Hacker News Reader

Comment by jddj | Hacker News Reader

Thanks for desensationalizing that awful title.

I also found the part about effective accelerationists and "decels" to be amusing. Sometimes it feels like we're living in a badly written sci-fi.

View on HN

> $260 Million AI Company Releases Undeletable Chatbot That Gives Detailed Instructions on Murder, Ethnic Cleansing

Holy cow, you weren't kidding.

Are there any alternative sources so we can deprive this publisher of attention and resources?

It's safe. Before they can commit Murder and Ethnic Cleansing they have to get their hands on a NVIDIA RTX 4090...

Except it’s already available at labs.perplexity.ai so you can easily indulge.

Its a 7B model with GGUFs already on Hugging face, with down to a 2-bit quantization. It’ll probably run on an RPi; you definitely don’t need an RTX 4090.

Well (doesn’t addressed “they released on it BitTorrent” hysteria bit, though):

https://dataconomy.com/2023/09/28/mistral-7b-mistral-llm-lla...

Yes, the original tweet with a magnet link seems like a good source.

They're amusing but they're definitely dangerous to the potential of AI in the medium term. People like to say that "these things will happen anyway" even if it's made illegal, but there's no way 250 million will be invested into a company to develop LLMs if they're made illegal or heavily regulated. ML is resource intensive as it is (in terms of number of PhDs required and compute) and if the decels and censors have their way they definitely can make a dent in progress (i.e. "it'll happen anyway" isn't true, or it'll be slowed down so much nothing interesting will happen within our lifetimes).

The philosophical arguments are definitely real and valid, but I still find the clans and labels funny.

It's over for decels.

If that company moved to the US, their model could never be regulated because of Bernstein v DOJ.

We shouldn't need corporations for this anyway. Why can't we decentralize training? Model training à la folding@home.

Training needs insane amounts of inter node bandwidth, to the extent that training clusters use specially built hardware for it. Decentralised training isn't physically possible anytime soon, maybe never.

Reply on news.ycombinator.com