Hard Truth About Using and Building AI Systems

Why It Matters

Because AI hallucinations can drive faulty business decisions, firms must secure reliable data pipelines before deployment.

Key Takeaways

• AI reflects average internet content, not superior intelligence.
• Without curated data, AI will frequently hallucinate answers.
• Enterprise AI deployments often ignore essential data‑management practices.
• Relying on free‑form models risks misleading business decisions.
• Invest in data pipelines now to avoid future AI failures.

Summary

The video bluntly labels current generative AI as “mid,” arguing it merely aggregates the average of publicly available content—from Reddit to Google—making it a mediocre reflection of human knowledge.

The speaker warns that when firms plug such models into enterprise workflows without feeding them high‑quality, domain‑specific data, the systems will inevitably hallucinate, producing fabricated answers that can mislead decision‑makers.

He cites examples like the model being trained on obscure sites such as unpublishedbooks.com and recalls a conversation from two decades ago about the necessity of proper data collection that many companies still ignore.

The takeaway is clear: businesses must prioritize robust data infrastructure and governance before relying on AI, or they risk costly errors and erode trust in automated solutions.

Original Description

At SXSW, Microsoft’s Dona Sarkar dropped some tough truths about artificial intelligence and what happens when the data it is trained on is not relevant or high quality.