Yes its a problem for the grid and the power companies don't allow large clusters to oscillate their power like this. The solution that AI have to do during their training big runs is to fill in the idle time on the GPUs with dummy operations to keep the power load constant. Having capacitors would be able to save on power usage.
Building their own models gives them more leverage for negotiating favorable terms with openai. The key to a good negotiation is having a strong BATNA, best alternative to a negotiated agreement.
Hmm I saw a different analysis recently indicating that there was a large reduction in fast food jobs in California after the raise to $20/hr. Which one is right?
Something I'm curious to know: How does the efficiency of this new process compare to using regular solar panels to generate electricity and then using that electrical energy to synthesize the same chemicals?
That was a voluntary pledge the company made to the users, right? It wasn't a legally binding commitment that there would never ever be any data sharing.
What I heard was that it was that the poor fellow tried to shoot himself but didn't get a direct hit. Because there was some time passed between the shooting and the death, they didn't find the gun and the body in the same place that you would expect.
You can do that but the amount of incremental data will be negligible compared to the rest of the data. Think of the knowledge cutoff more like a soft value.
10^17g is about 20,000 times heavier than the great pyramid of giza. Funny to imagine something as massive as a pyramid pass through your body and not noticing it.
I'd like their to be a single payer for my healthcare but I'd like that single payer to be me and not the government and not an insurance company. Just like I am the single payer for all my food and my housing why can't we have a world where I pay for my healthcare too?
By the time these academic studies get published they are usually already several months out of date. o3-mini was released yesterday and if one wants to know about the limitations of current technology they are much better to check twitter than some research paper
The idea I heard was to make liquid nitrogen during the day when solar power is abundant and then run the chips at greater efficiency at night using your stored liquid nitrogen.
If you use chatgpt somehow saves you from making one trip to the doctor in your car it can offset the entire year worth of chatgpt usage in terms of co2 impact.
Visiting Gettysburg was one of the few times in my life where I’ve done a tour with a private guide. It was way better than I expected. The guide actually drove me and my family in my car and gave us a great experience.
Yes it is. In RLHF and DPO you are optimizing the model output for human preferences. In the reinforcement fine tuning that was announced today you are optimizing the hidden chain of thought to arrive at a correct answer, as judged by a predefined grader.
It's illegal to destroy evidence of a crime but it's not illegal to avoid creating evidence in the first place especially if you genuinely believe that you're not doing anything wrong. Generally speaking, companies are not obligated to preserve every chat forever just in case they get sued later on.
This one makes sense to me cause the price (interest) is related to the risk or repayment. The way the universities do it is they want to find out how much money is in your parents bank account and then take as much of it as possible.
Currently, talented engineers flock to google to contribute their skills to making the best web browser. My concern for a publicly owned utility is that the top talent won't want to work there.
What’s the difference between a ghost job and opportunistic hiring? I’ve never seen any team I work for put out a job posting with no intention of hiring and I’ve certainly never participated in interviewing a candidate where there is no intention to hire. However, I have seen my team put out a job posting where we only intend to hire if we get an applicant that is unexpectedly good.
> You can't actually prove the assertion that being in office makes the difference here at all
That was the point that they were trying to make. You can't prove such a thing with data one way or another, i.e. it's not possible to a/b test company culture.