Personal news: I've decided to take on a new challenge at
@scale_AI
as their Director of Product! I had a blast building Tesla Autopilot's data engine and I'm excited to generalize this product for multiple industries from computer vision to large language models.
My team is hiring Project Managers excited to run Tesla's Data Engine and curate real, large, diverse, high quality datasets to improve Autopilot. The role bridges our vision engineering team and in house data annotation team - apply here
I know many of my followers are Tesla fans (I remain a big one too!). Part of the reason I knew I could leave was because the team is in great hands and on track to magic. Looking forward to all the updates on my Model <3 ๐
Personal news: I've decided to take on a new challenge at
@scale_AI
as their Director of Product! I had a blast building Tesla Autopilot's data engine and I'm excited to generalize this product for multiple industries from computer vision to large language models.
Back when I was taking Stanford's CS224S Spoken NLP course, RNNs and CNNs were the talk of the town. Now I can beat the perf of our final project (RNN trained on acoustic features and language features) by finetuning a transformer, even without audio :)
this role is in person in Palo Alto, CA. We look for sharpness and grit and concise comms skills and the eagerness to work with a variety of teammates :) ML experience is a plus
The video of the panel I moderated last week on self-driving ๐๐ค is now up!
Thanks again to
@AnguelovDrago
,
@alexgkendall
and
@drmapavone
for an excellent conversation, the topics we covered:
1/4 Got a dopamine hit from ch.2 of
@huggingface
NLP with transformers: finetuned transformer on dataset of 16k tweets labelled with sentiments. For my sentiment analyzer, even the single word "Tesla" has a most probable class of "joy" :)
I recently celebrated my 4 yr Teslaversary ๐ and re-listened to this episode where I reflect on creating my role, my first exec meeting and the meritocracy of the team: Statement Mondays: Interviews With Women Who Wear โHeelsโ to Work
The more pain I endure, the more I learn. I resisted the urge to look at an answer key for chapter 3 of nlp with transformers since none existed. Let my solution not tempt you :)
Back from International Computer Vision Summer School
@icvss
which opened with โhow to eat pasta in Sicilyโ and featured lectures on generative models for synthetic data and building chatGPT from scratch. Plus poster session with wine! ๐คฉ
#icvss23
I'm hiring technical product managers! If you are a technical go-getter wanting to rally teams and build solutions for large language model and generative model datasets, please reach out to me or apply here:
Also, we'd love to see you at Scale's flash session, "Creating a Data Engine," led by our Director of Product Management,
@goldenkatepark
, today at 12pm PT at Booth
#1718
. See you there!
1/3 I learned infants are "universal listeners" who can discriminate differences in sounds from any language but at 6-12 months they begin to specialize in sounds of their native languages. Thus a Korean-raised child will start to struggle to distinguish between "L" and"R"
1:30: reflections on last decade of deep learning since the ImageNet moment
9:30: modular vs. end-to-end approaches
18:00: how much do large labelled datasets matter now with advent of large language models and emphasis on self-supervised learning and efficient fine-tuning?
If you are a hardcore C++/C engineer, wanting to write and optimize software for real-time systems, please reach out to me or apply here:
SW Engineer -
AI Compiler Engineer -
AI Integration SW Engineer -
24:00: what is the role of open source in autonomous vehicle development?
32:00: role of simulation
43:30: how would you design the perfect hiding spot from a self driving car? :)
Amazing start to the
@Redpoint
@agihouse_org
Imagine Hackathon with surprise guest Google Founder Sergey Brin chatting RLHF, quantum computing, and more with
@karpathy
2/4 To my surprise, there were actually no mentions of "Tesla" in the emotion-tweet dataset so the feature representation of "Tesla" in the base model (distilbert-base-uncased) must be sensible
2/3 Is there a parallel of this "phonetic perceptual narrowing" to NLP's tokenization? From a larger stream of input (sounds for phonetic narrowing and characters of text for tokenization), both draw boundaries to shortcut perception of the most frequent occurrences
3/3 I used chatGPT to find the name of this phenomenon :) While Google search masters title -> description, I appreciate Chat's performance the other way, description -> title
Our team is truly wonderful and stacked! Prior experiences at Tesla / Apple / Stripe / Microsoft and degrees in computer science or math from Stanford / MIT / Princeton... plus we know how to have a good time :)
We are thrilled to announce the incredible momentum behind our new AP CSA curriculum, which has support from 6 state governments, 17 corporations, and 14 higher education institutions. Thank you for joining us on this journey! ๐ช๐ปโค๏ธ
My favorite moment of recruiting is the introโs of the first call with a candidate. During this moment I make a judgement call (intuition) and throughout the interview process I calibrate that intuitionโฆ
The role is in person in SF
@scale_AI
HQ. Looking for intellect, hustle, concise comms skills, willingness to learn and natural ability to work with a variety of stakeholders
this episode from
@heykojo
and a recent conversation with another incredible woman holding the same position have made me appreciate the impact of a โHead of Peopleโโฆ
What a way to kick-start the week! Check out
@nataliemuenster
's new podcast Statement Mondays, a podcast about badass women by badass women.
Head to / spotify / apple podcasts to listen to the first three episodes.
#statementmondays
@nikunj
engineer turned PM is valid especially for the AI Product Manager role whereas I'm looking for marketplace or consumer growth for the Growth PM role