It would. But it’s a good option when you have computationally heavy tasks and communication is relatively light.

Once configured, Tor Hidden Services also just work (you may need to use some fresh bridges in certain countries if ISPs block Tor there though). You don’t have to trust any specific third party in this case.

If config prompt = system prompt, its hijacking works more often than not. The creators of a prompt injection game (https://tensortrust.ai/) have discovered that system/user roles don’t matter too much in determining the final behaviour: see appendix H in https://arxiv.org/abs/2311.01011.

Like Firefox ScreenshotGo? (I think it only supports English though)

https://arxiv.org/abs/2305.17743

xkcd.com is best viewed with Netscape Navigator 4.0 or below on a Pentium 3±1 emulated in Javascript on an Apple IIGS at a screen resolution of 1024x1. Please enable your ad blockers, disable high-heat drying, and remove your device from Airplane Mode and set it to Boat Mode. For security reasons, please leave caps lock on while browsing.

• https://xkcd.com

CVEs are constantly found in complex software, that’s why security updates are important. If not these, it’d have been other ones a couple of weeks or months later. And government users can’t exactly opt out of security updates, even if they come with feature regressions.

You also shouldn’t keep using software with known vulnerabilities. You can find a maintained fork of Chromium with continued Manifest V2 support or choose another browser like Firefox.

Very cool and impressive, but I’d rather be able to share arbitrary files.

And looks like you can only send images in DMs, but not in groups/forums.

If your CPU isn’t ancient, it’s mostly about memory speed. VRAM is very fast, DDR5 RAM is reasonably fast, swap is slow even on a modern SSD.

8x7B is mixtral, yeah.

Mostly via terminal, yeah. It’s convenient when you’re used to it - I am.

Let’s see, my inference speed now is:

• ~60-65 tok/s for a 8B model in Q_5_K/Q6_K (entirely in VRAM);
• ~36 tok/s for a 14B model in Q6_K (entirely in VRAM);
• ~4.5 tok/s for a 35B model in Q5_K_M (16/41 layers in VRAM);
• ~12.5 tok/s for a 8x7B model in Q4_K_M (18/33 layers in VRAM);
• ~4.5 tok/s for a 70B model in Q2_K (44/81 layers in VRAM);
• ~2.5 tok/s for a 70B model in Q3_K_L (28/81 layers in VRAM).

As of quality, I try to avoid quantisation below Q5 or at least Q4. I also don’t see any point in using Q8/f16/f32 - the difference with Q6 is minimal. Other than that, it really depends on the model - for instance, llama-3 8B is smarter than many older 30B+ models.

Have been using llama.cpp, whisper.cpp, Stable Diffusion for a long while (most often the first one). My “hub” is a collection of bash scripts and a ssh server running.

I typically use LLMs for translation, interactive technical troubleshooting, advice on obscure topics, sometimes coding, sometimes mathematics (though local models are mostly terrible for this), sometimes just talking. Also music generation with ChatMusician.

I use the hardware I already have - a 16GB AMD card (using ROCm) and some DDR5 RAM. ROCm might be tricky to set up for various libraries and inference engines, but then it just works. I don’t rent hardware - don’t want any data to leave my machine.

My use isn’t intensive enough to warrant measuring energy costs.

The article isn’t about automatic proofs, but it’d be interesting to see a LLM that can write formal proofs in Coq/Lean/whatever and call external computer algebra systems like SageMath or Mathematica.

I see, thanks. Will check. I just thought perhaps you figured out something other than those from your experience.

Any guidance on choosing appropriate conservative settings for i7-13700K? I may be hit with the same as you in the future (sometimes I have to do some heavy multithreaded combinatorial computations which run several days with 100°C temperature, using all cores). The motherboard has options for customising pretty much everything there is, but I didn’t touch anything overclocking-related, so I have Asus defaults.

I’m still waiting for the day when actual ads across the internet drown in AI-generated advertisements pointing to no real product or service. Perhaps that’ll make attention industry collapse?

If you’re looking for a side project idea, here’s one.

You’re welcome!

As far as I understand, all of them can be made to work locally (especially if your local model is served via an OpenAI-compatible API, e.g. see llama.cpp’s server binary) with varying degrees of effort required.

• https://github.com/neuml/txtai
• https://docs.llamaindex.ai/en/stable/
• https://docs.gpt4all.io/gpt4all_chat.html#localdocs-plugin-chat-with-your-data
• https://github.com/atisharma/llama_farm
• https://github.com/jonfairbanks/local-rag
• https://docs.useanything.com/
• https://python.langchain.com/docs/get_started/introduction
• https://docs.neum.ai/get-started/introduction

Never ran RAG, so unfortunately no. But there’re quite a few projects doing the necessary handling already - I’d expect them to have manuals.

I’m using local models. Why pay somebody else or hand them my data?

• Sometimes you need to search for something and it’s impossible because of SEO, however you word it. A LLM won’t necessarily give you a useful answer, but it’ll at least take your query at face value, and usually tell you some context around your question that’ll make web search easier, should you decide to look further.
• Sometimes you need to troubleshoot something unobvious, and using a local LLM is the most straightforward option.
• Using a LLM in scripts adds a semantic layer to whatever you’re trying to automate: you can process a large number of small files in a way that’s hard to script, as it depends on what’s inside.
• Some put together a LLM, a speech-to-text model, a text-to-speech model and function calling to make an assistant that can do something you tell it without touching your computer. Sounds like plenty of work to make it work together, but I may try that later.
• Some use RAG to query large amounts of information. I think it’s a hopeless struggle, and the real solution is an architecture other than a variation of Transformer/SSM: it should address real-time learning, long-term memory and agency properly.
• Some use LLMs as editor-integrated coding assistants. Never tried anything like that yet (I do ask coding questions sometimes though), but I’m going to at some point. The 8B version of LLaMA 3 should be good and quick enough.

Rounding?