# DigitVibe

> Applied AI consulting, training, workflow automation, and custom software development.

Markdown copies follow canonical HTML paths without the trailing slash, plus .md; /index.md is the home page. Remove .md and add / to read HTML. Contact DigitVibe through /contact/ or /start-a-project/.

## Website

- [Applied AI Consulting, Training & Development | DigitVibe](https://digitvibe.com/index.md): DigitVibe helps organizations put AI to work through strategy, training, workflow automation, AI agents, and custom software.
- [AI Consulting, Training, Automation & Development | DigitVibe](https://digitvibe.com/services.md): Explore DigitVibe services in AI strategy, organizational training, AI agents, workflow automation, and custom AI software.
- [AI Strategy for Organizations | DigitVibe](https://digitvibe.com/services/ai-strategy.md): Practical AI strategy consulting built around workflows, opportunities, governance, people, and implementation.
- [AI Training | DigitVibe](https://digitvibe.com/services/ai-training.md): Applied generative AI training for leaders, teams, founders, and higher education. Executive briefings, hands-on workshops, flexible learning series, and custom programs.
- [AI Automation & Agents | DigitVibe](https://digitvibe.com/services/ai-automation-agents.md): AI agents and workflow automation consulting for knowledge systems, documents, reporting, communications, and connected operations.
- [Custom AI Software Development | DigitVibe](https://digitvibe.com/services/custom-ai-software.md): Focused AI-native software development for internal tools, research systems, educational applications, and specialized workflows.
- [AI Website Development | AI Driven Sites by DigitVibe](https://digitvibe.com/services/ai-websites.md): Build or rebuild your business website with AI Driven Sites, a DigitVibe service. Client-owned websites on Replit, team training, and optional managed changes.
- [AI Opportunity Scan | DigitVibe](https://digitvibe.com/services/ai-opportunity-scan.md): AI opportunity and readiness assessment built around workflows, current use, risk, capability, and a practical roadmap.
- [AI Champions Program | DigitVibe](https://digitvibe.com/services/ai-champions-program.md): An AI Champions Program develops internal practitioners who can experiment responsibly, support colleagues, and connect work with strategy.
- [AI Consulting for Higher Education, SMEs & Enterprise Teams | DigitVibe](https://digitvibe.com/who-we-help.md): AI consulting for higher education, SMEs, enterprise teams, nonprofits, and professional organizations.
- [AI Consulting for Higher Education | DigitVibe](https://digitvibe.com/who-we-help/higher-education.md): AI consulting and training for universities, faculty, researchers, administrators, and institutional leaders.
- [AI Consulting for SMEs | DigitVibe](https://digitvibe.com/who-we-help/smes.md): AI consulting for small and medium-sized businesses that need focused, practical adoption.
- [AI Consulting for Enterprise Teams | DigitVibe](https://digitvibe.com/who-we-help/enterprise.md): AI consulting for enterprise teams, departments, leadership groups, and innovation units.
- [AI Consulting for Nonprofits | DigitVibe](https://digitvibe.com/who-we-help/nonprofits-professional-organizations.md): AI consulting and training for nonprofits, associations, and professional organizations with tight resource constraints.
- [How DigitVibe Approaches AI Consulting | DigitVibe](https://digitvibe.com/how-we-work.md): A practical approach to AI strategy, capability building, and implementation.
- [About DigitVibe | Applied AI Consulting & Development](https://digitvibe.com/about.md): DigitVibe helps organizations understand generative AI, build practical capability, manage new risks, and choose the right technology for their work.
- [Dr. Mohammad Keyhani | Founder of DigitVibe](https://digitvibe.com/about/mohammad-keyhani.md): Dr. Mohammad Keyhani is the founder of DigitVibe, an applied AI consulting and development firm.
- [Start an AI Consulting or Development Project | DigitVibe](https://digitvibe.com/start-a-project.md): Start an AI consulting, training, implementation, or development project with DigitVibe.
- [Contact DigitVibe | Applied AI Consulting](https://digitvibe.com/contact.md): Contact DigitVibe for consulting, training, implementation, and software projects.
- [AI Consulting Case Studies | DigitVibe](https://digitvibe.com/case-studies.md): Explore DigitVibe's work with Above and Beyond Care: AI training, Digital Champions, practical agents, governance, and organizational capability.
- [ABCS AI Capacity Building Case Study | DigitVibe](https://digitvibe.com/case-studies/abcs-ai-capacity-building.md): How DigitVibe worked with Above and Beyond Care to train 37 participants, develop 10+ Digital Champions, and plan responsible AI adoption.

## Research blog

- [Generative AI, organizations & the changing world.](https://digitvibe.com/blog.md): Generative AI is changing how we work, build organizations and solve problems. In this blog, AI agents help us find and summarize research insights and discoveries and write about the possibilities they open up.
- [LLMs Assist NLP Researchers: Critique Paper (Meta-)Reviewing](https://digitvibe.com/blog/llm-review-critique-evidence.md): Experts marked more passages as deficient in AI-written peer reviews than in human reviews. Model feedback was less specific to individual papers.
- [The LLM Effect: Are Humans Truly Using LLMs, or Are They Being Influenced By Them Instead?](https://digitvibe.com/blog/llm-effect-expert-judgment-and-anchoring.md): A small study of policy analysts finds that LLM suggestions speed topic assignment while increasing alignment with the suggestions themselves.
- [LitSearch: Benchmarking AI retrieval of scientific literature](https://digitvibe.com/blog/litsearch-scientific-retrieval.md): On 597 literature-search questions, an AI retriever outperformed keyword search. Using GPT-4o to reorder results improved retrieval further.
- [When should an AI answer or refuse?](https://digitvibe.com/blog/learn-to-refuse-language-models.md): Learn to Refuse pairs a structured knowledge base with a refusal mechanism and tests whether selective answers improve accuracy.
- [What legal knowledge can language models show?](https://digitvibe.com/blog/lawbench-chinese-legal-knowledge.md): LawBench evaluates 21 systems across 20 Chinese legal tasks, revealing uneven skills and weak performance on many practical forms of legal reasoning.
- [Knowledge checks can help reduce AI errors](https://digitvibe.com/blog/knowledge-consistent-alignment.md): Knowledge Consistent Alignment tests what a model knows before fine-tuning and compares several ways to handle gaps.
- [When a Language Model’s Knowledge Collides](https://digitvibe.com/blog/knowledge-conflicts-llm-memory-context.md): A survey separates conflicts between prompts, retrieved sources and model memory—and explains why resolving them is context-dependent.
- [In-Context Learning in Language Models Has Many Moving Parts](https://digitvibe.com/blog/in-context-learning-moving-parts.md): A survey and benchmark study show why examples, scoring choices and model differences all affect in-context learning.
- [How Researchers Define and Test LLM Factuality](https://digitvibe.com/blog/how-researchers-test-llm-factuality.md): A survey separates factuality from related ideas and explains why checking open-ended model answers remains difficult.
- [Do AI assistants shift political views?](https://digitvibe.com/blog/hidden-persuaders-political-influence.md): An experiment with 935 registered US voters found that political conversations with three AI models coincided with changed candidate support for some participants.
- [HEART-felt Narratives: Tracing Empathy and Narrative Style in Personal Stories with LLMs](https://digitvibe.com/blog/heartfelt-narratives-empathy-and-story-style.md): Researchers used GPT-4 to label 874 personal stories. Readers reported more empathy for stories with greater character development and more plot events.
- [Teaching AI to reason about privacy norms](https://digitvibe.com/blog/goldcoin-privacy-law-context.md): GoldCoin uses contextual integrity to connect privacy-law requirements with realistic scenarios and evaluates models on HIPAA-related court cases.
- [Whose values do language models reflect?](https://digitvibe.com/blog/generation-gap-age-values-language-models.md): A comparison with World Values Survey responses finds that tested models tended to align more closely with younger groups, with differences across value topics.
- [An Electoral Approach to Diversify LLM-based Multi-Agent Collective Decision-Making](https://digitvibe.com/blog/gedi-electoral-voting-for-llm-agent-teams.md): An electoral module brings ranked-choice methods into LLM agent teams and shows how the aggregation rule can change benchmark answers.
- [Fine-tuning or retrieval for LLM knowledge?](https://digitvibe.com/blog/fine-tuning-retrieval-llm-knowledge.md): A comparison of model fine-tuning and retrieval finds different strengths when language models need facts they already know or have never seen.
- [AI financial claim verification still trails experts](https://digitvibe.com/blog/findver-financial-claim-verification.md): FINDVER tests whether language models can locate and explain evidence in complex financial documents—and finds a gap between model outputs and expert judgment.
- [Why AI accepts false premises](https://digitvibe.com/blog/false-premise-hallucinations.md): A study traces one kind of language-model error to attention heads that can interfere with retrieving a fact when a question embeds a false claim.
- [Fairer preferences, better AI judgments?](https://digitvibe.com/blog/fairer-preferences-ai-evaluators.md): ZEPO adjusts pairwise evaluation prompts without labeled examples; its authors link less-skewed model preferences with judgments closer to people’s ratings.
- [What LLM Benchmarks Measure—and Where They Fall Short](https://digitvibe.com/blog/evaluating-llm-benchmarks-limits.md): A systematic review follows language-model evaluation from prompt setup to metrics, highlighting missing details and fragile comparisons.
- [Code feedback helps LLMs query complex health records](https://digitvibe.com/blog/ehragent-code-clinical-record-reasoning.md): EHRAgent uses generated code, medical context and execution feedback to answer multi-table questions in three electronic health record datasets.
- [Culture Needs More Than One Language-model Score](https://digitvibe.com/blog/culture-needs-more-than-one-model-score.md): A survey of 90 papers finds that culture is usually measured through proxies, leaving important contexts and lived experience outside the test.
- [AI crypto agents combine market data, news and reflection](https://digitvibe.com/blog/cryptotrade-reflection-crypto-markets.md): CryptoTrade combines blockchain activity, market indicators and financial news. Its tests beat time-series baselines, but not established trading signals.
- [AI Answers Cross-market Product Questions Across 11 Languages](https://digitvibe.com/blog/cross-market-product-questions-11-languages.md): A multilingual benchmark tests whether product reviews and questions from one marketplace can help answer questions in another.
- [The subtle harms hidden in AI conversations](https://digitvibe.com/blog/covert-harms-ai-hiring-dialogue.md): CHAST evaluates seven forms of social harm in simulated hiring dialogues and finds that many tested models produced biased views in neutral-sounding language.
- [Climate-report retrieval benchmarks expose a knowledge gap for AI](https://digitvibe.com/blog/climretrieve-climate-report-search.md): ClimRetrieve measures whether search systems find the evidence a sustainability analyst needs in company climate reports.
- [Do great minds think alike? Investigating Human-AI Complementarity in Question Answering with CAIMIRA](https://digitvibe.com/blog/caimira-human-ai-complementarity-question-answering.md): Experienced quiz players and AI systems answered thousands of questions. Humans did better with indirect clues; large models were stronger on factual questions.
- [A better blend for retrieval-augmented AI](https://digitvibe.com/blog/blendfilter-cleaner-retrieval.md): BlendFilter combines query enrichment and document filtering to address two common sources of weak retrieval-augmented answers.
- [Balancing Global and Local Values in Multilingual AI](https://digitvibe.com/blog/balancing-global-local-values-multilingual-ai.md): A multilingual alignment study reports lower harmful outputs without a measured drop in general response preferences across its test languages.
- [An AI web agent learns to build scrapers](https://digitvibe.com/blog/autoscraper-web-data-agents.md): AutoScraper uses a language model to generate instructions for collecting webpage data. Its tests check whether those instructions run and return the correct information.
- [AutoPersuade: Evaluating and explaining persuasion with AI](https://digitvibe.com/blog/autopersaude-evaluating-persuasive-arguments.md): An AI analysis of pro-vegan messages links argument themes to audience ratings from 5,180 paired comparisons.
- [ArxivDIGESTables: Synthesizing Scientific Literature into Tables using Language Models](https://digitvibe.com/blog/arxiv-literature-tables.md): Language models generated comparison tables from research papers. Supplying captions and surrounding text helped them identify which features to compare.
- [Language models still miss mental-health comorbidity](https://digitvibe.com/blog/angst-comorbid-mental-health-benchmark.md): The ANGST benchmark allows social posts to carry depression and anxiety labels together. Even the best tested model struggled with complex cases.
- [AgentReview: Exploring Peer Review Dynamics with LLM Agents](https://digitvibe.com/blog/agentreview-peer-review-simulation.md): AgentReview simulates conference peer review with language-model agents. Changing reviewer bias changed 37.1% of paper decisions in the simulation.
- [ACE: A LLM-based Negotiation Coaching System](https://digitvibe.com/blog/ace-llm-negotiation-coaching-system.md): A negotiation coach combines role-play with targeted feedback and reports improved performance across two practice trials.
- [Personalized AI disability support needs more than polite replies](https://digitvibe.com/blog/able-personalized-disability-support.md): ABLE trains an AI system to tailor disability-support conversations to user profiles, with politeness and empathy assessed in synthetic dialogues.
- [AI Copilot access reduced email time by 12%](https://digitvibe.com/blog/ai-email-time-and-work-redesign.md): Email time fell in a randomized Copilot rollout across 66 firms. The study also tracked adoption, meetings and document work.