|
Top-level
@i0null @coupland @velinion @oliverherold @porru some see AI as a tool of a few big companies. but some big companies have an incentive to opensource their models (e.g. Meta is behind OpenAI , so to counter that threat, they've released free opensrouce LLMs to reduce demand for OpenAI) I'd like to see everyone running their own AI locally (text+images) instead of leaving it to cloud services ironically anti-AI campaigners are going to kill this possibility if they win 9 comments
@i0null @coupland @velinion @oliverherold @porru Local models have a secret weapon - finetunes which can be done on high end gaming rigs. The bigger AI's do 'MoE' where they branch between several experts. Opensource models can in effect be the ultimate 'MoE' , distributed across the web. IMO open models are the way to ensure we all have maximum input inguiding how they're used, what they are, and ensuring widest benefit @i0null @walter4096 @velinion @oliverherold @porru The kind of prescient risks we should be talking about wrt AI are things like job displacement and wealth disparity. AI will create trillions in wealth how do we prevent those trillions from going to, like 2 people. But that's an economic & ethics discussion, it really has nothing to do with AI itself. Instead we're distracted by "how do I get paid for every ChatGPT search" or "AI=evil" because it drives "likes" or "ZOMG! Extinction event!" @coupland @i0null @velinion @oliverherold @porru yeah this is why I back opensourced AI. I'm resisting using GPT and giving mindshare to the less capable LLaMa's that you can run locally. people kind of sleepwalk into this centralization of power by becoming *too* dependent on cloud services. the good outcome is where we all keep buying powerful PC's and do as much locally as possible you could be paying for a big tech service (->disparity) or it could be saving you money (lift all boats) @walter4096 @i0null @velinion @oliverherold @porru > people kind of sleepwalk into this centralization of power by becoming *too* dependent on cloud services Just like how we've sleep-walked into surveillance capitalism and climate crisis. Why tackle tough issues when we can be lulled to sleep by shiny toys? Really appreciate all the perspectives, great discussion. @i0null @coupland @velinion @oliverherold @porru yeah regulatory capture is exactly my fear. I'm just a lone tinkerer at the minute - I see huge potential being unlocked with these opensourced models , and a bunch of people arguing they should be shut down :/ I've been pro-tech all my life ever since my first 8bit machine. I've always seen it as an individual enabler, and without being 'left or right wing', opensource is very inline with that.  @walter4096 @coupland @velinion @oliverherold @porru Yeah it's such a shame that a technology with such emancipatory potential is more likely to hand greater power to the biggest players. Many people say technology is neutral but the application of it certainly isn't. I’ve been fairly surprized by Meta's commitment to open source but after that whole Cambridge Analytica scandal IMHO they're going to need to do a whole lot more to regain some reputation. @i0null @coupland @velinion @oliverherold @porru it's great how this emerged as a balancing mechanism meta, google, microsoft,apple all saw what OpenAI were doing, and panicked. MS bought OpenAI to stop them growing as the next tech giant(ie conversational interface) Meta then opensourced their own (inferior) LLMs ("ok, we dont think we can get to 1st place, lets at least empower opensource community to stop Microsoft dominating completely") |
Gregory's Server
@walter4096 @coupland @velinion @oliverherold @porru
If the models and datasets are open i think that does exemplify to some extent. However soon they will be so large that it will be impossible to run locally. obviously open models are concern to those who want to gatekeep but also those who want to circumvent restrictions. There are legit safety concerns about all this but I’m convinced a few in the industry would rather talks about sci-fi X-risk as a way of diverting attention away from the more prescient risks.