I don’t know why this is being taken at face value with so many upvotes. The Gensler SEC was right to go after actual scammers and ponzis, but they went much further and clearly had an agenda.

Gensler targeted the most reputable exchange in the US alleging that their core business is illegal, because the Gensler SEC decided to classify crypto assets as securities rather than define a new regulatory framework that actually fits.

https://www.sec.gov/newsroom/press-releases/2023-102

Coinbase wanted to follow the rules and spent years asking for clarity. Rather than provide clear rules, the SEC provided a lawsuit.

deleted by creator

The one thing I’m holding out hope for is that things like this won’t lead a significant number new people into getting duped.

There’s a certain portion of the population that seems to question nothing and got duped long ago by all of those “every immigrant is a rapist and murder”-type Facebook posts with links to totallyrealnews.com. That segment of idiots remains unchanged. They will believe any wild claim and never fact check as long as the misinformation fits their world view.

The question is, how many people who aren’t already lost see these low-effort AI posts and take them at face value?

Disclaimer: I say this knowing full well that well that better quality and more subtle misinformation is also possible with AI, but this ain’t it.

Yes, fuck Israel and fuck Russia. Not sure why I’m responding to this dumb bait, but here we are. It’s not a straw man argument when both countries are run by literal human feces

This is a major problem for all democracies, and LLM driven troll accounts probably do exist. But this xitter post is a fake error message. It’s clearly a troll.

Blocking fake accounts would help with the misinformation problem, but it’s a cat and mouse game. It could ultimately give additional credence to the trolls who slip through if the platform is assumed to be safe. The reality is that there will always be ways for fake accounts to avoid detection and to spoof account verification. Making it harder would help, but it’s not a comprehensive solution. Not to mention the fact that the platform itself has the power to manipulate public opinion, amplify their preferred narrative, etc.

The solution I’ve always preferred is the mentality the 4chan community had when I was younger and frequented it. Basically, and I’m paraphrasing:

Everyone here needs to grow up and understand that no post should ever be presumed to be true or legitimate. This is an anonymous forum. Assume that everything was written by a bot or a troll in the absence of proof that it wasn’t.

I think people put too much trust in social media precisely because they assume that there’s a real person behind every post. They assume that a face and a few photos gives an account legitimacy, despite the fact that it’s trivial to copy photos from a random account (2015/16 pro-Trump Facebook style) or just generate all of the content from scratch with AI (to avoid duplicate detection).

Trust itself is driver of misinformation. On social media, people should only fully trust posts made by people they know. That is the simplest and most comprehensive solution to the problem.

100%. I subscribe here to learn about new advancements—to learn about technology.

The finances and politics of the tech industry have a home in those respective communities.

Yep. It bothers me, but I respect the decision.

First of all, they spelled Heelys wrong. Second, Heelys are a great idea, even better as an adult in an office with polished concrete floors.

This one is only 7B parameters, but it punches far above its weight for such a little model:
https://huggingface.co/berkeley-nest/Starling-LM-7B-alpha

My personal setup is capable of running larger models, but for everyday use like summarization and brainstorming, I find myself coming back to Starling the most. Since it’s so small, it runs inference blazing fast on my hardware. I don’t rely on it for writing code. Deepseek-Coder-33B is my pick for that.

Others have said Starling’s overall performance rivals LLaMA 70B. YMMV.

Of all of the code specific LLMs I’m familiar with Deepseek-Coder-33B is my favorite. There are multiple pre-quantized versions available here:
https://huggingface.co/TheBloke/deepseek-coder-33B-base-GGUF/tree/main

In my experience a minimum of 5-bit quantization performs best.

Lower Wacker in Chicago is a black hole

Gergely Orosz Sep 7, 2023 Paid

Paperclip

I’ll sell you the same albums for $1.19 🤫