sounds impressive. Now i want to see similarly impressive story of debugging and troubleshooting there - which most probably is possible and would involve Gemini doing that troubleshooting, and that would be great.
8,597 karma · joined February 14, 2014
“Project Panama is our effort to destructively scan all the books in the world,” an internal planning document said. “We don’t want it to be known that we are working on this.” ..."
https://encrypted-tbn0.gstatic.com/images?q=tbn:ANd9GcQqEY5M...
some get very creative https://kvartira.mirtesen.ru/blog/43698892000/15-sumasshedsh... , https://masterok.livejournal.com/7601098.html?
Anyway, that demonstrates another way of datacenters getting modularized. The point is that the actual volume of datacenters to be produced calls for the mass production assembly line style. The one-off custom built approach becomes unsustainable.
In such a module-based construction you can optimize layout and the fabrication sequence so that the labor is minimized and simplified - like for example you'd be simply laying down wire instead of running it.
Thus the modules can be fabricated where the labor is cheap and plentiful and with high degree of robots usage while only final assembly would require high cost labor at the target location.
For an exec to talk about therapy - that is allowed only to the top dogs, like CEO, who can pass any personal crap upon his reports.
Now i'm starting to get scared - these cash positions are basically gone if matched against the new debt and various creative financing taken on for the AI buildouts. That looks a lot like 2000 - very promising tech everybody is piling money in. And i'm sure that several years later it will provide several decades of tremendous success like Internet did in the last 20 years. It is just those few initial choppy years of the hockey stick trough that we may be coming upon and that many may not survive not having that fat cash position anymore, and thus those years may happen to be very painful for the tech and for the whole economy.
25 years ago there were a lot (estimates put even as high as 95%) of dark fiber left as a result of the dotcom buildout and crash. It was successfully put back into action several years later. I wonder whether we will have similar dark datacenters in a few years.
questions of a layman - couldn't they initially do a small infusion of the [may be even weakened version of that] viruses to check for the immune reaction? May be such infusion would really serve as a vaccine prepping the immune system for the main infusion later? - though immune system killing the viruses may be nullifying the treatment - then may be when doing such therapy the patient needs to be [somewhat] immunosuppressed similar to transplant situations?
International distillers doesn't use that premium content, so they don't pay for it. They do pay for their access to the models they are distilling. Thus providing the revenue stream to those models. Thus those models make profit off the content they used for training. The content they mostly have't paid for.
>It's a cost that American open models will seem to have to pay but not international.
It goes both ways - American companies and their business are protected by American laws and have access to the market protected by those laws, etc.
Also note that the OpenAI/Anthropic argument is that the model training is sufficiently transformative to satisfy the fair use of the original content for training.
By that same argument, when distilling the distillers aren't using the original content the OpenAI/Anthropic models were trained on - the distillers are interacting only with the "sufficiently transformed" content of the OpenAI/Anthropic models and are normally paying for that.
There is also that old phonebook rule that facts can't be copyrighted. So, if i asked the model about bunch of phone numbers, i can publish the resulting list, can train my model on it, etc. Such approach doesn't allow to reproduce copyrighted works of course - and as we know the AI output isn't copyrightable, so it looks like basically any output i get i can use whatever way i like.
The same principle can be applied to distillation - it is a fair use. You just shouldn't use illegal ways to access the models being distilled.
To the commenter below: if it is illegal - has the police/FBI report been made? Otherwise it is just a civil court matter.
the same argument - a level of creativity in the world knowledge creation that ins't present in the model training on that knowledge.
Or in other words - model creation and training is just a distilling of the world knowledge.
So, new excuse seems to be emerging - "it was an AI". One can imagine a law enforcement questioning the AI to find out whether the AI did it accidentally on its own or was specifically prompted by some human to commit the crime.
"Fragment on Machines":
"he explores how human knowledge and collective intellect become embedded into machines, divorcing the worker from their own creativity."
"General Intellect":
"These texts are widely discussed for his concept of the General Intellect—the idea that society's shared, collective knowledge increasingly drives production rather than raw manual labor, and that this knowledge is alienated from workers and used as an instrument of capital."
(note: my point isn't to pass any political judgement here, like what real communism in China or not real, is it good or bad, i just find it interesting that pure political discussions by people with no technical credentials bring AI as a major factor today)
human brain has 100T synaptic connections. We already have 2 trillion parameter AI models. Parameter count grows more 3x per year. It means that in 4 years we'll have 100T model. My point here is that by the time we have probes to send even to the closest star, the biological brain would be the inferior option, and most probably we'd even be able to upload, just as fun small companions-observers, a bunch of people brains into what by the time would be a superintelligent 100000T+ parameter AI brain of the probe.
What we have here is a buildout of the due process violation system not a due process implementation system.
While it haven't been built yet, nothing seems to prevent ion drive even with 100x ISP of chemical rocket. That means we can get 1000-2000 km/s (acceleration with existing reactors would take about 100 years) and get to the closest star in 1000 years.
more so than not. We are highly adaptable, and that is our strength as a species. We are that adaptable because of our brain, and we have such brain as a result of adapting again and again. We should pay more attention to how to adapt to new tech at individual and social levels, and that adaptation would in turn again advance us. Whereis strict-prohibition-like-measures really play against our adaptability as they favor slow-to-adapt traits of our species, again at individual and social level.
There are of course cases where our adaptation is very limited - like for example highly radioactive environments, and so we chose to strictly regulate nuclear tech.
Practically speaking i'm all for government funded job retraining where it is possible, and for strong safety net to soften social impact of new tech. And in particular nothing prevents the government to charge a modest tax on each token and use those collected taxes to support the affected people.
Basically the Homo Sapiens history is the history of making and using tools. That is for example how we got the hands we've got, and our social organization was shaped by agriculture and later by the mass production and now more and more by information technology with AI becoming the major part of it.
>I am also afraid that it will create a worse future than the one that we had until now.
were hunters-gatherers better of worse than agricultural village dwellers? There is no clear answer. I feel that each stage of the progress made our lives better while some people want "back to the caves" though i think they probably never spent a night under open skies.
>I think deployment should slow down and research should be financed and diversified.
You can't slow down it in all countries at the same time. And thus slowing down it in any given country would just put that country behind.
I don't see it that way. I see it as the routine and toil that we perceived as everyday creativity and professional work is getting compressed out like repetitive values and patterns out of a data set - that isn't surprising giving that the core nature of AI training is encouraging pattern recognition/generalization and producing internally de-facto a compressor and a decompressor as a result.
A human may have seen 10 instances of something with minor variations - and would think that is creative. AI after seeing it 1000000 times and generalizing accordingly - this is a recipe for producing a 1000 more with desired variations according to the specification.
That naturally doesn't end creativity. It just raises the perceived bar, which really has always been there, for what is real creativity.