{"url": "https://www.youtube.com/watch?v=GGyq6-h8_VI", "page_url": "https://www.youtube.com/watch?v=GGyq6-h8_VI", "title": "Lightning Talk: Kubes 4 Nubes - Tom Brown, OpenAI", "show_or_channel": "CNCF [Cloud Native Computing Foundation]", "date": "2017-03-28", "duration_s": 297, "type": "talk", "speakers": ["Tom Brown"], "confidence": "high", "notes": "KubeCon + CloudNativeCon Europe 2017 (Berlin) 5-min lightning talk; sched bio confirms Grouper/MoPub Tom Brown.", "people": ["tom-brown"], "id": "2017-03-28_cncf-cloud-native-comput-lightning-talk-kubes-4-nubes-tom-brow", "audio_duration_s": 296.0544375, "worker": "jc-w05", "transcribe_s": 3.5, "model": "large-v3", "segments": [{"start": 0.0, "end": 46.08, "text": "Please welcome Tom Brown of OpenAI. Let me get this a little higher. Hey, so I am Tom Brown from OpenAI, and today we're going to learn in five minutes koobs for noobs. So first a quick word on noobs. Noobs are just people who are learning a product for the first time. This talk is not directed at noobs."}, {"start": 46.08, "end": 71.3, "text": "So if you're hoping to learn about kubectl get pods, you will not learn about that today. Instead, I'm going to talk about some product building stuff and how you can make tools that noobs actually love and continue to use until they no longer are noobs. So first, a quick word about OpenAI. OpenAI is a nonprofit research company that is trying to build safe AI."}, {"start": 71.3, "end": 96.94, "text": "And so we're about like 80% researchers. Everyone at the company is technical. It's about 60 people-ish now. And so most of the work that we do is running experiments. So basically each thing that we deploy is this beautiful and unique snowflake. So everybody has to learn how to use Kube. And then we have a couple other things like"}, {"start": 96.94, "end": 114.45, "text": "Since we're very bursty, we have an EC2 autoscaler and a bunch of other goodies that Vicky and Jonas, my colleagues, are going to talk about in their keynote tomorrow at 10. So my first couple weeks at OpenAI..."}, {"start": 114.45, "end": 138.56, "text": "Vicky and Jonas had basically built out our kube cluster and I had never used it before so my experience was kind of like reading through the docs or like when I got stuck seeing whatever magic command it was that they used to unblock themselves and I think originally when I submitted this I was going to talk about all the ways that the kube docs were just failing their users but"}, {"start": 138.56, "end": 166.16, "text": "Over the last two months, they've actually become very good. So good work, Kube people. I'm going to cut that entire section from this talk. So what I was doing then during that time was noticing places where I got stuck, and I would just write it in a doc. And so the first thing, if you want to make a product that's great for users and for noobs, is to write docs for yourself. I called this doc Kube for Noobs."}, {"start": 166.16, "end": 193.52, "text": "And basically then whenever someone else went through to, they would add stuff to that doc. And what's nice about that is that it's literally no effort. You are already figuring out the stuff, you just like dump it in a doc, and what you end up with then is actually good technical documentation for exactly the tasks that people within your org care about. And then the other type of way"}, {"start": 193.52, "end": 222.45, "text": "Yeah, the other type of docs that you will build is many of you are building stuff on top of Kube or building other products. You'll want to write docs for users, and this is a little bit more difficult because you have the curse of knowledge. So the tried and true trick for this is basically user interviews. So you write docs as best as you can, and then you go and physically sit with a user and see where it is that they get stuck."}, {"start": 222.45, "end": 251.71, "text": "And often it will be in surprising ways. So the classic things that you'll want to go and ask them are, what are they trying to do? Where is it that they're confused? What are the questions that they have that aren't immediately answered? And you can also look at where they go and start trying to look for documentation as a hint that your product isn't usable yet. So basically, write your docs."}, {"start": 251.71, "end": 279.71, "text": "Watch a user, and then you just repeat that. And what we try to do at OpenAI is to make sure that any product that we build isn't finished until any new user can come and figure out how to do the key workflows for it with just 10 minutes of reading the docs. And basically the practice for that is just watch a noob do it, then watch another noob. And if you follow these simple things, you'll end up building"}, {"start": 279.71, "end": 292.75, "text": "Thank you. I'm Tom Brown. If you want to learn more about building different products for particular research, feel free to email me, Tom, at OpenAI. Thanks."}]}