Gen AI Buy vs Build, Commercial vs Open SourceClaimOrganizations may use different language models for different use cases, which creates additional data-governance and risk-management challenges.25:50
Guardrails, governance and security in 2024
39 sessions
Gen AI Buy vs Build, Commercial vs Open Source
PodcastData Governance and AIClaimAlexandra Diem says sensitive customer data makes data governance a major barrier to buying generative-AI products from external vendors.46:43
PodcastManaging Data for Effective GenAI ApplicationClaimLanguage models can hallucinate answers even when the requested data does not exist, so their outputs require guardrails and human oversight.29:22
PodcastA Decade of AI Safety and TrustPushed backPetar argued that robustness verification cannot by itself establish that a machine-learning model is trustworthy because data and labeling problems must be addressed first.26:16
Security and PrivacyClaimDiego Oppenheimer says privacy and security discussions are important because data scientists may import personally identifiable information into scripts despite model security measures.1:21
Building the Next Generation of Reliable AIClaimShreya Rajpal says retrieval-augmented generation grounds responses in trusted data but does not provide strong reliability guarantees.9:08
PodcastDesigning ML Infra for ML & LLM Use CasesClaimEngineers scaling AI systems need to consider data scalability, model bias and outliers, model serving performance, and compliance and privacy.49:07
The State of Production Machine Learning in 2024ClaimProduction machine learning is challenging because it involves specialized hardware, complex data flows, compliance requirements, reproducibility, bias, outages, personal data, and cybersecurity.2:29
PodcastData Engineering in the Federal SectorClaimShane Morris says federal technology changes require security, authorization, and stakeholder approval rather than only a business case.17:57
Making Sense of LLMOpsClaimCustomer-facing LLM applications have risks including hallucinations, privacy concerns, bias, data-security breaches, dependence on external APIs, lack of human oversight, and misuse.11:11
Reliable Hallucination Detection in Large Language ModelsClaimHallucinations are an inherent limitation of large language models and are important to address when assessing reliability and trustworthiness.4:36
PodcastFedML Nexus AI: Your Generative AI Platform at ScaleClaimSalman Avestimehr says ownership, scalability, observability, privacy, and safety are major challenges when building generative AI applications.5:25
PodcastUber's Michelangelo: Strategic AI Overhaul and ImpactClaimThe generative AI gateway includes logging, auditing, cost guardrails, usage attribution, overspending alerts, safety and policy guardrails, and personally identifiable information redaction for data sent to external models.26:35
PodcastAll Data Scientists Should Learn Software Engineering PrinciplesClaimCatherine Nelson says security, package safety, model serialization, and adversarial attacks are software and machine learning concerns that data scientists should understand.13:56
PodcastExtending AI: From Industry to InnovationClaimModels that are no longer used can continue to consume compute and storage resources and create security exposure.14:44
PodcastAI in HealthcarePushed backDemetrios Brinkmann asked whether using multiple guardrail systems would be redundant overkill, while Eric Landry described using Ragas for evaluation and NeMo Guardrails for runtime protections and intent-based responses.41:36
A Blueprint for Scalable & Reliable Enterprise AI/ML SystemsPushed backNitin Aggarwal said there is no single established framework that serves as a universal or golden standard for evaluating and governing AI systems.32:52
PodcastHarnessing AI APIs for Safer, Accurate, & Reliable ApplicationsClaimRon Heichman says indirect prompt injection is especially dangerous when malicious instructions are retrieved from a web page or another source and enter the model's context as if the model found them itself.1:03:00
Vision and Strategies for Attracting & Driving AI Talents in High GrowthClaimAshley Antonides says an AI ethics or governance role works better when it is part of the technical team and participates in dialogue about requirements.4:42
PodcastMLSecOps is Fundamental to Robust AI Security Posture ManagementClaimSecurity is often one of the first casualties when AI and machine-learning organizations prioritize building quickly.3:17
Data Quality Management Techniques - The Complete GuideClaimLow-quality data creates costs through wasted resources, reconciliation work, lost revenue, lost opportunities, and damaged trust.2:00
The Evolution of Lyft's Feature StoreClaimFeature platforms should include strong data governance from the beginning, and deprecating features should be as easy as creating them.8:25
The Only Constant is (Data) ChangeClaimOrganizations have not kept governance and compliance practices up with the speed at which new data resources can be created.19:05
Unified Data + AI Governance with Unity CatalogClaimUsing separate catalogs such as Hive, an Iceberg REST catalog, and Glue creates fragmented discovery, governance, auditing, and lineage.8:41
PodcastThe EU AI Act: Navigating New LegislationClaimAI governance is becoming a practical organizational task as companies expand from a handful of AI applications to many more.3:04
PodcastAI-Driven Code: Navigating Due Diligence & Transparency in MLOpsPushed backMatt disputes the idea that every codebase should have zero security warnings, arguing that the appropriate amount of technical and security debt depends on the company’s size and stage.25:55
The Future of Healthcare: AI is HereClaimShaun says HeyRevia's agent has perception, prediction, planning, and control layers, allowing it to identify hold music, pause processing while waiting, plan future steps, and follow guardrails.11:33
How to Make AI Agents that ACTUALLY WORKClaimPatrick Marlo identifies meta-prompting, safety and guardrails, and evaluations as three recurring themes among high-quality production agents.6:20
Goal Oriented Retrieval AgentsClaimZoe Weil says Faber Labs runs its backend almost entirely in Rust, which has improved memory safety, concurrent processing, privacy, security, efficiency, and cost.10:12
