titaniczero

Well, my guess is that updates and inserts are much more frequent than searches in their use case. You're assuming a balanced frequency for these operations and it hardly ever happens.

ipv4 space is very limited and you can easily compute all the hashes. There is salting and combined with rotating salts it could work but no one guarantees that you’re not storing them

But the idea that the brain functions on a basis of complex statistical processes doesn't imply that statistics created the brain itself, it just suggests that the brain's processes can be modeled/understood through the lens of statistical methods. Statistics is just how we describe a tool we invented to analyze observable data, the brain could be a similar tool and it doesn't need to be the same tool.

It's akin to saying that we created the concept of "physics", yes we created it, and physics govern, for example, how a car moves, but it doesn't mean our concept of physics created the car or physics itself, we just use physics to describe and understand the car's movement.

Maybe statistics can't generate anything, but if you imagine everything we can do as very complex, unimaginable multi-dimensional functions that can generate outputs based on inputs, we can use statistics to find functions that fit any real (ground truth) function in the observable universe.

When you fine-tune it, do you train just the head/last few layers or do you also unfreeze the model afterwards and retrain the whole model with a very small LR for a few epochs?

10% of the population in spain was using netflix before this, as a reference 12% in germany, not that much difference. We'll see, but I think it has more to do with the decline in quality of the content and people sending a message than anything else.

My experience is similar but tbh I feel it's mainly due to the fact that oauth is implementing an important security layer (authentication/authorization) and that's hard by definition. There are lots of steps and things you can't afford to overlook or you'll be vulnerable to multiple different attacks

It’s probably provided as system instructions for rejecting things. You can use the API and feed it with different instructions with the system role

Because I was in a hurry and I don't have unlimited time, while providing info about the past that I already have was still valuable for the point

Why? Genuinely asking.

As a software engineer this all is very inspiring. I’m tickled and excited to do things again. I’m also learning deep learning and statistics and having a blast with toy projects.

I’ve noticed that bing chat is better if you instruct it not to search anything, that way it will use the model knowledge. I’ve learned to use the model knowledge or the web search results summary depending on what I want. But ChatGPT is still way better for model knowledge because it has fewer restrictions.

I wish they would make this distinction clearer in the UI. Most of the time it can answer without resorting to search, I think it would be better if the user explicitly specifies that they need web results.

OpenAI should implement an oauth authorization server and allow developers to use "Login with OpenAI account" into their apps.

I would be more concerned if we don't get to the point of decent local models for every task, because of the power these companies would have

> Actually nowadays I bet higher percentage of Swedes know English than know Swedish.

Is that hyperbole? Because that would be very sad if true, culturally speaking. There are 5 different official languages in my country (Spain) and there are several laws to prevent them from becoming extinct for this very reason.