How would such a ban even be enforced? This makes no sense?
Google has banned the training of deepfakes in Colab
91–100 of 137 posts
Re: Google has banned the training of deepfakes in Colab
#92How would such a ban even be enforced? This makes no sense?
It probably just gives them the grounds to justify a ban, if necessary.
Re: Google has banned the training of deepfakes in Colab
#93I wonder if we'll end up in a future where gaming GPUs will be restricted from running ML algorithms, to prevent private individuals from running advanced deepfakes/DALLE-10 or whatever.
Graphics is increasingly using ML, so I doubt it. But still, don't give them any ideas :)
Re: Google has banned the training of deepfakes in Colab
#94Earlier quoted context omitted.
They’d need to be very careful in doing so, as they both operate services that are certified and audited to meet US Gov FedRAMP High control standards, and doing that would violate several audited control standards. A breach in that area would cost them potentially billions in losses.
Public clouds are separated from their gov versions, with gov versions using older, modified versions of what everyone else uses. public clouds ARE NOT fedramp compliant
This is not true. Google Cloud ‘s FedRAMP offering uses the same regions, zones and binaries and configurations as the regular public cloud offering.
Source: I work on FedRAMP compliance for GCP Databases.
Re: Google has banned the training of deepfakes in Colab
#95If it's something that you think can be stopped, go ahead and try to stop it. Just know that you will be Open Sourced- even artificial general intelligence, when someday invented, will eventually have its design released in the clear.
Re: Google has banned the training of deepfakes in Colab
#96Earlier quoted context omitted.
Quoted post unavailable.
Yes, that's how society works. If you use a service respectfully you get rewarded with the ability to keep using that service.
Subservient lapdog attitude toward corporations is a modern thing not a generic property of "society."
Re: Google has banned the training of deepfakes in Colab
#97Earlier quoted context omitted.
Is there a good alternative to Colab you can just pay for, that's good for hobbyist individuals? Especially one that gives you shell access? I tried using PaperSpace for a while but it works maybe 1 in 3 times I tried to use it, they (temporarily) banned me for no reason, they send me "itemized" bills with empty line items, and their support was no help.
Disclaimer: I work there. https://www.grid.ai/
Yes, I like them (and I feel like I've seen some other similar services but haven't looked into them closely). Its easy to start and stop interactive instances (easier than AWS imo) for doing stuff in notebooks, and to run training jobs on more powerful GPUs when you need to, because you're only paying for it when you're actually using it.
OVH actually has a (distantly) similar service (much less user friendly than grid and without the whole "grid" feature, but that matters less for hobbyists if you're not doing a lot of parallel runs). I liked the OVH one a lot in principle, but in practice found it too buggy to use properly (and they don't have customer support). For a budget project it could be worth trying.
Re: Google has banned the training of deepfakes in Colab
#98I wonder if we'll end up in a future where gaming GPUs will be restricted from running ML algorithms, to prevent private individuals from running advanced deepfakes/DALLE-10 or whatever.
More likely the high end ML accelerators become so specialized as to diverge from GPUs. Carry on trying to use your gaming GPU for ML, with the knowledge that professionals are using hardware 100x as powerful.
Re: Google has banned the training of deepfakes in Colab
#99I have wondered in the past what kind of monitoring / data harvesting google is doing with colab. I don't use them for anything nontrivial or work related precisely because I worry about google stealing my stuff or that I would be violating agreements with my customers by exposing their data to google. This story reinforces that they are snooping. That aside, I don't think I find this too upsetting. They don't let pe…
Is there a good alternative to Colab you can just pay for, that's good for hobbyist individuals? Especially one that gives you shell access? I tried using PaperSpace for a while but it works maybe 1 in 3 times I tried to use it, they (temporarily) banned me for no reason, they send me "itemized" bills with empty line items, and their support was no help.
https://deepnote.com is exactly this!
Disclaimer: I work for Deepnote
Re: Google has banned the training of deepfakes in Colab
#100The results of GPT-3, Imagen, Github Copilot, etc, have shown me two things: 1. A critical part of AI that was missing no longer is: the model of our world. In natural language processing (more generally anything that interacts with humans) it has been an eternal issue that to understand sentences you must know the content, and therefore must, in theory, know 'everything' there is to our world and culture. As far as…
While I partially agree with the theses, one should note that in absolute terms the compute spent even on the largest models of their kind is still pretty tame, compared to SV software engineer salaries: you could train your own Imagen for ~200000$. This is pretty much kickstarter-tier, and not at the higher end of the distribution. The web-scale data is also available: we already have a dataset superior to one Image…
> you could train your own Imagen for ~200000$
If my experience in machine learning (which admittedly is small scale) is applicable to these large-scale projects, I'd say that for each press-ready model there are probably dozens of previous attempts/versions that didn't make the cut. So I'd always add an order of magnitude on top of the final model cost.
> The big tech exceptionalism is overblown, you don't really need exotic engineering to train an LLM with Jax and a TPU pod, or whatever accelerator you managed to procure.
While I agree with you that the current best models are still within reach of motivated small-medium sized groups outside of big tech, if the scaling hypothesis holds up, the next big step forward is likely even larger and more expensive models. I was skeptical of the scaling hypothesis before, but much less so now.
> petition our governments to train & make available large models under permissive licenses
I would say that if it comes to this we are already firmly in the "centralization and monopolization" territory if we have to ask nation-states for their computational generosity.