases with safeguards.
Swedish Offentlighets- och sekretesslagen (Public Access to Information and Secrecy Act) - Chapter 21 Section 7: Educational Records
Swedish law establishing secrecy obligations for student personal data in public education. Information about students' personal circumstances is protected unless disclosure is necessary for: (1) their education, (2) healthcare/social services, (3) guardianship supervision, or (4) criminal investigation. Public information: registration status, final grades, disciplinary decisions. For AI systems: requires secrecy by design, strict access controls, disclosure limitations, and balancing transparency with confidentiality. Creates additional constraints beyond GDPR for Swedish public universities like Uppsala.
NIS2 Directive (EU) 2022/2555 + Swedish Cybersäkerhetslagen (SFS 2024:163)
NIS2 Directive (EU) 2022/2555 raises baseline cybersecurity across the EU and replaces the original NIS Directive. Member State transposition deadline was 17 October 2024; Sweden transposed via Cybersäkerhetslagen (SFS 2024:163) which entered into force 1 January 2025, supervised by MSB. Covers "essential" and "important" entities across expanded sectors — including research and higher education in many Member States — though Sweden's scope decisions for universities should be checked against MSB's guidance. Core obligations: risk management measures (Art 21 — policy, incident handling, business continuity, supply chain, encryption, MFA, training), incident reporting (Art 23 — 24h early warning, 72h notification, 1-month final report), governance accountability (Art 20 — management bodies are personally accountable and must take cybersecurity training), and supplier obligations. Heavy fines (€10M / 2% global turnover for essential entities; €7M / 1.4% for important). Pairs with EU AI Act (cyber resilience of AI systems is part of high-risk obligations) and GDPR Art 32 (security of processing).
Great American Artificial Intelligence Act of 2026 (discussion draft) — 3-year federal preemption of state AI-development laws
A 269-page bipartisan US House discussion draft (Reps. Jay Obernolte R-CA and Lori Trahan D-MA), released June 4, 2026 — the most comprehensive federal AI framework yet proposed in the US. It would require large/frontier AI developers to undergo semi-annual third-party audits while imposing a three-year preemption barring states from passing NEW laws governing how frontier AI models are developed. States retain authority over deployment/use. Named targets for preemption include California AB 2013 (training-data disclosure) and part of SB 942 (content watermarking); Colorado's AI Act (effective June 30) would also be frozen. Drawing intense opposition (June 6) from unions (AFT), Public Citizen, and even safety advocates (Alliance for Secure AI) who back the audit/catastrophic-risk focus but reject preempting state safeguards. Relevance for a Sweden/EU-based AI practice: not directly binding, but a leading indicator of a US regulatory model that diverges sharply from the EU AI Act's risk-tiered, deployment-focused approach. A US "light-touch federal floor + preemption" regime would widen the transatlantic compliance gap, shaping how global model providers (whose products you build on) document training data, watermark content, and disclose audits. Watch the audit-mandate language as a possible convergence point with EU obligations.
Förvaltningslagen (Administrative Procedure Act, SFS 2017:900)
Swedish Administrative Procedure Act (in force since 1 July 2018). Governs how Swedish public authorities — including Uppsala University as a state agency — handle cases and decisions affecting individuals. Establishes foundational principles every UU-built system must respect: legality, objectivity, proportionality (§5); service obligation toward citizens (§6); inter-authority cooperation (§8); right to access one's own case documents (§10); duty to handle cases without unnecessary delay (§11–12); right to be heard before adverse decisions (§25); duty to give written reasons for decisions (§32). Crucially, §28 permits *fully automated decisions* — but legality, objectivity, proportionality, the right to be heard, and the reasoning duty all still apply. This makes Förvaltningslagen the backbone for any AI-augmented decision system at UU: automation is allowed, but explainability, due process, and human appealability are not optional.
US Treasury Financial Services AI Risk Management Framework (FS AI RMF)
Treasury establishes the de facto AI governance standard for US financial services. The FS AI RMF gives banks, insurers, and fintechs a structured compliance framework for AI risk. Aligns with NIST AI RMF but adds sector-specific requirements (systemic risk, fair lending). For consulting: this is the most concrete US sector-specific AI governance framework — creates immediate demand for AI risk assessment, bias auditing, and vendor oversight in financial services.
Offentlighets- och sekretesslagen (Public Access & Secrecy Act, SFS 2009:400) — Overview
Top-level entry for the full Swedish Public Access & Secrecy Act (in force since 30 June 2009). OSL operationalises Tryckfrihetsförordningen Ch 2 by defining when secrecy applies and how it is balanced against the constitutional right of access. Key chapters for Uppsala University work: Ch 2 (secrecy regime and "skadebedömning" harm assessment), Ch 6 (documentation obligations), Ch 21 (general secrecy for personal circumstances), Ch 23 (education-specific secrecy — pairs with the existing Ch 21 §7 entry in Field Kit), Ch 39 (personnel and employment matters). For AI/data systems at UU, OSL determines which student or employee data can flow to which systems, what counts as a public document, and how access controls must be designed. Pairs with TF Ch 2 (offentlighetsprincipen), Förvaltningslagen, Arkivlagen, and GDPR — typically the source of conflict between transparency and data protection in Swedish public-sector AI work.
EU Council Agreement to Streamline AI Act — High-Risk System Rules Delayed Up to 16 Months
EU Council tacitly acknowledges the AI Act's compliance infrastructure isn't ready. High-risk system rules delayed up to 16 months while standards mature. New teeth added: explicit NCII and CSAM prohibition using AI. Buys compliance time for companies but the direction of travel doesn't change. For consulting: clients get breathing room on high-risk classification but should not slow down preparation.
US National AI Legislative Framework — Federal Preemption of State AI Laws
White House blueprint urging Congress to preempt state-level AI regulation, channel oversight through existing sector regulators, and take an innovation-first approach. Non-binding but sets the US regulatory direction. Sharp contrast to the EU AI Act's prescriptive model. Key implication: California-style AI bills (SB 1047 etc.) would be effectively neutralized if enacted. For EU-exposed clients, transatlantic compliance divergence widens significantly.
EU AI Act Article 50 — Final Code of Practice on Marking & Labelling of AI-Generated Content
On 10 June 2026 the European Commission published the FINAL Code of Practice on marking and labelling of AI-generated content. It is voluntary and gives generative-AI providers and professional deployers practical steps to comply with the AI Act's Article 50 transparency obligations, which become applicable on 2 August 2026. Two core duties: (1) providers must mark AI-generated/manipulated outputs in a machine-readable format; (2) professional deployers must clearly label deepfakes and AI-generated text published on matters of public interest. The Commission will issue further guidelines to clarify scope. For an EU-based builder of generative AI agents and client apps, this is a near-term, concrete compliance trigger: any system that produces synthetic text/image/audio/video output for clients needs a machine-readable watermarking/provenance plan (e.g. C2PA-style metadata) before August.
Tryckfrihetsförordningen Chapter 2 — Offentlighetsprincipen (Principle of Public Access)
Tryckfrihetsförordningen Chapter 2 — the constitutional layer above OSL. Establishes every Swede's right to access "allmänna handlingar" (official documents) held by public authorities, including Uppsala University. A document becomes "allmän" when it is received by or completed within an authority. The authority must hand it out promptly and free of charge unless secrecy applies under OSL. Requesters have the right to remain anonymous. For AI systems at UU, this is the *generative* law — it creates the obligation to register, find, and disclose documents (the "diarié" obligation), and it determines when AI prompts, intermediate outputs, and generated drafts become public records. Pairs with OSL (which carves out secrecy exceptions), Förvaltningslagen (procedural), and Arkivlagen (long-term preservation). The canonical Swedish public-sector design constraint and the source of most "GDPR vs. transparency" conflicts.
Arkivlagen (Archives Act, SFS 1990:782)
Swedish Archives Act (in force since 1 July 1991). Establishes that public authorities' archives — including Uppsala University's — form part of the national cultural heritage and must be preserved. Core duties: preserve allmänna handlingar (§3), organise the archive so retrieval is possible (§4), dispose ("gallra") only when explicitly permitted by Riksarkivet's rules or by a specific gallringsbeslut (§10), and hand over to the appropriate archival authority on dissolution (§9). Pairs with Tryckfrihetsförordningen Ch 2 (which creates access right), OSL (which carves secrecy), and GDPR (whose storage-limitation principle is the canonical conflict point). For AI systems at UU, this creates a retention floor that GDPR's retention ceiling cannot lower without explicit legal support. Practically: AI prompts, intermediate outputs, and generated documents that become allmänna handlingar enter the archival regime and cannot be casually deleted — they need gallringsbeslut.
EU AI Act — August 2026 enforcement milestone
The EU AI Act's full General-Purpose AI (GPAI) and high-risk system obligations reach a major enforcement milestone in August 2026. Smartr cites this explicitly in their implementation-stage checklist as a governance item every AI project must prep for. For andhumans, this matters on two fronts: 1. Any client pilot that crosses into production during or after Aug 2026 needs an AI Act readiness pass. This is a natural add-on to the feasibility/scoping deliverable. 2. Readiness checklists, model cards, transparency documentation, and risk classification are exactly the kind of "thin iceberg" artifacts a team-of-one could productize — they're repeatable, compliance-driven, and hard for clients to DIY.
Brands Tracked
Powered by 4 MCP servers · Supabase + pgvector · Telegram bot · Refreshed every minute
Full Case Study →