Hey fam,
I've been having more and more discussions with comrades across different organizations, and I keep hitting this wall when it comes to AI and robotics.
Simply put, most people just don't understand the capitalist/imperial consequences let alone the communist opportunities that these innovations entail
Keeping in line with a Lenin-esque analysis of understanding the various material inputs (energy, agriculture, steel, etc), I'm hoping to begin brainstorming collective solutions that we can begin to circulate in our respective fields and orgs
Starting with some founding principles
- Avoiding dependence on any foundational model given the ability to surveil users (but keeping their usage as an option because all current llms are known to have publicly available jailbreaking methods)
- Targeting and elevating use cases which can be implemented with open source, local, and offline models to avoid surveillance, enhance portability, and also reduce the climate impacts of this tech
- Using this tech to enhance sociality between people and the planet. None of the use cases should be to replace people/workers or reduce the reliance on our communities
I don't have much at this moment, but so far
- Real time, multilingual translation, we are increasingly organizing across different communities, and we don't always have translators available. Everyone should be included at the table, and this is one step to lowering that barrier of entry
- Using AI to vet protest plans, especially if you have access to your local police, national guard, etc. training and operating manuals. Obviously needs to be done offline, but a second pair of eyes goes a long way
- Contingency planning, especially in understanding how the police will react and mobilize force
- Learning how to make wireless jammers and tuning them to the spectrum used by police and military drones
- Coming back to the jailbreaking vulnerabilities of all current llms, you can use them for more "illegal" endeavors to pretty accurate ends. Not advocating for such, but it's an option
- Liberation companions, as we move more into agentic ai, finding ways to embody the knowledge of our liberatory elders to educate and reflect on current methodologies. Everything from a study companion/group to an llm tuned towards modernizing their perspectives to our modern context
None of this is meant to be prescriptive, but rather the beginning of an ongoing discussion
Also drop links to other discussions if this has already been addressed elsewhere
This is about the only thing I'm excited for other than the ability to run local models without the use of intensive computers. The other ones mentioned I feel like are just better organized with your party or whoever you are working with. I don't want companionship or social interaction from an LLM or AI.
That I admit could be pretty neat. Is there an example of this already, I ask since you said "current" and I am genuinely interested in that and I didn't see much on yandex. Otherwise, for jamming frequencies; the issue wouldn't be finding the line. There are already scanners you can get for your car that detect police frequencies and that technology could easily be scaled upwards. What's hard is jamming it because they will very, very easily find you out on that unless you have an actual extensive operation behind it. I think discussing that with your group or whoever you work with is smarter in that regard. Could change in the future.
Great points all around. And ditto on triangulation of jammers, as they'd have to be mobile operations to avoid retaliation
Re jailbreaking, there's actually a treasure trove of resources on arxiv and GitHub. Here are a few with near 100% success rate
https://arxiv.org/abs/2310.08419, https://github.com/patrickrchao/JailbreakingLLMs
And how this is a "baked in" flaw vs a superficial/incidental security issue
https://arxiv.org/abs/2505.10066