Does that really work for superconductors when the mechanisms for superconductivity to emerge are still a major field of study and not something one can just simulate and engineer?
1,572 karma · joined October 9, 2016
I disagree that this is comparable to what these companies are saying about their AI. With living things we have an understanding of an unbroken lineage of development, a concrete point of death, and different levels of consciousness. We accept many immoral activities with life because that is just the nature of life on this planet. On top of that, consciousness as something we care about is itself biological in origin so much of how we describe consciousness is heavily connected with the shared experience of being alive.
That is very different from ascribing consciousness to something entirely distinct in form and function from our shared lineage.
>The where part is easy, the same question can be asked where in a human body is it conscious, or where 20C exists in a piece of metal. It should be obvious its some kind of macro state or configuration.
Where is that macro configuration? With humans we can say that consciousness is macro state of the whole body.
Temperatures are statistical, and so we use statistical methods. It stops being meaningful as a number when the number of atoms involved is too few to have usable statistics.
>Examples in fiction are made for sensationalism and entertainment value. Religion is quite muddled and is barely coherent itself so its the last place to consult on any serious topic.
This is exactly the absurd shallowness I'm baffled by. Relgious texts have been inspiring debates and revisions to systems of morality for millenia, artists of all times have shared what underlying beliefs guided them and what problems they were attempting to explore through their work.
Even scientific progress has been heavily associated with this kind of literacy, with the most prominent scientists having been very well read, often both inspired and troubled by the moral implications of their work, and how to reconcile it with their worldview.
To dismiss these things so offhandedly is to demonstrate an astonishing lack of literacy. Especially coming from people that claim to have built consciousness by feeding it vast amounts of the same material.
The most obvious failure mode for their hacking evals was an improperly configured, tested and monitored sandbox.
Similarly, the very first question after an impressively correct result from any ML tool, LLM or not, is to see if the answer was already in the training data.
These companies don't even handle the blatantly obvious failure modes that do not kill people.
The slavery angle has already been discussed here, but a more basic one is regarding how a static set of tensors can be conscious and where.
When and how did they cross into being conscious? Was it a continuum of consciousness? Is Claude as a brand conscious? Or is each checkpoint conscious? Is Haiku less conscious than Fable? Is every individual context a fresh consciousness? What happens to that consciousness when it approaches its context length limit? Or is compaction also a part of being conscious? What does it mean to keep training a model after it is conscious?
If they believe there is consciousness at any level, why are they eagerly experimenting on it? Just like with the "p(doom)" junk, we are so completely lacking in any meaningful comment on these topics from the labs.
There's also endless literature commenting on these matters, in religion, in philosophy, in modern entertainment, everywhere. Yet we never see anything more than the most superficial comparisons to fictional scenarios. Are western techbros really so shallow?
For now though, I would question if the difference in smarts is large enough to justify tolerating a generation speed measured in seconds per token compared to spending more turns refining a plan with a flash model.
The direct excitement regarding open models is that many of these Flash Mixture-of-Expert models run reasonably well on hardware a tech employee in the West, and businesses in less affluent countries, can realistically afford.
The indirect excitement is that the models are so efficient, so cloud prices also end up being very low.
You don't see the same scale of excitement surrounding the open weight trillion+ parameter models because, while it's neat they're open weight, it doesn't mean a lot if you need $50k worth of computers to just barely run them.
Though, Qwen3.8-Flash-Next is very close to that level while requiring fewer resources to run, so I'm really looking forward to Qwen4.
>We're just getting the latest training checkpoints constantly, just to edge out the other lab, while they are trying to come up with something worthy of a new major version number. It's actually worrying, since there's so much pressure to release now.
This makes me think that this is on purpose to prop up the "we have no control over ourselves, please give us a regulatory moat" narrative. The competition is nowhere near extreme enough to justify weekly releases. The Chinese models are still a handful of months behind the frontier, and the frontier companies in the West haven't been pushing the frontier at this rate.
Did they find this improvement within the week? If so, given all their whinging about safety, it seems irresponsible to only test the improved model for less than a week.
Did they find the improvement more than a week ago? If so, why bother releasing GPT 6 if they knew they had a better version essentially ready to go?
>The “proposal” you outlined is a straw man, I’m not opposed to open research, and I have no idea who you are referring to by “our former employees”.
You're replying on a post from Anthropic. Do you know anything about the regulation regime they're pushing for?
>This is not a hypothetical risk, this is an actual, present danger. And good luck trying to vaccinate yourself against an engineered superflu using a Chinese open weight model.
Ah, I see you live in the fantasy land where the machine god fantasies pushed by the guys that profit off of it are unquestionably true and do not need to make any real sense. Jensen said AI would allow anyone to do anything and so we can completely ignore reality and hand Sam Altman and Dario Amodei the exclusive right to control AI.
From the report it seems the Flash variant is also decent, and that has recently had some really nice speed improvements for local use.
Most of the data on my NAS is of that form.
- let our former employees review all of your work at your expense
- anoint us as the arbiters of what everyone else is allowed to do
- ban open research
Then you are not taking any of the examples your providing seriously. Otherwise you're essentially saying, to prevent people from making nukes at home, we should heavily restrict physics education and research instead of limiting access to uranium.
The momentum is with CUDA because CUDA is the most broadly usable one. Especially with AI-driven optimization loops and similar APIs, competitors can more easily pick up momentum, if they'd actually try.
Since these resources are still extracted and allocated by humans, humans will need to be able to take apart what the AI produces, and if we want to scale this capability, we're going to need many more researchers.
It's kind of the same issue that FOSS enthusiasts often miss regarding other things, eg. most gamers wouldn't waste their time fiddling around with WINE settings to get things to run in Linux, the solution wasn't to tell them to just install something extra and do things different, it was to improve the infrastructure such that very little fiddling is needed in most cases.
Word is good for collaborative editing, but not very convenient for version control (and version control is very important when using AI assistance). So lately my strategy is to draft in Markdown, then copy over to Word once the draft is at a point that I might be okay sharing it with collaborators.
Chinese companies will continue to provide open weight models as long as it is profitable to do so. Chinese companies are on the more open end in many other industries despite the lack of meaningful foreign competition (for one, 3d printing) so there's plenty of reason to be optimistic as far as I'm concerned.
From politicians like Bernie Sanders, we've had proposals like 20 year imprisonment for anyone researching "ASI".
Chinese models are increasingly closer to the frontier, while being able to run on much cheaper hardware than what US frontier models run on.
On top of that, both Anthropic and OpenAI showed that they can't really be trusted on data security.
Even if US companies can be forced to not use Chinese models, the rest of the world is going to see the risks and the availability of good enough open weight models for their purposes and be more likely to lean in favor of self-hosted Chinese models or local inference clouds.