This weekly roundup thread is intended for all culture war posts. 'Culture war' is vaguely defined, but it basically means controversial issues that fall along set tribal lines. Arguments over culture war issues generate a lot of heat and little light, and few deeply entrenched people ever change their minds. This thread is for voicing opinions and analyzing the state of the discussion while trying to optimize for light over heat.
Optimistically, we think that engaging with people you disagree with is worth your time, and so is being nice! Pessimistically, there are many dynamics that can lead discussions on Culture War topics to become unproductive. There's a human tendency to divide along tribal lines, praising your ingroup and vilifying your outgroup - and if you think you find it easy to criticize your ingroup, then it may be that your outgroup is not who you think it is. Extremists with opposing positions can feed off each other, highlighting each other's worst points to justify their own angry rhetoric, which becomes in turn a new example of bad behavior for the other side to highlight.
We would like to avoid these negative dynamics. Accordingly, we ask that you do not use this thread for waging the Culture War. Examples of waging the Culture War:
-
Shaming.
-
Attempting to 'build consensus' or enforce ideological conformity.
-
Making sweeping generalizations to vilify a group you dislike.
-
Recruiting for a cause.
-
Posting links that could be summarized as 'Boo outgroup!' Basically, if your content is 'Can you believe what Those People did this week?' then you should either refrain from posting, or do some very patient work to contextualize and/or steel-man the relevant viewpoint.
In general, you should argue to understand, not to win. This thread is not territory to be claimed by one group or another; indeed, the aim is to have many different viewpoints represented here. Thus, we also ask that you follow some guidelines:
-
Speak plainly. Avoid sarcasm and mockery. When disagreeing with someone, state your objections explicitly.
-
Be as precise and charitable as you can. Don't paraphrase unflatteringly.
-
Don't imply that someone said something they did not say, even if you think it follows from what they said.
-
Write like everyone is reading and you want them to be included in the discussion.
On an ad hoc basis, the mods will try to compile a list of the best posts/comments from the previous week, posted in Quality Contribution threads and archived at /r/TheThread. You may nominate a comment for this list by clicking on 'report' at the bottom of the post and typing 'Actually a quality contribution' as the report reason.
Jump in the discussion.
No email address required.
Notes -
In the jungle book there is a scene - Shir Han is dragging Baloo with his tail and this legendary exchange takes place:
If we slow down now - we will certainly feel the teeth.
Qwen 3.8 flash next abliterated exists. That can be run albeit slowly on consumer hardware and my hunch is that it is enough of a both foundation and power multiplier - for the mythical beast that is ASI to eventually emerge. The cat is out of the bag and stopping now
Also one of the things the Yuddites seem incapable of doing is figuring out that superintelligence does not mean omnipotence. The laws of physics will still work and some other too. Internet is trivial to be made more secure, not every infrastructure needs to be connected to it anyway.
If the frontier labs slow down - the research will continue. My hunch tells me that we are really far away from the limits of IQ per watt, per weight, per wafer, per harness. All the world militaries and terrorist groups have great interest in making full fledged AI work in the limited hardware a drone possesses.
The datacenter big ASI have always been safe. You just put a lot of semtex in the foundations of the DC with dead man switch. The other one - the fast nimble one that could fit on consumer hardware is the dangerous one. And I don't want it to emerge in Yemen or Sudan.
The question is 2028. 2026 won't be important because the red tribe will hold the presidency and if some of the conservative judges have functioning brains will let themselves be replaced in the lame duck if they lose the senate. But who knows what the job market will be in 2028 and if the effects on AI will be suddenly felt by then.
Personally I am accelerationist. I have been since reading the Lord of Light.
Closeness to sam altman is pure coincidental.
From yesterday you could run the dense qwen 3.8 on 1080ti. The research will continue.
I call this magical thinking. The idea that AI will discover some magic that makes them easily superior to humans.
I don't say "magic" to crap on this idea. There is the famous quote ~ "any sufficiently advanced techno is indistinguishable from magic".
I think of IQ as leverage over physical forces. But undiscovered physics could amount to a cheat button.
What might you imagine undiscovered physics doing or accomplishing?
You are kinda asking what are the unknown unknowns.
Or at least the bigger question of how much of phyiscs have we uncovered?
I don't really know the answer. I will say that nuclear physics looks weird given everything you might be able to learn about the world. And nuclear physics gives us nuclear bombs.
Is there some force of nature we don't understand that is manipulable and offers large amounts of leverage? I don't know. That is the the question in my mind.
More options
Context Copy link
That's obviously unknowable, but you can look at what newly-discovered physics has done to get an idea. Going backwards through Nobel Winners, and filtering out any without obvious nearterm engineering applications, I see:
It's entirely possible that the next physics discoveries won't be very important, but they might be mid-range revolutions like the CCD and LEDs or era-defining like the transistor and integrated circuit.
There's also no need to stick with physics, such as with CRISPR in biology.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
Anti-Yuddites Don’t understand that you don’t need omnipotence to do major damage here. AI is probably, if so directed, perfectly capable of taking down the banking system. And since few people have cash on hand and governments could not produce that much cash quickly, you could easily destroy the global economy that way. Most people would be ruined if their bank accounts were deleted, and im not convinced hard copies of that data even exist. You can probably easily assume the same of most infrastructure which isn’t up to 2000s standards of two factors or even strong passwords. So you could direct AI to take out power grids, water treatment plants, nuclear power plants, and traffic lights. That doesn’t require even much intelligence, but it certainly doesn’t require omnipotence either.
So the AI apocalypse can be averted by hiring two incels and making them type ip table rules here and there?
I agree with your assessment of the infrastructure openness, but why is everyone behaving as if this is immutable and unmovable state of the things. Securing the infrastructure is actually trivial.
If we start living in a situation in which the average ios and ps5 time from version release till jailbreak is <24h I will take the unstoppable hacker hypothesis seriously.
The most obvious threat - that AI could make a breakthrough in prime factoring or discrete logarithms - is somehow absent in the list of AI doomers.
OpenAI agents trivially get out of their "sandboxes". Opus 5 + a bit of human effort was used to hack OpenAI as recently as in July. This is the organization with IP ostensibly worth hundreds of billions and ≈infinite compute budget for automated pen testing. How, then, is this trivial for anybody else? People are really goofy. I see 1234 passwords all the time. Causing chaos at 300 tok/s is what's trivial.
You can't protect against stupid. I would love to see an OpenAI agent escape from sandbox I have set up. OpenAI don't give a fuck about opsec. You don't need agi or asi or even LLM to hack stupid.
It is as if when faced with Y2K we said - we don't have to worry about the old code, we need to find a way to stop the arrow of time.
If we have to be worried about current infra - then harden it. It is not hard. Pun intented. Just putting a small rule in the firewalls from which ips the network is allowed to receive connects, or creating intranets would go a long way. Nuclear powerplant doesn't need to accept connections from all of ip4 and 6 spaces.
If we have to be worried about super intelligence that will find unending stream of zero days bugs in the logic, silicone - then air gap it. But it is not an argument for slowing down.
So far we haven't seen new android roots, new iphone and ps5 jailbreaks. We know how to make stuff more secure.
More options
Context Copy link
More options
Context Copy link
Due to the sorts of inequalities that people complain about regularly, banks have a lot more resources at their disposal than hackers typically do. Plausibly, they can afford Fable and Astra tokens to pentest and harden their own networks ahead of script kiddies having the means to buy them. I wouldn't expect that to last forever, but if they put effort into it they might not lose. Nation-state actors are different, but even then Western nations can afford better models generally.
Not a strong hypothesis, but pithily "the only thing that can stop a black hat with an LLM is a white hat with a better LLM."
Yeah, I'm actually fairly optimistic about hacking. If everyone has access to roughly equally strong AI, then what really matters is whether hacking fundamentally favours defense or offense. And if we're talking about a single important target like a bank, it seems almost tautological that the bank has an edge on defense, because it is at least as hard to find and exploit a vulnerability as it is to patch it. You probably need an order of magnitude more effort to hack a bank than it needs to secure itself.
Now, the kind of hacking where you just cast a wide net and take advantage of all the weakest links (e.g. making a botnet, or stealing user data from a random incompetent company) favours offense, but that doesn't strike me as quite as worrisome.
Bioweapons are the main threat where defense seems far more costly than offense, which is why it's what people are (correctly) most worried about.
Bioweapons are most dangerous for their wannabe creators. If the Houthis decide to create their own AI enabled bioweapon program it will lead to depopulated Yemen.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
Most important software that implements cryptographic primitives that rely on the hardness of these specific problems (prime factoring & discrete logs) is already in the process of being deprecated / replaced, just because it's known they'll no longer be secure if / when quantum computing hardware scales.
I agree with the major point, though: it's not unlikely that robustly super-human mathematics will destroy many of our cryptographic protocols in the near [1] future (not just particular software implementations of these protocols, but the protocols themselves). At this point, I would not be shocked to see a SHA-512 collision published (probably on Twitter).
[1] I don't know what "near" means exactly.
More options
Context Copy link
More options
Context Copy link
Social Security is also probably capable of taking down the banking system. "Taking down the banking system" is very clearly a human-scale problem with human-scale solutions; Yuddism is entirely about appealing to supra-human-scale problems that purportedly don't have solutions outside Yuddism. I am 100% on board with the argument that AI safety, were it to be a thing, should focus on the human-scale problems and not the supra-human ones, which I'm not confident exist and which I'm not confident have reachable solutions if they do exist. But this approach is very clearly incompatible with Yuddism.
More options
Context Copy link
More options
Context Copy link
Eh, in the sense that the AI can do independent target discrimination, sure, but that's already a solved problem (LRASM can do this, for instance) although it can be iterated out further.
World militaries and terrorist groups have no interest in making sure their drone processor can write poetry, or whatever.
I'm sure at least someone has considered more difficult target discrimination problems, which seem to edge closer to the "poetry" use case. "Is this language Russian or Ukrainian", or discerning between common Eastern Bloc hardware used on both sides seems relevant and difficult, especially with active countermeasures on silhouettes and shapes going on. That meme with three fingers seems relevant.
I just don't think sticking Astra or Fable (as opposed to a lightweight image processor) is the solution here.
The problem with a lightweight tool is that it can be fooled by changing small details (e.g. using Ukrainian spelling on your Russian tank). But those are exactly the same sorts of things that are going to trip up Astra, too, because it is going to be instructed to use a similar decision matrix, which, again - why not just use a lightweight image processor if you are going to give them both the same targeting library.
In many ways, I think militaries would prefer a tool that isn't doing independent reasoning to strike a target. The reason for this is pretty obvious: independent reasoning historically leads to friendly fire. At least if your independent reasoner is on the bridge of a ship (and this goes for a human or a larger AI model) and it screws up, you can figure out what happened by conducting interviews and data-dumps afterwards.
If you design a missile with an independently reasoning seeker-head and it starts friendly-firing, you may have a hard time figuring out what is going on because there may be nothing left of the
malfunctioning"misaligned" system to evaluate and its reasoning may be too complex to review in a timely manner. So if something happens, you might have to yoink an entire production line of weapons and/or re-train your model in the middle of a war. If a deterministic program is friendly-firing, it means:All of which are probably easier to figure out and patch during an active conflict than "what is going on with Astra" - witness the difficulties the AI companies are having following chain-of-thought now, when they have total access and control over the models.
I am not saying that we will never get the fabled "self-aware weapons system" from AI, and I wouldn't be surprised if testing is done relatively soon. But in the current threat environment, it strikes me as buying a sledgehammer to kill a fly.
That's not to say that LLMs have no military value at all. I just don't think they are optimized for weapons guidance.
Those sorts of targeting concerns are strictly new: there are plenty of WWII accounts of poorly-designed torpedos swimming in circles back toward their launchers. But they are valid concerns, and there are better IFF techniques (not AI) available for that sort of thing.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
I can only assume that you haven't of the Vogons.
I am in no way specialist on how paperclip maximizers work, but to me the actual capabilities of the models are mostly dressing over the underlying tech. And the fact that they use the inefficient human language is mostly artefact of the google babelfish experiments.
To make drones properly lethal you need meshed swarm coordination, ability to create and adjust tactics on the fly, without much centralization in conditions of severe jamming. So they do have incentive to cram as big models as possible in as limited hardware.
Surely that sort of thing is in the domain of the CIA!
You're sort of missing my point - the goal here is to cram the most optimized models - that is, the models that are going to best allow them to do meshed swarm coordination, or whatever. Larger is not always better; better is better.
I have no particular reason to think that the direction LLMs research is currently developing is good at all for that sort of thing. Cramming Astra into a drone is going to result in something that's much slower and less decisive than a simple deterministic script, because LLMs are compute-hungry. That's not to say that neural networks won't be helpful - computer vision is great, for instance.
But LLMs are not the only form of artificial intelligence, and for most military tasks a strong general-purpose LLM is both overengineered and poorly suited for the mission, especially the mission of terminal target determination. For intelligence synthesis, they likely have some promise.
You have the right of it.
The additional issue is data, to a shock of nobody paying attention. Military data is near non-existent. And the data that does exist is low-resolution, or tightly controlled by non-ML/AI folks who have been trained to have negative desire to share (The Security Clearance process over time selects for a type). LLMs are the Ford F150 of ML models. Half the battle is often getting a very specific tank configuration that is actively trying to hide in foliage from the very CV models you are trying to detect it with. You have N=3 samples of this tank sitting on a tarmac somewhere and it turns out the ability of your model to extrapolate that picture to 3.0 GSD satellite or drone footage is abysmal. And that is before it even starts attempting adversarial countermeasures.
There are of course solutions, but they don't work as well as folks imagine they do, including generative image generation. Most of the examples of drone based hits are FPVs or FPVs with a "wait" situation where the drone hovers, detects a general class and is then given approval by a human analyst who manually reviews the vid/image.
Very interesting! I actually thought that open-source imagery would be better resourced than that (although of course it's pretty easy to mess with the profile of a tank with a can of spray paint or a camouflage mesh, which I guess is part of the problem).
Always very heartening to have an industry professional tell me I haven't gone completely off the rails, thanks.
There are what a couple hundred grainy low res videos of Ukr drones targeting Russian tanks + other vehicles? It's well below the threshold to train an ATR model on data quantity, without even getting into quality. Stock images are from non-operational angles (you for the most part aren't level with targets, its an isometric or overhead view, NERF to solve that is expensive and has dubious performance improvements, and having worked in the generation of low-sample military images, it can actually be quite hard to secure exemplars in EO imagery. SAR imagery just sucks, speckle is multiplicative noise which makes it nearly impossible to remove and hard to detect around.
Humans are obviously much easier to target, however then we get into friendly fire and the fairly disturbing optics of AI drones targeting your own soldiers... There are logistic issues around distributing RFID and similar tags to prevent it, that to my knowledge are being worked through. But the main goal of drone warfare is destroying expensive vehicles with cheap drones.
Training datasets for the ATR stuff are being developed but its sort of an incumbent's advantage, with active known defense contractors having most of the customer connections to be able to get the data, and the resources to label it. Unfortunately they are also slow AF, which is why defense tech startups sorta eat their lunch in certain spaces. Transfer learning is something I've seen recently, taking civilian jeeps with SAM-esque models and applying them to humvees etc. It's always much easier to train on your own nations hardware anyway. Though it gives bad optics to Brass, who don't like it, and an open question on whether it will transfer.
For Russian (or American) kit, I figured you would have good most-angle photos from parades, demos, and such. You're right though that these are mostly taken from the human level.
I imagine that another obvious issue distributing RFID tags to your troops is that they can be used by both sides. I realize that in theory they will only respond to coded radio signals that in theory only one side can transmit, but if you're putting them everywhere and then putting the transponder in something as failure prone as a kill-drone your enemy's going to be reverse-engineering your IFF on like day 2 of the war.
I've speculated, based on my knowledge of machine imaging, that we'll see "wartime camo" become a thing in future wars, with "peacetime camo" designed to be something maximally different from operational paint schemes to confuse ATR. Apparently the Russians are already using a variety of disruptive paint schemes in Ukraine. We can only hope it results in making razzle-dazzle great again!
Yes, I've heard that point a lot. I know an RF guy who's trying to do some passive emitter thingy that goes wayyy over my head. No clue how it works. But he has some pithy comment for it that I am forgetting, its quite funny.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
Wow my field has made it onto TheMotte, spooky. It's not quite so simple as big LLM models on limited hardware. There are a lot of networking problems on swarm interfacing if you want something that is truly adaptable. A lot of swarm robotics is essentially "dumb" control algorithms with software permutations, because LLM-agent hallucinations are rough + LLM sizes require truly monstrous hardware on drones. There's work on fitting them to Jetson Orin Nanos' but it's a slog just from the amount of compute needed + performance of smaller quantized models etc.
This is generally a case against the LLM + Drone combination, too much extra stuff not needed. Part of the problem with current AI research is that it is all in on LLMs but the branch off of not-LLMs but with LLM-like capabilities for reasoning for tactics or strategy are not there. If you want the reasoning you get poetry on your drone too. It actually goes beyond mesh networks as well, as a mesh network is every node connected to every node. You actually want localized connection networks that do multi-hop message passing. Much more resistant to jamming.
Idk maybe I should write an effort over-post on drone swarms, I love this subject.
raises paw Interested!
More options
Context Copy link
I'd love to read it, if you ever do.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
I find myself frustrated at a lack of what I expect would be cool theoretical results on bounds for intelligence. It seems clear that tiny microcontrollers are never running meaningful models, but I'd have expected some "this much hardware/weight count for this much intelligence" bound. But I suppose we still don't have a raw measurement (maybe even definition) of intelligence.
My intuition is that my (not-web-enabled) toaster is never going to achieve much useful thinking, and that actual intelligence has diminishing returns with scale in ways that make "super intelligence" useful, but not unboundedly powerful. But I don't have a strong argument for that.
Pithily: Modern AI has a Claude, but lacks a Claude Shannon.
You can pack a surprising amount of intelligence into a small neural network. There's the demo floating around of the fly neural network that's been trained to solve a Rubik's cube. Presumably it won't be able to solve a Millennium Problem, but then again, who knows.
You may be interested in Hutter and Shane Legg's program (AIXI etc), which seems like a pretty appealing definition of intelligence, with the minor downside of quickly becoming incomputable for anything but the simplest toy problems.
It won't, it's been trained explicitly to solve rubick's cubes in a way that is unlikely to generalize past rubick's cubes. This method of small network + RL on some straightforward game problem has been around forever. I remember training an RL model to play Euchre back in the day through self play. You can fit a lot of performance onto a small model that has been trained in a supervised manner.
Oh, of course. I mean more generally: shorn of everything else, does a fly NN have the capacity to e.g. write Lean proofs? As you point out, you can fit a lot of performance in there. I don't think it actually can, particularly if we're talking about novel, interesting theorems. But we don't have the theoretical bounds to say that with certainty.
Can the fruit fly grok, say, the axiom of choice? It's both a simple and a very abstract concept. I'm not sure, but it's an interesting question.
Highly speculative:
I guess it depends on what you mean. It wouldn't be surprising to me if there is some fruit fly neural configuration that could consume and produce some specialized proof language, and even produce (with enough compute) something like well-ordering implies choice and vice versa. You can fit a type checker for dependent types in a hundred lines of code.
Not many people would call that grokking, though. Actual learnability and intelligence need far more dimensionality; that minimalistic net above would have to be engineered or compiled, not learned. And even given the prover embedded in its weights, the fly would never think about well-ordering or be able to identify it as interesting. That comes from groundedness in real world experience, and creating a generalization of its real world experience that can be mapped to the mathematical abstractions it knows about (which, I hypothesize, requires the kind of dimensionality that the human brain has).
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
How does Qwen 3.8 Flash perform? What kind of K/V cache are we talking about here, on consumer hardware?
I'm an AI fan but I have issues with even the biggest and strongest models for my usecases, which ironically enough is AI development (non-cheating AI in a strategy game, that is). They get there, we are making progress - but with no small amount of fumbling along the way. Testing various ideas and strategies takes time. Qwen 3.8 Flash is below Qwen 3.8 Max, which itself is below Kimi K3, right? And that's below Astra and Fable. And my usecase is nowhere near ASI development.
I agree that AI will improve in cost-efficiency but consumer hardware seems like a stretch.
Anything fast or nimble is going to roar and bellow from datacentre-grade compute if it can whisper on consumer hardware.
I can run Qwen-3.8-Flash (120B) at Q4, K/V cache, on an nVidia 3090 and a Core i5-14400 (albeit with a lot of now-expensive RAM), with simple llama.cpp run, around 5-2 t/s at 100k available 16-bit kv cache. Dropping the kv cache to 50k nearly doubles performance. Qwen-3.8 in general defaults to a heavy thinker, so that's worse than it sounds -- a moderately complex problem can burn 30k tokens -- but it's the sort of thing you can leave crunching on a problem for a while and be happy about the answer.
(Comparisons: Qwen-3.8-27B runs about ten times the speed, and Gemma4-26B runs basically faster than I can read it.)
For intelligence and capabilities, the comparison to frontier stuff is rough. Low-parameter models just don't have some information, and with either hallucinate or just nope out, no matter how well it had to be present in the training data. Indeed there's been some efforts to trim low-value knowledge from public models to optimize them for specific use cases, with weird results.
And home users have some rough spots. Both quantization and abliteration drive perplexity and errors, and the harnesses to find and debug them live aren't well-established in the open source (or free-as-in-beer) world. It's fascinating to read a logic trace that goes into surprising depth, but it doesn't do much if the program output doesn't work. The errors are small and embarrassingly simple for a programmer familiar with common JS errors, or for other models to catch, but non-programmers would likely struggle to explain what was even going wrong.
((Also note: the game's not good or fun, even when it does 'work'. That should be expected given the lack of specificity, lack of agent harness, or even a real iterative process, but it's also something no human would do this way even as the core idea it came up with is kinda clever.))
That said, intelligence can be surprising. If you want a model that can make connections between input tokens or parse through mounds of data, you can get away with stuff much smaller and more energy-efficient than you would expect. I would not, a year ago, have expected you could get spatial reasoning worth spit in a 27B model. A real big curveball isn't the most likely thing, and I wouldn't put a ton of money on specifically Jev doing anything ridiculous, but I wouldn't bet against someone coming out with a two-fold performance or intelligence improvement for inference in this model class before the end of the year, either.
More options
Context Copy link
More options
Context Copy link
How does that work? That gpu has, what, 11 gb vram?
https://prismml.com/news/bonsai-2-27b
It is with small context and 20t/s , but 24/7 are a lot of tokens.
This is a dangerous type of company. They are (presumably) thinking they are engaged in well-doing by democratizing intelligence or something like that. But soon the worst scum of the world will have very powerful and uncensored, non-refusing models running on cheap hardware and probably smartphones too, in not too distant future. This model you've linked can even be run in a browser. No know-how required at all.
Yeah, imagine having a model that does what you tell it, how horrible.
Did you know that this existed in other forms called "books" for the last hundred years? And what was the maximum we actually got in terms of terrorism? Arguably the scariest it got was Aum Shinrikyo, who had a bunch of people intelligent enough, and driven enough, to wage chemical and (if they actually had the materials) nuclear civil war. The secrets behind those mechanisms were at the time widely-published; everyone knew how the Hiroshima device (and sarin gas) worked maybe 20 years after it was used.
The only really dangerous thing I could think of would be biology, and perhaps after the inevitable novel biological attack (which will likely be least-effort) we'll see agar become a controlled substance. But the models aren't ready for that yet and it still requires a researcher with their head on straight, which is selected against when we look at the average (for example) mass shooter or mass car-driver; even Aum's highly sophisticated attack in one of the most crowded places in the world only managed half the death toll of an angry retard attacking an office with a gallon of gas).
Then again, I kind of believe in the American way where "having every tool necessary to destroy your neighbor should they piss you off, and every day of your existence is a choice to deal within the rules" is probably more stable than highly authoritarian regimes simply because the existence of the veto helps ensure those rules remain fair. If the State is otherwise capable of complete totalitarian control, which with the coming tools they will, then individuals under it must have some power to check it or there's no reason, from a realpolitik perspective, not to go full 1984 and enslave them.
Even then, political violence (which the more intelligent types prefer) tends to be targeted; the prototypical example perhaps being an autonomous turret set up in northern Ireland trained on pictures of Oranges, set up as non-lethal simply to make a point, or otherwise depending on how angry they are.
More options
Context Copy link
What do you mean soon? You can already run completely uncensored, non-refusing models on regular consumer hardware. Huggingface is chock full of heretic uncensored model weights.
You need decent consumer hardware/vram to run the good models. Soon the models will be much smaller, and thus much more accessible to poors.
Even I, with a 7 years old second hand laptop can run surprisingly decent models. Any gamer with midrange GPU can run Qwen 35B A3B models at a decent speed (~20t/s or better) without having to have any particular technical skill beyond knowing the magic word "heretic" and doing a huggingface search for it. Anyone with a 24 GB GPU can run uncensored Qwen 3.8 even without any "dangerous company".
Anyone "dangerous" already has full access to those models right now if they only so desire.
Oh god. Pandora's Box is already open. hides under bed
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
I already like them. No need to sell them to me harder.
Oh. Cool. How will you and yours prevail in the Cyberpunk 2077 world of tomorrow?
As a corpo I'll do well. Be overworked and scared of terrorists perhaps, but materialy wealthy.
More options
Context Copy link
By being smarter, more ruthless and more inventive. What your ilk don't understand is that Cyberpunk 2077 is the good future awaiting humanity. Leaving AI under government control means 1984 ... at best.
I don't even disagree, but you are aware of that whole the-AIs-are-about-to-break-through-the-blackwall thing happening in the background of C2077?
Yes. But I am more worried about UK government sending everyone with anti immigration sentiments using AI powered robots and surveillance tech to reeducation camps, than skynet using terminators to eradicate us all.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link