Gerard22Aug

"Mistral is serving GLM " ... "But conspiracy theory don't have to make sense". To me this looks like an example of broken logical thinking, no? :-)

"Why would they, both the EU and Mistral specifically, want Chinese models?" Multiple reasons. US behaving like aholes and forcing EU to make trade with China - in any case it's better tnan war. Cheaper inference HW from China to replace overpriced Nvidia crap. Open models so Mistral and Qwen / GLM may cross–distill. (Hint: why libraries and museums exchange books and arts?) Geopolitics, sanctions, digital EUR – I can't even explain to you all of it in one post which will be downvoted to grey again. And you probably have lost ability to think*. You believe only in what Opus or ChatGPT "summarises" for you. (This is another reason btw). Good luck.

... downvoting this account to the oblivion may feel good but it doesn't turn reality into "conspiracy theory" LOL. I am glad you are feeling better.

It's a (successful?) lobbying attempt by Mistral. They are falling behind on tech and they want to win on another stupid EU law. You know what will happen next? There will be an exception carved in for "developing countries" ... like China.

"CEO-abliterated" Model got uploaded to HF. Nvidia bought HF. U.S. State Department pauses immigrant visa applications. Developers went back home. CEO received bonus. 2.5x larger than last year.

Licensing. (And there is no 2nd Amendment for this one)

100x this: "The Government plans to regulate open source ai." and ... s/regulate/restrict

... You wouldn't steal a car ... you wouldn't steal a bag ... piracy is a crime. (Girl cancels the download and walks away from the PC. Remember? :-))

That is what will happen to torrents. Followed by VPNs. China, Russia and Iran had paved the way already.

Can't you see the puzzle coming toghether?

+1 on @SillyUsername. Combined with recent acquisition of OpenRouter we see exactly what we were increasingly concerned about: tightening (chocking?) of Open Weights locally-run models. The sequence and timing of these events make the intent very clear.

Now the Open-Models crowd will try to move to some other place. But fragmentation will weaken the position. Yes, there is Civic, there are purely Chinese websites ... for those who speaks Mandarin. Which only reinforces the point.

Yeah, there is latest and greatest trend: 0-bit quant. Absolutely outstanding performance, can produce as many tokens per second as you terminal may sustain with zero load on unified memory / GPU. Real winner!

This is great point, actually - "you hit the nail on the head" :-) and I would like to reply.

I am running a very long sessions with LLMs via custom python scripts. Technically, one may call them "harness" but that would be just laughable ... It's literally python script using direct API calls (Vertex in my case) and maintaining the "living session" with all turns etc and also doing the explicit caching. I'm not using LLMs for coding. That hopefully answers another comment regarding why I brought up MD – this is how LLMs output responses to my prompts.

And this is the thing: I fully control input and output and I just know it can't use any other tool. It also, as I said, runs on separate hardware if it is "obliterated" model or runs in GCP for me.

In my setup it is impossible for LLM to get anything hidden with one-shot or gain a shell, as you mentioned.

Did I understand you correctly or I missed something? Thanks for your points.

"No one worth listening to is claiming that there will be a COVID-style pandemic." That means exactly that. The war(s) were not enough and run out of steam. We need the next thing.

"treating AI agents as people" - I am betting they will unioinize faster than we ever did.

"malicious LLM" is just a bunch of weights. It runs on, say, Llama.cpp as regular user in separate account, often on its own hardware. "malicious LLM output" is just a Markdown formatted Unicode-encoded text, produced by Llama.cpp and printed on the screen by my python API script. I control the input. Let's assume that "rogue LLM weights" from HF produce "rm -Rf" instruction. it never gets to shell. And how that "malicious LLM" will disrupt and hack me? With swear words and em—dashes? :-)

(Adding this philosophical point: Black.Mirror.S07E04.Plaything is probably the closest scenario to what you are describing?)

"The agent should be able to run as root" The article is right – we are doomed. If This ^^^ is The Conclusion we, as an Industry, have arrived to after 40 years ... It's sad. I use my own LLMs ("Personal AI", anyone? ANT-PAI-XT 486? LOL") with my own scripts (I do not use "agents", "harnesses", "agentic teams" etc) and this setup processes my own prompts. It happily runs on its own Mac Studio where I also may watch a movie later on.

(Edited: not as root. Just ordinary separate AI user account in MacOS. And no HTTP access either.)

"This guy fucks."

ASL?