Hit 5 years at OpenAI this week. I've given my best mentally and physically, still I struggled with imposter syndrome—more so in the past year.
Tao Xu
301 posts
- I want to work with greatest folks I have ever worked with; I want to contribute to the 21st century Manhattan project like my heroes did 70 years ago; I wanted AGI to benefit all of us, not just a few billionaires.The most interesting thing about OpenAI is: How did the company build a culture that not only shipped GPT4, DALLE, and ChatGPT — but also made every staff willing to follow Sam out the door? Do any OpenAI staff want to share?
- Back in the days, @alexandr_wang helped recruit me to OpenAI. Now I still chose to stay at OpenAI after 6 years and all the craziness. Long time ago, I also worked at Meta and built a few things there, I am definitely happier here.
- I quit, officially unknown employment status now.
- I could testify that, the output from 2-3 person team here are often comparable or even superior to 10X or even 100X bigger teams at other places in tech.i know im not supposed to brag about openai, but the talent density at this scale (375 people) is 🤯 and i dont think has happened in the tech industry in recent memory
- Human brain is way under-trained under Chinchilla scaling law. Eye has highest bandwidth, roughly 50 frames/sec, each frame worths 1k tokens; within the entire human life time, it will collect less than 1e14 tokens; human brain has 1e14 weights, requires 2e15 tokens to train.
- Let's go!! OpenAI 2.0 will 🚀🚀🚀🚀.We have reached an agreement in principle for Sam Altman to return to OpenAI as CEO with a new initial board of Bret Taylor (Chair), Larry Summers, and Adam D'Angelo. We are collaborating to figure out the details. Thank you so much for your patience through this.
- OpenAI is nothing without its people
- I completely agree. Personally I think that my work at OpenAI is not just my job, is something much greater.
- “I say unto you: one must still have chaos in oneself to be able to give birth to a dancing star. I say unto you: you still have chaos in yourselves.". I think that he is talking about OpenAI.
- proud of my work here.Whisper V2 might be "just a little more training" but from our early tests, the improvement is huge! On our particularly difficult internal dataset, we're down from 47% word error rate to 37% (Google speech-to-text is at 51%)
- pretty proud of our work here, though I joined late into the efforts. The final model is after 13 iterations, I joined at v9. The model's English audio understanding capability beats me as nonnative speaker 2x-3x in terms of WER.We've trained a neural net called Whisper that approaches human-level robustness and accuracy on English speech recognition. It performs well even on diverse accents and technical language. Whisper is open source for all to use. openai.com/blog/whisper/
