Hacker News Reader

ttt3tson EleutherAI announces it has become a non-profit

The problem with LLMs is the big ones don't run on consumer hardware. If 12GB or 24GB ram would run them you would see just as much interest.

The open source models will keep up as hardware allows IMO.

ttt3tson Beating OpenAI CLIP with 100x less data and compute

Any plans to document how to fine tune your models then?

ttt3tson Google Has ‘Delusions of Exceptionalism’

Yea, I figured Google was the average mega corp when they removed "don't be evil".

Still, they have areas where they need regulation. E.g. Paying Mozzila and Apple to be default search is anticompetitive.

ttt3tson Big Tech's future is up to a Supreme Court that doesn't understand it

Alright, given the volume of messages on FB, twitter, etc, isn't choosing what messages you promote approximately equivalent to publishing anything? I am pretty sure you can find any opinion/view on these platforms.

ttt3tson Oakland declares state of emergency due to ransomware attack

For regular backups of 6TB why would you even consider tape? Seriously asking because I have priced tapes and the break even point vs hard drives is high.

ttt3tson Tesla goes down across Europe leaving some drivers unable to charge cars

Just wondering but what is the lifespan of an EV? How much does it cost to replace the batteries after 10 years? I ask because I fear only the rich will be able to afford cars with such a ban. My first car cost $2k and was a clunker but it was all I could afford to get to school.

Even now, in the US, Tesla's get perks like using the car pool lane. Great deal for the rich.

ttt3tson Ask HN: How do you find stuff on the internet now?

Hardly, reddit, in its current form, will die because of LLMs. It is too easy to get an account and post. They will have to wall the garden or trash will fill it. :(

ttt3tson Ask HN: Is anyone building a question answering system using the HN corpus?

Hmm. I have full size bloom running on a server in my basement. It can be ran naively with about 400GB of ram. Using used hardware you can get that for about $1200. Still, with CPU inference, you're looking at about 5 mins per response.

With optimization, I have it down to 140GB of ram. Trying to get it under 120GB without loosing too much accuracy so it can be ran on standard desktop consumer hardware (who's limits are usually 128GB).

Given the lack of resources I have found I figured the general intrest was low? Maybe I will open source it and do a write up.

ttt3tson Opening Date for Colorado's Casa Bonita Announced

Lol, then you probably don't live in Denver and are not their target audience.

It is a famous restaurant in Denver featured on south park. I can only hope their food is better.

ttt3tson Facebook increased satisfaction and usage by sending fewer notifications

My web based email client has notifications allowed.

I am pretty happy browsers added support for this. It is disabled by default and easy to say no forever if a site is annoying about it.

ttt3tson Django makes it easier to build better web apps more quickly and with less code

I think you made your own point. It has been around and maintained for 20 years. The number of web stacks that can claim that are very few. Worth mentioning once in a while.

ttt3tson Django makes it easier to build better web apps more quickly and with less code

Compared to the constant bitrot in "newer" frameworks I have grown to like Django more over the years. My old projects in Django still work. The meteor projects, flash apps, Android apps, less so. Modern software undervalues stability.