743 karma · joined July 6, 2022
But while I believe the general trends outlined here, it's important to remember that "the world" != "the West" != "the US", and even within a single country, the situation can be vastly different from one place to another, one socio-economic stratum to another, or even one family to another.
I also think kids escaping to digital places is not a very recent phenomenon, and has been happening for at least a generation. It's just that the platforms have changed. Back in my days, that place was Facebook, and we had entire digital lives there that our parents knew little about. And it's no coincidence that we started migrating elsewhere at the time when the adults started joining as well (though the general enshitification of the platform certainly didn't help either).
Of course Jobs blocked this, but it's insane that it was even proposed as a serious idea. I'm pretty sure this would have been a PR stain on Apple even in the pre-social media era.
Maybe TF has gotten better since but at the time it really felt like an internal tool that Google decided to just throw into the wild. By contrast PyTorch offered a more reasonable level of abstraction along with excellent API documentation and tutorials, so it's no wonder that machine learning engineers (who are generally more interested in the science of the model than the technical implementation) ended up favoring it.
[1] The worst part was that Google only hosted the docs for the latest version of TF, so if you were stuck on an older version (because, oh I don't know, you wanted a stable environment to serve models in production), well tough luck. That certainly didn't gain TF any favors.
Fine for some applications, but a massive regression from modern fiber infrastructure and definitely not suitable for everything (just think how slow the modern web is even with 15ms connections to datacenters). There's a reason why Starlink & co are trying to set up communication satellites closer to the ground.
> “In space, you get almost unlimited, low-cost renewable energy”
Low cost???????? Sending a solar array into space would probably rank among the most expensive forms of energy production.
> Starcloud’s space-based data centers can use the vacuum of deep space as an infinite heat sink.
Well, good luck getting the heat out first. I hope you planned for some big radiators to go along your 5GW solar array.
Take an OCR model with 99.9% character-wise accuracy. Sounds pretty good, right? Well if your use case is, say, digitalizing old printed novels, then yeah it's probably good enough.
But what if your documents are personal records with millions of names, to insert in some administrative database? Now 1 out of 1000 persons will have their name misspelled. Ooops.
What does that mean exactly? I assume the LLM was not left alone with its task for 3 months without human supervision.
But I also found that once you reach a certain level of expertise at $BIGCORP there are few opportunities to try to push the boundaries of your domain knowledge, lest you are dead-set on taking initiatives and swim against the current of a rigid organisation. That may just be specific to my job and employer though.
And while the part about navigating internal complexity is true as well, it's harder to highlight on a CV than "I have worked with technologies X, Y and Z". At best you can say "I have worked with X at a big corp, and also Y, which you probably don't know because it's an internal tool." You just have to hope that the name of your employer alone carries enough weight on its own.
The other thing about working for $BIGCORP is that it molds your skills to be hyper-specific to this company. It's less about learning to use cool technology stacks, than it is about learning the internal tools, procedures, and unspoken etiquette of the company. Skills that are vital to navigate the everyday complexity of $BIGCORP, but that you can't really export to any other job.
> It might interest you that newer versions of Wasmi (v0.45+) extended the resumable function call feature to make it possible to yield upon running out of fuel:
That is really interesting! I remember looking for something like that in the Wasmi docs at some point but it must have been before that feature was implemented. I would probably have chosen a different design for the WASM apps if I had it.