Strange. Had never read blog posts about individual engineers leaving Google on official Google Developers Blog before. Is this a first? Every day someone prominent leaves Google... Sounds like a big self-own if Google starts to post this kind of stuff. Looks like sole post by either of the (both new to Google) authors in the byline.
Francois Chollet is leaving Google
61–70 of 182 posts
Re: Francois Chollet is leaving Google
#62Earlier quoted context omitted.
$100M isn't enough capital for an AI startup that's training foundation models, sadly. A ton of folks of similar stature who raised that much burnt it within two years and took mediocre exits.
Interesting, I think $100M is totally enough to train a SotA "foundation model". It's all in the use case. I'd love to hear explicit arguments against this.
There's a bunch of failed AI companies who raised been $100M and $200M with the goal of training foundation models. What they discovered is that they were rapidly out paced by the large players, and didn't have any way to generate revenue.
You're right that it's enough to train one, but IMO you're wrong that it's enough to build a company around.
Re: Francois Chollet is leaving Google
#63Earlier quoted context omitted.
Google, in my experience, is a place where smart people go to retire. I have many brilliant friends who work there, but all of them have essentially stopped producing interesting work since the day they started. They all seem happy and comfortable, but not ambitious. I'm sure the pay is great, but it's not a place for smart people who are interested in doing something . I've followed Francois (and had the chance to c…
Am I almost alone in having no interest working for a large firm like Google? I've been in tech since the 90s. The only reason I'd go is to network and build a team to do a mass exodus with and that's literally it. I don't actually care about working on a product I have exactly zero executive control over.
Why zero executive control ? I’d expect a company like google ( like most large orgs ) to have a very large amount of internal code for internal clients, sometimes developer themselves. My experience of large orgs tells me you can have control over what you build - it depends on who you’re building it for ( external or internal)
Re: Francois Chollet is leaving Google
#64I loved Keras at the beginning of my PhD, 2017. But it was just the wrong abstraction: too easy to start with, too difficult to create custom things (e.g., custom loss function). I really tried to understand TensorFlow, I managed to make a for-loop in a week. Nested for-loop proved to be impossible. PyTorch was just perfect out of the box. I don't think I would have finished my PhD in time if it wasn't for PyTorch. I…
Keras was a miracle coming from writing stuff in Theano back in the day though.
And PyTorch was a miracle after coming from LuaTorch (or Torch7 iirc). We’ve made a lot of strides over the years.
Re: Francois Chollet is leaving Google
#65Earlier quoted context omitted.
Am I almost alone in having no interest working for a large firm like Google? I've been in tech since the 90s. The only reason I'd go is to network and build a team to do a mass exodus with and that's literally it. I don't actually care about working on a product I have exactly zero executive control over.
Why zero executive control ? I’d expect a company like google ( like most large orgs ) to have a very large amount of internal code for internal clients, sometimes developer themselves. My experience of large orgs tells me you can have control over what you build - it depends on who you’re building it for ( external or internal)
That's not what I mean. I've got a deep interest in how a product is used, fits in a market, designed, experienced AND built.
If I went to Google what I'd really want to do is gather up a bunch of people, rent out an away-from-Google office space and build say "search-next" - the response to the onslaught of entries currently successfully storming Google's castle.
Do this completely detached and unmoored from Google's existing product suite so that nobody can even tell it's a Google product. They've been responding shockingly poorly and it's time to make a discontinuous step.
And frankly I'd be more likely to waltz upon a winning lottery ticket than convincing Google execs this is necessary (and it absolutely is).
Re: Francois Chollet is leaving Google
#66Hi HN, Francois here. Happy to answer any questions!
Here's a start --
"Did you get poached by Anthropic/etc": No, I am starting a new company with a friend. We will announce more about it in due time!
"Who uses Keras in production": Off the top of my head the current list includes Midjourney, YouTube, Waymo, Google across many products (even Ads started moving to Keras recently!), Netflix, Spotify, Snap, GrubHub, Square/Block, X/Twitter, and many non-tech companies like United, JPM, Orange, Walmart, etc. In total Keras has ~2M developers and powers ML at many companies big and small. This isn't all TF -- many of our users have started running Keras on JAX or PyTorch.
"Why did you decide to merge Keras into TensorFlow in 2019": I didn't! The decision was made in 2018 by the TF leads -- I was a L5 IC at the time and that was an L8 decision. The TF team was huge at the time, 50+ people, while Keras was just me and the open-source community. In retrospect I think Keras would have been better off as an independent multi-backend framework -- but that would have required me quitting Google back then. Making Keras multi-backend again in 2023 has been one of my favorite projects to work on, both from the engineering & architecture side of things but also because the product is truly great (also, I love JAX)!
Re: Francois Chollet is leaving Google
#67I've always wondered how fchollet had authority to force keras into TF... https://github.com/tensorflow/community/pull/24
notably that link shows “@tensorflow tensorflow deleted a comment from fchollet on Nov 21, 2018” as well as other deleted comments
Re: Francois Chollet is leaving Google
#68Hi HN, Francois here. Happy to answer any questions! Here's a start -- "Did you get poached by Anthropic/etc": No, I am starting a new company with a friend. We will announce more about it in due time! "Who uses Keras in production": Off the top of my head the current list includes Midjourney, YouTube, Waymo, Google across many products (even Ads started moving to Keras recently!), Netflix, Spotify, Snap, GrubHub, Sq…
Congratulations Francois! Thanks for maintaining Keras for such a long time and overcoming the corporate politics to get it where it is now.
I've been using it since early 2016 and it has been present all my career. It is something I use as the definitive example of how to do things right in the Python ecosystem.
Obviously, all the best wishes for you and your friend in the new venture!!
Re: Francois Chollet is leaving Google
#69I loved Keras at the beginning of my PhD, 2017. But it was just the wrong abstraction: too easy to start with, too difficult to create custom things (e.g., custom loss function). I really tried to understand TensorFlow, I managed to make a for-loop in a week. Nested for-loop proved to be impossible. PyTorch was just perfect out of the box. I don't think I would have finished my PhD in time if it wasn't for PyTorch. I…
Keras 1.0 in 2016-2017 was much less flexible than Keras 3 is now! Keras is designed around the principle of "progressive disclosure of complexity": there are easy high-level workflows you can get started with, but you're always able to open up any component of the workflow and customize it with your own code.
For instance: you have the built-in `fit()` to train a model. But you can customize the training logic (while retaining access to all `fit()` features, like callbacks, step fusion, async logging and async prefetching, distribution) by writing your own `compute_loss()` method. And further, you can customize gradient handling by writing a custom `train_step()` method (this is low-level enough that you have to do it with backend APIs like `tf.GradientTape` or torch `backward()`). E.g. https://keras.io/guides/custom_train_step_in_torch/
Then, if you need even more control, you can just write your own training loop from scratch, etc. E.g. https://keras.io/guides/writing_a_custom_training_loop_in_ja...
Re: Francois Chollet is leaving Google
#70Hi HN, Francois here. Happy to answer any questions! Here's a start -- "Did you get poached by Anthropic/etc": No, I am starting a new company with a friend. We will announce more about it in due time! "Who uses Keras in production": Off the top of my head the current list includes Midjourney, YouTube, Waymo, Google across many products (even Ads started moving to Keras recently!), Netflix, Spotify, Snap, GrubHub, Sq…
Will you come back to Europe?