Live data from Hacker News

Problems that attract smart people

medium.com

1–10 of 106 posts

Re: Problems that attract smart people

#2
There are a couple assumptions implicit in our system of "problems" and "smart people".

There is the assumption that it will be written about.

There is the assumption that "solving" is the appropriate response.

Combined, they describe a constrained perspective. A volume which, compared to the alternative, is infinitely small.

So consider that the next time you feel a riddle's tug, smarty.

Re: Problems that attract smart people

#3
Was that implementation confirmed to work or was it a bug?

I saw people saying it only loads the weights necessary, but all the weights are necessary (unless the architecture is heavily modified). Does it stream or load only the currently necessary layers in a way?

Re: Problems that attract smart people

#4
post #3

Was that implementation confirmed to work or was it a bug? I saw people saying it only loads the weights necessary, but all the weights are necessary (unless the architecture is heavily modified). Does it stream or load only the currently necessary layers in a way?

On start up it does load a fraction of the weights, so RAM usage is low, but as you run inferences on the model, it tends to eat up the required amount of RAM. This was from my initial testing.

Re: Problems that attract smart people

#5
post #3

Was that implementation confirmed to work or was it a bug? I saw people saying it only loads the weights necessary, but all the weights are necessary (unless the architecture is heavily modified). Does it stream or load only the currently necessary layers in a way?

It mmaps the data so the page any particular bit of data is on is only loaded into memory if something on that page is accessed (at least that is my understanding).

Re: Problems that attract smart people

#6
Does enabling to run LLMs at home really strengthen democracy? People may believe that, but it potentially enables AI to run on any device with a low footprint for surveillance purposes as well, and governments don't even have to pay for the development...

Re: Problems that attract smart people

#8
post #6

Does enabling to run LLMs at home really strengthen democracy? People may believe that, but it potentially enables AI to run on any device with a low footprint for surveillance purposes as well, and governments don't even have to pay for the development...

> Does enabling to run LLMs at home really strengthen democracy?

Well, I can't speak about strengthening democracy for sure, but what I see is that AI will become a significant means of digital production, and we wouldn't want those means to be in the hands of a few people.

> but it potentially enables AI to run on any device with a low footprint for surveillance purposes as well,

Yes, a low barrier for inference cuts both ways. That is a risk.

Re: Problems that attract smart people

#9
post #6

Does enabling to run LLMs at home really strengthen democracy? People may believe that, but it potentially enables AI to run on any device with a low footprint for surveillance purposes as well, and governments don't even have to pay for the development...

If you're referring to this passage

> democratising the means of digital production is the difference between a dystopian authoritarian future and an empowered (and admittedly chaotic) humanity

The usage here means make accessible to everyone.

Re: Problems that attract smart people

#10
post #6

Does enabling to run LLMs at home really strengthen democracy? People may believe that, but it potentially enables AI to run on any device with a low footprint for surveillance purposes as well, and governments don't even have to pay for the development...

A low barrier to entry for surveillance means citizens can more easily watch their governments too.

As long as surveillance tech requires big resources, governments will have the edge. Miniaturize it, decentralize it, and it means citizens and public interest organizations can hold the government accountable more easily (this is, notably, what OSINT is all about).

Post reply on HN