data.world reposted this
Principal Scientist & Head of AI Lab at data.world; co-host of Catalog & Cocktails, the honest, no-bs, non-salesy data podcast. Scientist. Interests: Knowledge Graphs, AI, LLMs, Data Integration & Data Catalogs
Excited to be back CDOIQ again. I'm giving a talk on Wednesday and the takeaway is: Invest and treat knowledge as a first class citizen (the WHAT) But Why? 1. Because we are doing the same thing over and over again (more data, more tech) and expecting different results. That's Einstein's definition of insanity. 2. We need to find a balance between efficiency and resilience in order to address the known use cases of today (efficiency) and the unknown uses cases of tomorrow (resilience) 3. Operational Efficiency: by being efficient and resilient, we can reduce cost, improve productivity, be agile and flexible to deal with those unknown use cases, optimize resources But haven't we been already been investing in knowledge? 1. Yes, but by chance: we find issues when folks complain 2. Yes, but in an ad-hoc manner: it's not someones dedicated task 3. Yes, but inconsistent investment: ups and downs to deal with data quality But why NOW? Simple: (Gen)AI! This is an opportunity to ride the hype, in order to address the pains we've been constantly been complaining about. If it's not now, then when??? How do you address the trust issues for AI: Accuracy, Explainability and Governance? By investing in Knowledge! How to treat knowledge as a first class citizen? People, Process and Technology, of course :) - Tech: Knowledge Graphs! Your first knowledge graph is your data catalog which brings together the technical and business metadata, building the brain of your organization. Our lab has been pioneering the research on studying how to use Knowledge Graphs to increase LLM accuracy, specifically for structured data. So if the "why now" is (Gen)AI, then this is the motivator to invest in knowledge graphs which provides the context of your organization. - Process: Follow a methodology. Success is defined by business questions, not technical questions. I'll be talking about our Pay as you go methodology and Iron thread approach. - People: Have the right people in the room: Business, Data and .. Knowledge Engineers! I'm also excited to host another #HonestNoBS dinner Wednesday night. It's going to be 6 months since we did the last one so it will be great to see how the data world has changed in the last half year. If you are not aware of these dinners, check out the report I wrote about the dinners I hosted in 2023. We can possibly squeeze a few more folks in, so if you are a data leader/executive and want a dinner under chatham house rules, comment below or ping me directly. Links in comments
Papers about our research on increasing accuracy of LLMs for question answering using Knowledge Graphs https://arxiv.org/abs/2311.07509 https://arxiv.org/abs/2405.11706
I heard this morning that the content update that Crowdstrike added was a genai chatbot. That caused the worldwide computer outage. En. Ai needs to be cancelled an open ai shutdown until its war criminal musk step down
Keep it up! We need more and more people repeating this so that eventually it becomes the reality 😇
So glad you are at this symposium! Great post. One of these days I want to make it to one of your honestnobs dinners!
FOMO level 100
This sounds like my kind of talk Juan. I wish I could be there!
Go, Juan!
Principal Scientist & Head of AI Lab at data.world; co-host of Catalog & Cocktails, the honest, no-bs, non-salesy data podcast. Scientist. Interests: Knowledge Graphs, AI, LLMs, Data Integration & Data Catalogs
1moHonest No BS Dinner Wednesday Night https://page.data.world/wwj-boston Recap report for the 2023 dinners https://data.world/mstatic/assets/pdf/datw_honest_recap_ebook_l1r3.pdf