18-08-15 Most Valuable Research Events
Category: Idea Lists (Upon Request)
<!-- gdoc-inlined -->
Original thought:
ML contrarian lunch.
- What do ml researchers commonly believe that you think is false?
- Which fundamentals are broken, in a way that has downstream effects on the way that we think and do research?
- What is the most important research idea that is currently not being worked on?
- What current beliefs are the largest stoppers of progress?
- Truths that few people agree with you on
- Independently generate answers to one of the above. (to the 10m mark)
- Create large doc with collective answers
- Choose one to two worth defending / debating / discussing
- Place selected ideas into public pool
- Public vote (to the 20m mark)
- Debate / Discuss / Defend 3-4 ideas (to the 55m mark)
- Choose next event’s prompt, debrief (to the 60m mark)
Other thoughts / seeds with potential:
A lot of this stuff requires psychological safety. One way to maintain safety is by having really strict guardrails on what can and can’t be said, should and shouldn’t be said, and how things are said. Another, better way is to only conduct these conversations in high trust environments between long-term friends. People you can expect to give you charity, to vouch for you, or who you know you can get along with after a dispute has come and gone.
- What do people believe that you think is false?
- Outstanding problems in <field x>
- Why I work on what I work on
- My path to google brain (/ short life stories)
- Assumption Questioning
- This idea must die
- What is fundamentally broken, in a way that stoppers progress?
- Who are the people in the way of that progress
- What is the most important research that is currently not being done?
- How would your research (and your behavior) change if you didn’t have any requirements to publish, to make money and your position was guaranteed?
- We have lots of ‘talks’, not enough conversations
- Not enough debates
- What is the next generation Tensorflow / DL library?
- How do its assumptions differ?
- What is after DL? What is the DL paradigm in the way of?
- What could replace / bring down DL as the primary ml research paradigm? Should google invest in that, make it happen and own it?, stopper it?
Unrefined Ideation
Most valuable parts of conversations can generalize here. Take a valuable conversational mental motion, and create an event around it. x
Frame has to be tough love.
A lot of this stuff requires psychological safety. One way to maintain safety is by having really strict guardrails on what can and can’t be said, should and shouldn’t be said, and how things are said. Another, better way is to only conduct these conversations in high trust environments between long-term friends. People you can expect to give you charity, to vouch for you, or who you know you can get along with after a dispute has come and gone.
And so, I turn to exclusivity. Within brain. Which can be damning… I can see why this doesn’t exist. I may want to stay a lot much hush-hush about this kind of thing.
- What do people believe that you think is false?
- Truths that few people agree with you on
- Outstanding problems in <field x>
- Why I work on what I work on
- My path to google brain (/ short life stories)
- Assumption Questioning
- What is fundamentally broken, in a way that stoppers progress?
- This idea must die
- Who are the people in the way of that progress
- What is the most important research that is currently not being done?
- How would your research (and your behavior) change if you didn’t have any requirements to publish, to make money and your position was guaranteed?
- We have lots of ‘talks’, not enough conversations
- Not enough debates
- What is the next Tensorflow / DL library?
- How do its assumptions differ fundamentally?
- What are the deep outstanding questions in machine learning (not just DL)
- What is after DL? What is the DL paradigm in the way of?
- What could replace / bring down DL as the primary ml research paradigm? Should google invest in that, make it happen and own it?, stopper it?
Source: Original Google Doc