I do wonder, hypothetically, if free Linux distros had 80% of the consumer market, would we see just as many dangerous exploits and malware as we do on Windows today? It seems to me that the consumer community is so small that it’s hard to say if it’s secure or just obscure.

I understand in theory Linux is more secure… But are individual users really not opening themselves up to attacks, downloading foss software right and left? Using built in stores? Wine emulation?

I’m so happy that new pipe is working again.

I use virtual credit cards from privacy.com. for trials I set a $1 limit and forget about it. It’s pretty useful for legitimate subscriptions too, since I can pause or end them just by pausing the card.

The attrition is slow, but every user lost to Linux is likely lost forever. After a year or so of totally free software, who is going to build a new windows compatible PC, buy a Windows 11 license, and pay for subscription service just to do word processing, or play a few incompatible games?

Windows completely overestimates people’s willingness to throw out their laptop or PC just to get a new OS paintjob. For every person who does it, another one will leave their ecosystem forever.

I believe Librechat would achieve your goals, but you’d need a PC or server to host it in. It supports all major API.

Kobold light might with as well, and doesn’t need to be hosted locally, but I don’t think it supports Claude haiku specifically, for unknown reasons.

Additionally, the official Claude api workshop is pretty good on desktop, but it only supports Claude.

I mean, be careful. These llms can be honeypots for data. Like, if you’re using it for cover letters, or work, you’re sending tons of personal info to random websites.

I would recommend sticking to actual, reputable vendors for llms, or running your own. I have a GTX 1070 and can run some pretty decent models these days locally using koboldai.

Bing is probably the only way to use gpt 4 without paying for it, and Microsoft probably won’t steal your bank account info.

There is almost no chance that it is truthfully based on gpt 4. If you want a free, open source llm with 32k context and generous limits, I recommend using huggingface.co/chat/

The nous-hermes model (you can select different models) is uncensored, and performs really well for a open source model. Plus, they have data controls so you can turn off data gathering per model. Huggingface is a reputable vendor, and doesn’t claim to be something it isn’t.

This feels… Scammy? Not to be accusatory, but gpt 4 is expensive to run. It is impossible for people to use it for free.

What llm is actually providing the response here? Either someone is footing the bill for an API and acting as a proxy, a situation which raises many red flags, or the model you’re talking to is something far cheaper to run, like a mistral model.

Even the second case is sketchy. 😅