roger (@rogertoor) header
roger (@rogertoor) profile picture

0 → 1 obsessive • 3x founder

San Francisco
Joined December 2008
Photo by roger (@rogertoor)Photo by roger (@rogertoor)Photo by roger (@rogertoor)Photo by roger (@rogertoor)Photo by roger (@rogertoor)Photo by roger (@rogertoor)
“At its core, all business is about making bets on human behavior.” — Ben Wiseman, Wall Street Journal via @mulegirl
10
roger retweeted
interpretability might be the most important scientific problem in the world right now, as I don't think we can solve alignment if we don't know what we're building. I think I have good news about it. I co-authored early work in mechanistic interpretability at OpenAI and came away pessimistic, because neural networks appeared to fold into arbitrary tangled knots. years later, I started openai's first large-scale automated interpretability project, finding about 2% of neurons legible with the primitive tools we had with modern agentic tools and a much closer look, the picture looks totally different. the vast majority in models up to trillions of parameters are legible, and their circuits beautiful. many famous examples of tangled neurons turned out to be misunderstood. The roadblocks I expected mostly aren't there. I now think ambitious interpretability will be solved by the time we need it most, and possibly quite soon, as automated scientists keep improving. it may be a bitter lesson for interpretability: the simple thing, at absurd scale, works already-clean.goodfire.com/
889
2,657
1,464
19,994
4,819,796
It’s tempting to understand consciousness as a fully flexible but resource limited attention head that performs local operations on mental contents, minimizing its prediction error by maximizing the improvement of expected reward, and that can take parts of itself into its focus.
30
16
2
216
Predictive Early Stopping – A Meta Learning Approach – @Cometml #MachineLearning buff.ly/2Vs5O9l
15
13
It is mindblowing that roughly the same protocol design for the Internet Protocol (IP) has taken the internet from nothing to ~15B connected devices and has survived for 40 years through the period of fastest technological change in history. What lessons can we draw? A thread 👇
68
583
125
2,377
Google's Hinton outlines new AI advance that requires less data reut.rs/2A1PZJb
1
“The Secret to Mastering Your Time is to Systematically Focus on Importance And Suppress Urgency” by @Alltopstartups medium.com/the-mission/the-s…
1
"Time Well Spent is a movement to align technology with our humanity." cc @tristanharris timewellspent.io/take-contro… youtube.com/watch?list=PLtLe…
Switch your home power to clean energy for free producthunt.com/tech/arcadia… via @owen_yin
“‘The typical path of how people respond to life-changing inventions:” by @cdixon medium.com/@cdixon/the-typic…
1
“🎈🎉 The Complete List of User Experience (UX) Resources & Tools 🎉🎈” by @jhreha medium.com/@jhreha/the-compl…
1
ouch, wtf happened in 1984? "When Women Stopped Coding" @NPR @planetmoney (tldr: home PCs were marketed to boys)
16
239
93
Thanks to @bethanye, @rogertoor, & Chakri Uddaraju for bringing great insights to our panel last night. It was a great event. #findyourCTO
Remember, software is a medium to express your value proposition. Iterate pre-tech to race ahead (where possible), then scale! #leanstartup
3
After thinking about homelessness for 10+ yrs, but not having the bandwidth to help, I'm delighted to join @NewStoryCharity as an advisor
2
8
16
Love @lyft even more now
2
Gokul Rajaram of Square Drops Product Management Knowledge blog.wizeline.com/2014/09/30…
7
This morning stopped by HP Labs where I worked 15 years ago. Travel to new places broadens you; travel to old places deepens you.