Vinayak Hegde, Microsoft AI advocate and former CTO, argues that non-deterministic model outputs are beneficial for creative applications despite risks in regulated domains.
Insight
AI Ethics Frameworks Over-Index on Western Norms Over Asian Contexts
“Mostly when we think about ethics, we're thinking from a Western point of view some things that might be acceptable, maybe an Indian context or maybe a Chinese context may not be very similar to what we look from a different aspect.”
Assertion Not checkable as stated
Vinayak Hegde: DeepSeek showed models could be trained at a fraction of expected costs
“Deep Seek when it came out it created like, I wouldn't say chaos, but created a sensation because how good it was and it was strained at like a fraction of the cost, which people thought wasn't possible”
Prediction Not checkable as stated
Hegde: Achieving AGI will take longer than 10 years
“I'm a little bit of a skeptic. I think it will take longer than that.”
Insight
Hegde: B2B buyers tolerate poor UX, but consumers demand great design
“UX, for example, you know, and design is like, you know, something that is is like right at the top. In B to B, you can sometimes get away by not doing such a great job. And there will be more tolerance as long as you solve the business problem pretty well, bu…”
Insight
Hegde: LLMs Introduce Probabilistic Behavior Unprecedented in Traditional Software
“What is so different about LLMs is, always, from a software engineering perspective, we never had elements which were probabilistic. We always had elements which were very deterministic. If you did A, B, and C, you would get D. If you do A, B, and C, again, yo…”
Prediction Not checkable as stated
Hegde: Software Engineering Will Transform Over Five Years for AI
“Software engineering has, discipline has to change to accommodate for that. And I think we'll be going through this transformation. So when we have this you know, conversation, maybe five years later, I think we'll talk about it very differently.”