I found a Chromebook, well “chrome top” desktop did this well enough. It’s not technically immutable but it’s hard enough to install and break things that it worked well with my grandmother.
It’s partly this but also the artists benefit in other ways from large strata of their fan base being able to attend live shows, but merch in person and mingle with each other. The venue and Ticketmaster only benefit from the ticket sales.
Scalpers are also a source of guaranteed sales, so you're diluting the risk of a concert because someone is already buying up all seats already for you and running the risk of not being able to sell them at a higher price later.
Anecdotal but the perception in my corner around the time of the license switch was that Postgres was good enough now with bjson columns, and we could just use that for new projects. Projects that needed to scale beyond what is easy with Postgres are fairly rare and usually mean you’re doing well enough to throw money at citus or whoever to prevent everything from melting.
Infrastructure like that is a natural monopoly - I certainly would have an easier time switching from Google than I would Comcast, the latter would require me to move or put up with 90's era internet over copper lines.
I can't easily type that out - and once the format can't be read / editing in a simple text editor, I'm starting to lean towards a nice binary format like protobuf.
I think maybe the point is rather than faces and voices can be indexed like text now too. For a brief time this was probably computationally infeasible and so it offered privacy, but now a government or company could theoretically automatically analyze all video chats.
PGP+email still makes indexing this content computationally infeasible though.
Let me know what platform you use for encrypted video calls that isn’t based on trusting a third party with the private keys.
That is usually the requirement to get a UX people will use with their non technical family, but people accepted it, I think in part because it didn’t seem practical to index it all. It definitely is practical now though.
Not the GP, but I'd at least imagine that a self-hosted XMPP could get the job done. You'd probably also need your own STUN and TURN servers, but it should definitely be doable.
Hell, might even be doable with public XMPP and OMEMO, but I don't know if that works with the video call stuff.
It's sadly a lot of effort, but I feel that for stuff like family privacy stuff, it might just be worth it.
I self host prosody (an XMPP server) with a TURN server in my living room. With Guix System this is largely declarative, aside from the DNS records and initial user account registration (which I do on the command line).
We're using it for messaging, sharing files, video calls across continents, etc. Re video calls: the XMPP and TURN servers are only used for negotiating the connection, so you don't need a very powerful machine for any of this.
I can’t speak for Israeli tech, but the pentagon has an image problem in the valley, I don’t believe they are getting the best recruits even for contracting companies like Palintir. Our generation is closer to Iraq and Vietnam than WW2, and many of the bright minds are first generation immigrants. Despite the more recent image problems ad tech has (now that people are seeing more of how the sausage is made), it’s still sexier to work on big consumer companies than defense. You’d have to pay my colleagues more to work for the US government, even indirectly, instead it’s often less (and often with less freedoms of what they do off the clock).
And now, what I’m reading is that if you do go contract for the military in AI, your function is partially some kind of scapegoat insurance. Blame those eggheads with their computers who can be fooled, not the fools who hired them and acted on that signal above others I guess?
The idea that a chatGPT model would have been a deciding factor in preventing 10/7 is laughable on its face to anyone who works in the industry, except maybe a consultant selling LLMs to the IDF.
That image problem goes away when you want to close a 7-8 figure TCV Fed deal to make your quarterly sales KPI.
The bigger stumbling block is procurement.
Software Procurement by Federal standards is relatively straightforward so a Series E+ startup can make it if they spend around $7-10M and 1-1.5 years on a dedicated roadmap for FedRamp and FIPS compliance.
Once you step out of software, procurement becomes paperwork hell. Throw in the paperwork hell from R&D Grantmakers like the DoD and DoE, and you end up with a quasi-Soviet procurement system.
Ironically, most of these compliance and regulatory checks were added for good intentions - primarily to minimize corruption and graft, yet it basically clogged up the entire system, and dissuades startups and innovators from working directly with the Defense community.
Some projects like DIUx and and In-Q-Tel are trying to change that, but it's too little too late, and our defense base is entirely dependent on firms like Microsoft, Cisco, Crowdstrike, Zscaler, etc acquiring promising startups to evangelize their innovations internally.
> Software Procurement by Federal standards is relatively straightforward
> FedRamp and FIPS compliance
It’s odd to see these in the same sentence. FedRAMP is so insanely complex/difficult to achieve in a straightforward way. Even by your own estimate for a series E startup (with lots of capital and the ability to spend >18 months< on compliance) there’s a 3M$ variation in cost.
That rules out every startup or SME in software and that’s why you have Palantir, half baked tech that rarely delivers/is somehow more universally hated in USG than ServiceNow. Yet able to seize the space and hike prices endlessly due to compliance being so difficult to achieve — they realize/accept this as their edge as well and it’s why they so aggressively pursued IL6.
The good news is that this is going away and USG is strongly reconsidering its approach here. CMMC, imo, is a huge step in the right direction.
At the end of the day, most work done by technical teams within Defense Agencies is implementation, and the R&D related work is done by specific vendors or very autonomous labs (either National Labs or a specific PI at a University)
This is how it works at the Fed just like any other corporation, as well as with any other peer country.
While there are internal R&D projects, most agencies aren't having their engineers design and productionize bespoke environments from scratch - they're implementing existing tooling and buying it off the shelf.
For example, if you want an internal cloud platform, you'll just use Azure GovCloud. If you want to spin up a K8s cluster, you'll spin up an AKS cluster. Want to protect your cluster? You'll just purchase an off the shelf CNAPP.
For defense, R&D is important, but that isn't the DoD's forte and distracts from it's core mission, which is why they offload innovation to the private sector. Even the USSR did this to a certain extent by the 1970s by supporting defense corporations like Mikoyan and Sukhoi that basically operated as state owned corporations that competed with each other.
The issue is the amount of suppliers in the US has shrunk dramatically since the 1990s due to the compliance overhead and requirements such as a single platform DoD wide (a major reason for F35 cost overruns).
On top of that, any fundamental research requires a significant amount of paperwork to justify funding and sets limits on salaries for PIs and Postdocs that are significantly lower than market rate.
Basically, American private industry has largely been divorced from the MIC, and aside from a handful of major enterprises, there isn't an incentive to enter the procurement space. We've accidentally remade the entire 70s-80s Soviet procurement system in the US today.
There are some changes happening in Software and Satellite procurement, but not as much in other sectors like Avionics.
They aren't a significant portion of the DoD R&D infra.
Most FFRDCs and UARCs are staffed by civilians employed concurrently with a regional University or Industry Vendor, and these labs in turn are PPPs often operated by a private sector firm like Lockheed or a university like UCB.
On top of that, the bulk of the budget goes to funding research done outside of FFRDCs and UARCs via programs like DARPA, grants from the DCTO S&T, SBIR/STTR, etc
This fusion of university research, private sector research, and some limited in-house research is what's called Civil-Military Fusion.
The issue is the private sector portion has increasingly been divorced from the private sector, as up and coming private sector opportunities or promising startups don't have an easy on-ramp into the existing defense procurement or research infrastructure, and grantwriting+compliance overheads plus limited grant funding dissuade most companies aside from your Charles River Analytics types from going thru the hurdles.
That has nothing to do with it. You responded to a comment saying "X makes it hard for the DoD to hire people" by saying "X doesn't affect procurement". If you actually realized they were talking about hiring, what you should have said is "they never have to hire anyone so the difficulty with hiring you are talking about is not relevant".
The Pentagon has more image problems than being a difficult customer to work with.
The "mission" they tout as being the main driver to work for then is often ill-defined and what is best known typically has an atrocious public image problem surrounding it.
There are people in the Valley who will work for less money if it's for a cause they believe in.
The Pentagon's work? It isn't a cause they believe in. In-fact many see it as a more noble cause to thwart all military actors - our own included.
Most research is in some way shape or form is funded by the military. The era of having some general commanding eggheads is long gone.
Since the 1970s, it's almost all outsourced to the private sector via PPPs because private sector players can deploy capital and execute much quicker than the DoD or DoE who have regulatory requirements and need to have specific line items defined for them within budgets.
Also, I think you underestimate or don't realize how much DoD related work is done in the Valley today. I'd estimate that 30-40% of startups in the Bay Area are in some way funded by the DoD - either via Federal sales or via strategic investments via In-Q-Tel or their private sector counterparts.
On top of that, most STEM research grants at Bay Area universities come from either the DoD, DoE, or DHHS.
This public private partnership model is what China copied, which is unsurprising, as most of their middle level leadership and policymakers attended these programs and benefited from the US-China Science and Technology Agreement which evangelized the American PPP R&D model in China since the 1970s.
The difference is, the Chinese system is much more lax about compliance and graft, which allows for it to be much nimbler. The downside is graft can be MASSIVE, such as the corruption scandals surrounding China's Big Fund and fiascos such as the collapse of Tsinghua Unigroup
There's another issue here as well, which is that many of the tech folks who would be ok working for the government, even at reduced rates, cant get through the hiring morass that uncle sam puts up. The fed gov simply isnt set up to quickly acquired talent from industry. They also remain remarkably hidebound by old rules like requiring advanced degrees for senior positions.
For example, Naval Undersea Warfare Centers, Division Newport, had a job fair a few weeks ago. IIUC a number of attendees were given offers very soon after.
But NUWC is a DoD DEMO organization, so maybe it's easier for them than some other parts of the DoD.
And salary definitely is an issue. Even with the Boston pay scale, I think they have a hard salary cap for most software positions at about $150k + very small annual bonuses.
how many of those hires already had clearances and/or military experience?
you've got an active TS/SCI and we'll get you onboarded next week.
and if you don't... it'll be at least 6 months. and that's assuming people aren't too upset about ties to China, a polyamorous lifestyle, or how much weed you smoked.
FAANGs did a lot of stupid interview BS, whiteboarding and leet-code nonsense, but I got an offer letter a couple weeks after, or a rejection, and a start date a month later.
> how many of those hires already had clearances and/or military experience?
>
> you've got an active TS/SCI and we'll get you onboarded next week.
Defense contractors often want candidates to have an active clearance, but AFAIK that's not at all a requirement for DoD labs.
I'm guessing the contractors want to avoid the financial cost and scheduling uncertainty of applying for the clearance. Especially because the clearance follows a person when they change employers.
> and if you don't... it'll be at least 6 months.
I'm not sure where you got that information, but it doesn't match my experience. You get an interim (non-TS) clearance very quickly, and a permanent clearance eventually.
> and that's assuming people aren't too upset about ties to China, a polyamorous lifestyle, or how much weed you smoked.
I have no idea what exact criteria OPM uses for denying a clearance application.
But last I knew, DoD does do random drug testing. I'm not sure what the consequences are for failing a marijuana test, but it wouldn't shock me if it causes loss of clearance.
Often, no. This is serious work being carried out by adults that need to come together. There is no replacement for the water cooler yet. I made the decision to explicitly seek out in-office, on-location defense work. The seridiputous conversations and relationship building was not happening in remote work. I'm someone who has always worked from home and I still do every week but my career and life were going no where typing at people through Slack and building meaningless web apps--despite making enough money to be reticent to tell most people my earnings level.
Now I'm building software, involved intimately with designing and interfacing with specialized hardware, and travelling to interesting places doing interesting things with interesting people-- occasionally toppling off of combat machines. I took a 30% pay cut to do it. No regrets whatsoever, living life.
Most Top Secret work occurs in a SKIF. Basically you enter, lock your phone, smartwatch, and whatever else in a locker, then enter the area where the work gets done. This area is regularly swept for bugs and whatnot.
You can't work on "top secret" stuff on your own due to OpSec.
It's very real. Having smoked or taken other illicit drugs in the recent, or not so recent, past is a major source of stress for people applying for clearance. You have to be sponsored at a significant expense by a current employer and if you don't get clearance your career is going to be upended. It's up to the worker to judge if they pre-qualify based on opaque information and anecdotes you find on Reddit.
I'd venture to guess that more tech workers lack citizenship than lack the ability to pass a drug screen. More importantly, the problem you describe is problems with the opacity and risk of failure for a clearance: not "fuckin' druggies", which is what I responded to.
> Having smoked or taken other illicit drugs in the recent, or not so recent, past is a major source of stress for people applying for clearance.
If you have broken the law in the past, the clearance processes mostly seem to care that a) you acknowledge it and have stopped b) you are upfront about it, and it can't be used as leverage against you.
If you're currently routinely breaking the law, yes, it's going to be hard to get clearance. That seems pretty reasonable to me.
generally they don't take weed to seriously, but want to know you've been drug free for roughly a year.
By comparison, several / most of the Cali tech firms I've worked for / with / around had devs hitting a THC vape at lunch. Might have had to pass a piss test to get the job, but that's just 30 days, and no one is knocking on your neighbor's doors to verify your drug and employment history.
I am still not 100% convinced they didnt just let it happen on purpose (and then were surprised just by the scale), having an excuse to raze the place down for good, which is exactly what they are doing. The signs were there, everywhere, and mosad aint bunch of clueless paper pushers.
The guy in charge is former spec ops, murder of anybody without battling an eye is part of the deal so dont expect some humanism from that direction.
If I didnt read similar stories from other times and places, where it played almost exactly like this... AI is not going to solve political issues, just make them more complex than they already are
I believe this is similar to the Latent Consistency Modeling approach, where it’s a replacement for the “diffusion” process, not the underlying weights. Basically, they have a more efficient process for pulling images out of the weights, not necessarily a set of new weights.
The weights are different, because the model is different.
As jzbontar below mentions, the crucial point is that the random noise mask is the same. The diffusion models are trained to turn random noise to an image, and they are deterministic at that - the same noise leads to the same image.
What the authors did here was to find a smart way of training a new model able to "simulate" in a single step what diffusion achieves in many; to do so, they took many triplets of (prompt, noise, image) generated starting from random noise and a (fixed) pretrained stable diffusion checkpoint. The model is trained to replicate the results.
So, it is surprising that this works at all at creating meaningful images, but it would be _really_ surprising (i.e. probably impossible) if it generated meaningful images which were seriously different from the ones it was pretrained with!
Oh the images and prompts we see in article are from the training data?
Pardon my ignorance ...
Does MIT model then not work as a general text-to-image model to generate novel images based on arbitrary new text prompts that it has not seen before?
Nothing to pardon, asking questions is always the right thing to do :-) I also didn't look into the paper in great details, although I'm quite sure I am not fooling myself, but still take this with a grain of salt.
My understanding is that this paper by MIT doesn't train any new model from scratch. I takes a pretrained model (e.g. StableDiffusion), which however is trained to do "a small step" only: you fix a number of steps (e.g. 1000 in the MIT paper), and ask the model to predict how to "enhance" an image by a certain step (e.g. of size 1/1000); the constants are adjusted so that, if the model is "perfect", you get from pure white noise to an image in the exact number of steps you set. If I remember correctly how diffusion works, in theory you could set this number to any value, including 1, but in practice you need several hundreds to get a good result, i.e. the original StableDiffusion model is only able to fit a small adjustment.
This new paper shows how to "distil" the original model (in this case, StableDiffusion) into another model. However, unlike typical distillation, which is used to compress a big model into a smaller one, in this case the distilled model is basically the same as the one you start with; but it has been trained with a different objective, namely to transform random noise to the prediction that the original model (StableDiffusion) would make in 1000 steps. To do so, it is trained on a very large amount of triples (text, noise, image). But I don't think you can incorporate into this training procedure other "real" images that are not generated by the model you start with, because you don't have a corresponding noise (abstractly, there is no such concept as "corresponding noise" to a given image, because the relation noise -> image depends on the specific model you start with, and this map is not anywhere near invertible, since not all images can be generated by StableDiffusion, or any other model).
Once the model is trained, you can of course give it a new prompt and, in theory, it should generate something rather similar to what StableDiffusion would generate with the same prompt (hopefully, the example displayed on their web page are not from the training set! Otherwise it would be totally useless). But you should never obtain something "totally different" from what StableDiffusion would give you, so in that sense it's not "general", it is "just" a model that imitates StableDiffusion very well while being much faster. Which is already great of course :-)
reply