Ian Goodfellow, of the GAN fame, left Apple and rejoined Google DeepMind because of Apple’s mandatory RTO policy.
369 karma · joined September 22, 2020
[1] - https://blog.google/inside-google/company-announcements/buil...
[2] - https://www.cnbc.com/2024/04/17/google-workers-arrested-afte...
[1] - https://github.com/meta-llama/llama3/blob/main/MODEL_CARD.md...
Thank you for doing this!
He left recently, but he wrote this blog while still being employed there, as mentioned at the bottom of the post.
SLURM does that wrapping for you, where you essentially just point to the file that you want to run, along with some high level GPU and CPU resource allocation tags, and it just schedules and runs it for you.
I have seen some people trying to run GCP (lol) with SLURM, and wouldn't be surprised if it is possible with AWS/Lambda or any of the other cluster service providers (Cluster-as-a-service, CLaaS?).
Just through one Google search, looks like its definitely possible with AWS: https://docs.aws.amazon.com/parallelcluster/latest/ug/slurm-...
How successful have precious iterations of the XPrize been? IIRC, there was (is?) one for landing a rover on Mars and another for cleaning up the ocean.
Have those companies yielded tangible benefits or even spurred innovations in their respective fields?
Update from Greg
Btw, I was at R:SS 2022 and meeting the Skydio autonomy team was one of the highlights of my career as a robotics engineer!
The core of these advancements are powered by Diffusion Policy [1], which Prof. Shuran Song's lab at Columbia (before she moved recently to Stanford) developed and pioneered. I'd suggest everyone to view the original project website [2], it has a ton of amazing real world challenging experiments.
It was a community favorite for the Best Paper Award at the R:SS conference [3], this year. I remember our lab (and all other learning labs in our robotics department), absolutely dissecting this paper. I know of people who've entirely pivoted away from their projects involving behavior cloning/imitation learning, to this approach, which deals with multi-modal action spaces much more naturally than the aforementioned approaches.
Prof. Song is an absolute rockstar in robotics right now, with several wonderful approaches that scale elegantly to the real world, including IRP [4] (which won Best Paper at R:SS 2022), FlingBot [5], Scaling Up Distilling Down [6] and much more. I recommend checking out her lab website too.
[1] - https://arxiv.org/abs/2303.04137
[2] - https://diffusion-policy.cs.columbia.edu/
[3] - https://roboticsconference.org/program/awards/
[4] - https://irp.cs.columbia.edu/
They do some fantastic research in applied and theoretical ML. Rosanne Liu is a personal hero/role model of mine, and we all have a lot from the way she has organized this group and her career in general.
I really wish I'd known about this group back when I was an undergrad and just getting into DL research.
Let me know what your interests are, I can point you in the direction of some interesting people who might consider mentoring young undergrads.
All the best!
Back in 2017, I was just starting out as an intern at a Systems Design lab in a university. I was about to head into my first meeting with a potential advisor, when a college senior from my robotics club forwarded the link to DonkeyCar on the club group chat. IIRC, the first repo was hosted on wroscoe's personal account, and allenwells and AdamConway were the most active contributors, along with wroscoe of course.
I did not have any experience with Linux, Git, ML. Nothing! I walked into the meeting with the advisor and pitched working on DonkeyCar. It was new, it was shiny, it used TensorFlow v1! It was down to working on this, or some GIS thing, analyzing satellite spectral bands.
Anyways, my accidental blurring out of this project in the meeting led to the most fruitful summer of my life, where I managed to learn about package managers, machine earning, computer vision and how to ask the right questions on an open source communities Slack. I even remember badgering the project Slack so many times that wroscoe replies to me, saying that I should be more mindful about people's time, and that I should read the Python Error messages more carefully, before just copy pasting lines and lines of error code.
In addition to introducing me to Linux, this project introduced me to Deep Learning as well. I remember going through Josh Gordan's tutorials on KNN's, during my attempts to debug the Differential Drive DonkeyCar code. Ah the memories! One thing led to another, and I was following the launch of the first TPUs. I remember heading for lunch with a few folks from the department, where they began discussing the dedicated hardware that would be accelerating TF code. It was wonderful, as I tried to keep up with the announcements.
This brings back so many memories. If Will, Adam or Allan are reading this: Thank you. Your brilliant project absolutely changed my life and sparked a lifelong fascinating with robots, ML and tinkering!
A commenter below posted:
>Lehman Brothers is also not included because, even though it was a US bank, it was an investment bank with no FDIC insured deposits. It was around the size of all of this year's failures, combined.
So does this mean, that since Lehman had no customer deposits, it doesn't count?
Thank you again!
By commercializing this research, I mean why not integrate this into GMail for auto reply solutions, so that it can automatically suggest meetings? Why not integrate it into Slides to come up with better titles, summarization etc? Similarly for Google Docs, automatically summarize reports, make suggestions depending on crispness/clarity etc.
Why did they decide to just sit on these models, and endlessly prolong their product launches that had these features baked in.
I'd love any alternative view points of this.
It also showed me how SpaceX was born when Elon and his team crunched some numbers on a trip back from Russia, only to realize that it would be cheaper to build their own rockets rather than rent Russia's.
That book is filled with stories about complex systems that humans broke down into simpler problems and hacked them together into monolith companies. Contrary to popular beliefs (mostly held by people who haven't read the book), its not entirely about Elon.
My two cents.
I installed Ubuntu 16.04 initially and half of the drivers did not load. Every time I installed a new driver another cog stopped working. It was a mess to get Bluetooth working, WiFi just wouldn't work and my display driver straight up wouldn't work. After a week of getting stuck in this tangled mess, I decided to move to 18.04 assuming that drivers would be available on a later version of an LTS distribution. I was partialy right, I managed to get the WiFi and Bluetooth working but the god damn display drivers just won't load. This means I cannot suspend my PC, adjust brightness or connect to an external monitor. I will sell you this machine but looks like you need linux as well.
So I have a decently speced hexa-core PC with a dream keyboard, great screen but cannot run Ubuntu which is what my job as a robotics engineer most need. Everytime I look at it I'm infuriated at my utter lack of due diligence. There's a guide on the Ubuntu website with a list of supported PCs. Refer to that before making a purchase. Its hellish, frustrating and heartbreaking otherwise.
/rant