Enjoyable chat with Nifemi Aluko. Seems we went to the identical highschool in Lagos and later boxed on the similar gymnasium in Austin. We additionally bought into AI measurement, calibrated belief, and why complicated techniques…

Episode 9 of Practice Ground simply launched. And this one’s private.

My visitor is Sanmi Koyejo, an assistant professor of pc science at Stanford, founding father of Virtue AI, and one of many main voices in reliable AI analysis. He’s additionally writing a e-book on machine studying from human preferences.

We attend the identical highschool (International School of Lagos – ISL) – I used to be in his brother’s class.

We did not even piece that collectively till years later, after we’re already coaching on the similar boxing gymnasium in Austin. The dirty, no-AC storage gymnasium with our coach, Bruce, yelling at everybody. I keep in mind watching Sanmi go on the pace bag and pondering, rattling, this man can field.

He’s now main analysis on synthetic intelligence at Stanford.

We talked about:
– What reliable AI really means technically (calibrated belief, not an excessive amount of, not too little)
– Why the measurement techniques constructed for AI’s inner use are failing on the societal stage
– Why 50% of significant AI work is knowledge and engineering
– Whether people will ever cease babysitting complicated AI techniques (spoiler: he is by no means seen it occur)
– What serendipity really appears to be like like in a analysis profession

I requested Sanmi what proportion of actual AI work is knowledge vs. algorithms. He stated:
“Despite what I love, the actual thing is maybe fifty percent data and engineering, ten percent the stuff that we actually teach.”

He loves algorithms. That’s the work that excites him.

But throughout the largest AI labs on this planet, the work that really produced significant outcomes got here all the way down to knowledge curation and engineering self-discipline.

There’s an unbelievable quantity of “humanness” behind each advance we rejoice as automation.

The narrative we inform is that machines did it. The reality is individuals did the curating, debugging, and deciding.

That’s a compelling story about human’s position within the hype of automation.

I hope you benefit from the dialog.

Watch on YouTube or hear on Spotify/Apple.
Link in feedback.