Honest question, why do they need tens of thousands of GPUs? I would have thought they could run a Sol-level model on a few million dollars of hardware. Let's say it's ten million dollars, and let's also say they want to test multiple models in multiple different ways. You're still talking on the order of a hundred million dollars to build an airgapped test system. Happy to have my math proven wrong here.
One of my friends from grad school got a Masters of Public Policy then decided she would rather chalk menuboards for a living, spun up an LLC, and does amazing work. I admire people who are so willing to chase their dreams. I joke with my five year old that I still don't know what I want to be when I grow up, except it's increasingly not a joke.
I totally agree with you! I love seeing a flyer at my daughter's school that was obviously made by a student. It shows that they care enough about the thing to put real work into promoting it. Heck yes I'll buy a donut from coffee cart so you can go to some camp, fifth graders!
Now most people get their music through a paid service that streams content from a remote data center. OpenAI is trying to replicate the music industry's history.
It'll be easy to enforce for most corporate clients. I work in a profit-driven corporation that's now owned by a private equity firm. We try all sorts of things to make money. But there's no way we'd deploy a legally questionable model, even if we thought nobody on the outside would ever find out. I was our technical lead for GDPR implementation; we took all sorts of steps to make sure systems were compliant even when there was approximately zero chance of anybody on the outside ever finding out if we hadn't. The buy-cake-for-money-launderers companies flouting the law are an exception, not the rule.
I recently hired for a tech position on my team. I ended up hiring an amazing internal applicant. I consider myself lucky. The hiring process has gotten awful on both sides in recent years.
If you do crime, can you be arrested by the local police?
"I have diplomatic immunity! Also I live on a reservation with their own police force. So no."
It's pretty clear that's what the amendment means. The only ambiguous situation I've found is if you're born just inside American waters on your way between foreign countries, but that's a really narrow case.
What change would you suggest? A minimum period of time in the country before birthright citizenship applies? What about people who enter the United States illegally and then have a child twenty years later? What about my ancestors, who entered the country legally but never pursued American citizenship?
The 14th amendment was written the way it was to create a bright line that was easy to implement. I'm sure they considered other ways of framing the issue. I think it's brilliant just the way it is.
The United States is different from 95% of countries. We're the place you are supposed to flee to when those other countries oppress you. That's just not the history of a place like Italy or Japan.
My great great grandparents left Quebec in search of a place where they could earn enough money to make it. They immigrated to America. They lived in communities of other Quebec emigrants and spoke French their whole lives. They never pursued American citizenship. Without birthright citizenship, would my great grandma have been American? OK then what about my grandpa? What about me? I'm not sure if any of my immigrant ancestors formally pursued American citizenship.
This is real. If your app breaks because your AWS region crashed, the CEO probably isn't going to blame you. If you have similar uptime but the crashes are due to you picking a cheaper setup, you might be in more trouble.
I wonder if there's a single trainer in Kenya who is responsible for some of the conventions we see so often. Maybe (s)he just really likes full stops and used them in all of the training examples.
I'm not excusing Altman's many many lies, but it is worth noting how many algorithmic advances in the past 20 years were made with relatively little computing power. Think of things like xgboost (2016). A modest computer can run a pretty big dataset on CPU. When Tensorflow was launched in 2015, I played with it on an ancient laptop and it worked just fine. Then I upgraded to a mobile workstation and was still able to keep up with many SOTA models. Turns out LLMs are massively more power hungry than most earlier algorithms, even in NLP.
US law is more murky if they are responding to a perceived emergency. That can give probable cause. Imagine someone calls 911 due to smoke coming out of the neighbor's house. Clearly the drone can legally use IR to look for fire in the house. But it gets complicated fast, which is what worries me. Now imagine a car backfires, a microphone array reports it as a gunshot, and a drone shows up and starts scanning the nearest apartment building.
Nicely done! When I started moving up from being an individual contributor to corporate management years ago, the brass would casually say things like "P&L" or "EDITDA" and I had no idea what they meant. I read textbooks, took online classes, and it was a big deal when I finally could lead a conversation related to our finances.
Well said. There are so many better companies out there. It reminds me of Microstrategy before their Bitcoin madness started... why would anyone use it?