About
I am a highly experienced leader in AI/ML who knows how to create models that answer…
Activity
6K followers
-
David Whiting shared thisDavid Whiting shared thishttps://lnkd.in/gN4NMx3JH2O AI launches H2OGPT and LLM Studio to help companies make their own chatbotsH2O AI launches H2OGPT and LLM Studio to help companies make their own chatbots
-
David Whiting shared thisDavid Whiting shared this"Last year our CEO asked us to kind of do a state of the AI evaluation across AT&T. In that evaluation, what we found is that we were driving about $2.5 billion of value with AI that on an annualized basis across AT&T. The other great thing about it is that we found about 50% of that value came from our team--and this is a great thing--and about 50% of that value came from other parts of business. And so that shows the value, right? That we are democratizing this out, you know, that we have the competency across the business. And I think that's only the beginning, right? Things are becoming exponential in terms of what we can achieve. And we are rolling out generative AI as fast as we can that I think next time we do that evaluation, which will be here in another month, I think we'll talk $3 billion of value." -- Andy Markus, Chief Data Officer, AT&T #AI #generativeai #H2OWorldIndiaClip from Fireside Chat with Andy Markus, AT&T & Sri Ambati, H2O.aiClip from Fireside Chat with Andy Markus, AT&T & Sri Ambati, H2O.ai
-
David Whiting shared thisDavid Whiting shared thisSee how we at H2O.ai are democratizing the making of Large Language Models with #h2oGPT! https://lnkd.in/ghcSeYUw We open-sourced the largest and quite possibly the best instruction-tuned 20B LLM with Apache 2.0 license, for commercial use (no LLaMa or ShareGPT contamination): https://lnkd.in/gWU8H9sX Learn more about: - Data preparation for fine-tuning (2k LOC and growing) - LLM evaluation frameworks - Prompt engineering for best results - Fine-tuning with LoRA/8-bit on commodity hardware - ChatBot code - Our LLM roadmap Try the live ChatBot demo: https://gpt.h2o.ai/ Clone (and star :) ) our GitHub repos: https://lnkd.in/gtaHYvdb and https://lnkd.in/gZPqnJUr Get the model/data from Hugging Face: https://lnkd.in/gq99d7Zy Join us on Discord: https://lnkd.in/gWP7SRui Thanks for sharing! And check out our blog on H2O LLM Studio for no-code fine-tuning! https://lnkd.in/guE-Ze8x #GPT #ChatGPT #OpenSource #LLM #H2OBuilding the World’s Best Open-Source Large Language Model: H2O.ai’s JourneyBuilding the World’s Best Open-Source Large Language Model: H2O.ai’s Journey
-
David Whiting shared thisDavid Whiting shared thisTop Google engineer admits tech giant is falling behind in the A.I. arms raceTop Google engineer admits tech giant is falling behind in the A.I. arms race
-
David Whiting shared thisDavid Whiting shared thisUseful article on #h2oGPT - from H2O.ai, the leader in open source AI/ML. Watch this space! We have some big announcements coming your way on safe, private, and responsible #llms for healthcare & life sciences. The future of #ai & #generativeai in health requires more than what’s available currently - and we are proud to be working with the world’s leaders in health & care on expanding our #healthai community & movement. Sri Satish Ambati
-
David Whiting shared thisDavid Whiting shared thisWhat are some of the legal considerations when you're working with AI? According to Patrick Hall, Principal Scientist at BNH.AI, there are 7 key areas you should think about. #KnowledgeFirstAI #AIEthics #AILaw
-
David Whiting shared thisDavid Whiting shared thisAn educational synthetic credit decision dataset is made available in PiML. The dataset includes demographic information for the purpose of bias evaluation. This is an educational dataset to illustrate the end to end model exploration, development and validation in PiML including comparing black box model with post hoc explainability vs. inherently interpretable ML models, causal feature selection and various model testing in PiML: model weakness identification, reliability/uncertainty, robustness and resilience (distribution drift), model bias/fairness. https://lnkd.in/exDUvFde Special thanks to many contributors: Aijun Zhang Vijay Nair Zebin Yang Ningzhou Zeng Yu SU Nick Schmidt Patrick Hall Nengfeng Z. Anwesha Bhattacharyya Yaqun Wang Runze LiGitHub - SelfExplainML/PiML-Toolbox: PiML (Python Interpretable Machine Learning) toolbox for model development & diagnosticsGitHub - SelfExplainML/PiML-Toolbox: PiML (Python Interpretable Machine Learning) toolbox for model development & diagnostics
-
David Whiting shared thisDavid Whiting shared thisWow, very direct words of warning from the Information Commissioner's Office regarding emotion detection algorithms! "The information commissioner has warned companies to steer clear of “emotional analysis” technologies or face fines, because of the “pseudoscientific” nature of the field." "It’s the first time the regulator has issued a blanket warning on the ineffectiveness of a new technology, said Stephen Bonner, the deputy commissioner, but one that is justified by the harm that could be caused if companies made meaningful decisions based on meaningless data." #aiethics #airegulation #artificialintelligence h/t Bogdan Grigorescu and Robert Baugh Ryan Carrier, FHCA, Merve Hickok https://lnkd.in/gp2Hipu3Information commissioner warns firms over ‘emotional analysis’ technologiesInformation commissioner warns firms over ‘emotional analysis’ technologies
-
David Whiting liked thisDavid Whiting liked thisI clustered 3,887 insights & analytics jobs from last week into the fewest job descriptions that represent most of the market... Four hypothetical job descriptions, or 'job archetypes' represent 94% of every I&A role posted right now. 1. The Technical Data Scientist (39% of the market): ML, Python, statistics. ~4+ years experience. AI skills already required in 37% of these. This one seems pretty straight forward. It is basically your typical data scientist. 2. The Senior Analyst (26%): SQL, Tableau, Power BI. The classic. And notably (or strangely)... AI is largely absent here. Only 18% even mention it. The interesting thing here was the need for technical skills. There are a lot of people who were senior analysts 10 years ago who only knew Excel. The tech requirements are higher now, and I suspect they'll continue to grow... or this role (as it currently is) will disappear. 3. The Strategic Insights Manager (22%): The least technical and closest to the decision, and yet the roles ask for ability to 'create dashboards' or ability to 'automate production so you spend more time on judgment.' The top requirements: communication (72%), stakeholder management (67%), leadership (54%). But don't get too comfortable non-tech people... the interesting thing here is the 'automation' piece. They want you to have skills that allow you to automate the technical part of your job so you can focus on the strategic part. In other words, there isn't going to be some tech team to do that for you... you have to be self-sufficient, which is a big part of the game in this labor market. 4. The AI Product & Strategy Lead (13%): This one didn't exist a few years ago. 65% of these postings require AI skills outright. Not just "familiarity with"... building with it. Two things jump out at me. TECH SKILL IS BECOMING PRICE OF ADMISSION FOR NON-TECH JOBS First, this is the same story as my last post told from a different angle: the insights archetype closest to business decisions is the one defined by influence, not technique. And yet, the technical requirement is seemingly become the price of admission... the job is what you do with it. NEW ROLES ARE EMERGING; OLD ONES MAY BE DYING Second, AI isn't spreading evenly across our profession. It's concentrated, creating an entirely new job archetype at one end (1 in 8 I&A jobs!) while barely touching the traditional analyst role... yet. Time series data coming soon... If that LinkedIn projection holds, archetype #2 is where the disruption lands. Traditional analysts will become nearly obsolete in the "insights & analytics" space. Those types of roles will become more 'strategic' and require specific functional skills like marketing. Said another way: I suspect a marketer may be expected to be more data-savvy and analytical while the analytics person is expected to become more strategic and technical. More on this at TMRE and Michigan State. Come say hi.
-
David Whiting liked thisDavid Whiting liked thisA very personal and thoughtful film grew from a single statement, "I was MVP of the Super Bowl, but I didn't know who I was." Director Jeremy Williams followed that lead to document an intimate journey by Hines Ward to a powerfully emotional destination. I'm big on BOTH product AND process, and Jeremy has been exemplary on both fronts - a pleasure to work with and a great watch!
-
David Whiting liked thisDavid Whiting liked thisWe've never shared this data publicly before. 5 real clients. Real numbers. Nothing hidden. Client 1: SaaS CEO Before: 2,100 followers. Zero inbound. After (6 months): 84,000 followers. $1.8M pipeline. Client 2: FinTech Founder Before: 900 followers. Doing cold outreach daily. After (7 months): 63,000 followers. $2.3M pipeline. Client 3: Agency Owner Before: 4,500 followers. Posting inconsistently. After (5 months): 52,000 followers. $940K pipeline. Client 4: PE Partner Before: 1,800 followers. Invisible in his space. After (8 months): 143,000 followers. $4.1M pipeline. Client 5: Tech CEO Before: 3,200 followers. Getting 200 views per post. After (6 months): 91,000 followers. $1.7M pipeline. Combined results: 12,500 followers became 433,000. $10.8M in total pipeline generated. Average time to first enterprise deal: 3.2 months. None of them wrote a single post. We're sharing this because we're tired of "Linkedln growth" agencies that show one cherry-picked result. These are 5 consecutive clients. Not our best. Not our worst. Just the last 5 people who said yes. That's the difference between a system and a fluke. We don't get lucky. We engineer results. If you're a founder or executive sitting on expertise that nobody sees, you already know what's possible. The question is whether you'll be next or watch your competitor take the spot.
-
David Whiting liked thisDavid Whiting liked thisBig thanks to Gaspar Gonzalez and Castor Fernandez from Hammer and Nail Productions for bringing us this fun David (Bama) vs Goliath (Texas) saga full of soon to be Major League stars. Spoiler alert: Goliath wins.
-
David Whiting liked thisDavid Whiting liked thisMr. President: two civilizations are being destroyed, not one, but only one will rise with pride — the one that chooses honor and dignity over subjugation. Power can level cities, but it cannot command legitimacy; it cannot manufacture respect. In the end, what survives isn’t the force that compels obedience, but the one that refuses to surrender its humanity, even under pressure.
-
David Whiting liked thisDavid Whiting liked thisAnthropic wrote about a way to decrease RAG retrieval error rate by 67%. It’s called Contextual Retrieval and it improves RAG by adding meaningful context to each chunk before embedding and indexing with BM25. First, for each chunk of your document, a concise explanation of key details is generated. For example— In a medical study database, a chunk "Patient A showed symptoms of fatigue" becomes -- "In a 2022 clinical trial, Patient A (Group 1, receiving Drug X) showed symptoms of fatigue." These details situate every single chunk within the whole document. While creating the vectordb, embeddings are calculated on these contextualised chunks. Then, BM25 is also applied to capture exact keyword matches. BM25 is a ranking function that improves text retrieval by measuring word relevance using TF-IDF while also accounting for doct length and saturation effects. At query time, both methods are used to retrieve relevant chunks, which are ranked, deduplicated, reranked and passed to the model. Contextual Retrieval reduces retrieval failures by 49%, and combining it with reranking boosts the reduction to 67%. Adopting AI? Build a no-code AI tool in < 1 min on https://www.miskies.app ♻️ Share it with anyone who’s building AI agents :) I share tutorials on how to build + improve AI apps and agents, on my newsletter 𝑨𝑰 𝑨𝒈𝒆𝒏𝒕 𝑬𝒏𝒈𝒊𝒏𝒆𝒆𝒓𝒊𝒏𝒈: https://lnkd.in/gaJTcZBR Link to the full article is in the comments. #AI #LLMs #RAG https://lnkd.in/gubfPfvN
-
David Whiting liked thisDavid Whiting liked thisCLAUDE dot md files don't scale beyond modest codebases. If you're building serious software with Claude Code or any agentic tool, a single AGENTS dot md will eventually fail you. This paper shows what comes next. A 1,000-line prototype can be fully described in a single prompt. A 100,000-line system cannot. The AI must be told, repeatedly and reliably, how the project works, what patterns to follow, and what mistakes to avoid. Single-file manifests hit a ceiling fast. This new paper, Codified Context, documents a three-tier infrastructure built during real development of a 108,000-line C# distributed system across 283 sessions over 70 days. The system uses a three-tier memory architecture: a hot-memory constitution (660 lines, always loaded), 19 specialized domain-expert agents (9,300 lines total) invoked per task, and a cold-memory knowledge base of 34 specification documents (~16,250 lines) queried on demand via an MCP retrieval server. Across 283 sessions, this produced 2,801 human prompts, 1,197 agent invocations, and 16,522 autonomous agent turns, roughly 6 autonomous turns per human prompt, with a knowledge-to-code ratio of 24.2%. Crucially, none of it was designed upfront: each new agent and specification emerged from a real failure, a recurring bug, an architectural mistake, a convention forgotten, and was codified so it could never require re-explanation again, turning documentation into load-bearing infrastructure that agents depend on as memory, not reference.
Experience & Education
Publications
-
Machine Learning Methods for Detecting Patterns of Management Fraud
Computational Intelligence
See publicationDiscovery of financial fraud has profound social consequences. Loss of stockholder value, bankruptcy, and loss of confidence in the professional audit firms have resulted from failure to detect financial fraud. Previous studies that have attempted to discover fraud patterns from publicly available information have achieved only moderate levels of success. This study explores the capabilities of recently developed statistical learning and data mining methods in an attempt to advance fraud…
Discovery of financial fraud has profound social consequences. Loss of stockholder value, bankruptcy, and loss of confidence in the professional audit firms have resulted from failure to detect financial fraud. Previous studies that have attempted to discover fraud patterns from publicly available information have achieved only moderate levels of success. This study explores the capabilities of recently developed statistical learning and data mining methods in an attempt to advance fraud discovery performance to levels that have potential for proactive discovery or mitigation of financial fraud. The partially adaptive methods we test have achieved success in a number of complex problem domains and are easily interpretable. Ensemble methods, which combine predictions from multiple models via boosting, bagging, or related approaches, have emerged as among the most powerful data mining and machine learning methods. Our study includes random forests, stochastic gradient boosting, and rule ensembles. The results for ensemble models show marked improvement over past efforts, with accuracy approaching levels of practical potential. In particular, rule ensembles do so while maintaining a degree of interpretability absent in the other ensemble methods.
Languages
-
English
Native or bilingual proficiency
-
German
Professional working proficiency
Other similar profiles
Explore more posts
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content