I mean, this is a cool demo, but that's about it. It shows that you can build a model that holds secrets which are genuinely encrypted until the model gets the correct input. But it's also trivially detectable, so the framing seems a bit misleading.
This is a deflationary comment. You make a good point that the weights today are detectable, but I think proofs of concept around poison pills are very important. I am unaware of any theoretical reason why an attacker or misaligned model could not one day find a way to disguise a poison pill such that it is indistinguishable from the normal weights. There could be such limits, but research in this direction is very useful!
If (and that's a bit of an ask to be honest) I understand this correctly, the weights will look very different from what trained weights look like. Very sparse, not uniformly random.
This is true in the immediate case, but unfortunately there are relatively simple methods to obfuscate weight distributions; see "Obfuscating compiled weights" on page 13, or Figure 2 here: https://arxiv.org/pdf/2407.14494
This is such a bad video. Someone else's video, cut over multiple different times, a horrible slurred AI voice, shaky filmed screen with a guy talking over… Each layer adds nothing, but somehow this is the version worthy of sharing?
It's a Twitter link to a post of an NFT guy, my expectations were near zero before the video started. I'm disappointed he didn't even bother to hide that he was reading the output of a chatbot, though. At least the video didn't end with "buy #ingredientsplitcoin".
I'd post the link to the source but of course all attribution has been scratched off the video and the first level of overlays.
Agreed, I googled "Ingredient Splitting" to find a better description and interestingly got pages of pet food related content - it seems a bigger issue with dogs than people
With animals, particularly dogs, you got a very vocal minority of owners going hard on the "feed them naturally" idea - in Germany we call them "BARF" for "biologisch artgerechtes rohes Futter" or "biologically correct raw food". And oh boy. These people are factually correct in everything they say, but have an energy and persistence level that would put hardcore Evangelical proselytizers to shame.
"These people are factually correct in everything they say"
Are they? I don't know the German movement, but the talking points I hear coming from US BARF people seem to mostly be a mix of FUD, naturalistic fallacy, anti-vax, and misunderstanding of biology.
Your opinion is wrong in my opinion. BARF is very much practised in the US as feeding whole raw unprocessed food. Like, chicken (with feathers, poop, stomach contents and everything that came with the bird), lamb (again everything that came with the lamb), fish (everything that…… you get the idea). And that really keeps the dogs healthy, no teeth problems, no cancer, nothing. I have a 11 year old Tervuren that outruns and outperforms 3-4 year old working Malinois.
Right, the "this diet cures cancer!" claims are probably why the parent comment has their opinion.
Both are likely correct. Highly processed foods are likely as bad for dogs as they are for humans, AND the people who lean hard into those claims tend to be nutters who are coincidentally correct about this alongside the antivax, anti seed oil, naturalistic bullshit that they use as virtue signaling.
The phrasing makes it sound like you are using them through a chat interface. Have you tried something like Claude Code? The real value only starts materialising once it has sufficient access to your environment.
It's a function of scarcity. Generation (or "writing" as it used to be known back in the day) is now cheap, so judgement and taste are the new bottleneck and therefore the difference between slop and effort.
It’s overhyped but for a lot of people like me it was pretty fresh and different. Strong banter dialogue too, got a lot of praise for people “talking like people.”
Stargate is just one of those rare shows that (apart from, uh, certain episodes) aged incredibly well. I really miss optimistic television where people are competent at their jobs.
Even the whole "science person solves the problem under pressure after getting an insight" thing suddenly seems a lot more realistic now that LLMs have actually become good.
The real story here is that people are selling modded Pis as higher-memory versions. And this check actually caught them + is circumventable. The restriction is somewhat annoying to a small subset of enthusiasts but seems to catch genuine fraud.
The memory that they are placing in it has to be cheap enough to be worth their while. Which in this climate is highly probably faulty memory, or memory that runs at a different speed/spec to the one advertised to come with the RPi. That is fraud.
It's unfortunate, but given that there are so many ways you could end up with a poor quality knock-off (i.e. Amazon, sorry, 'third party marketplaces') it's probably something that skews the tradeoff quite significantly.
I guess there are more individuals who will be upset when they can't run something on their equipment as it was advertised to them than will be upset that they can't re-solder the RAM.
There are probably even fewer individuals who'll do due diligence in their own supply chain. Whilst those that suffer are technically suffering from some kind of crime / civil wrong - the chance of them getting a remedy for this is going to be non-existent.
Given the how overpriced the high RAM models sold by the RPi foundation are, I think they just can use good RAM and still make a nice profit. Like you know RPi foundation could, if they were not greedy.
> The memory that they are placing in it has to be cheap enough to be worth their while.
You're framing this as if no reputable vendor is willing or able to source cheaper parts.
If they can get a better deal in more RAM, why can't they upgrade the boards? As a consumer, doesn't that benefit you as well?
> Which in this climate is highly probably faulty memory, (...)
That's specious reasoning and blatant fear-mongering. Consumer protection laws require vendors to support their products. If RAM is faulty, whether it was hand soldered by Eben Upton himself or hacked from a Soviet fridge, vendors are compelled to support customers. This means vendors ultimately have an interest in minimizing faulty equipment, because it hurts their bottom line.
So why can't people have the right to upgrade their hardware? I thought we already went through this battle in the 1980s.
This is classic gaslighting technique corporations use to extract more profit. Modifications or upgrades made by small business = fraud, OEM selling different SKUs = legitimate business.
If a small business is modifying the product and not telling the eventual customer they've done so how is that not fraud? I don't want to buy a Raspberry Pi with a memory chip from who knows who which might have not passed QA modified by people who may or may not do a decent job at soldering it to the board.
A unit that rpi themselves sold with 8g is different than a unit that someone else modified. The reworked unit is not sold with the disclosure that it has been hacked and no longer covered by any warranty, and the memory installed is of totally unknown provenance and quality, timing specs, failed qa, etc. Even the simple process of heating everything up to desolder & resolder temps is not free, it degrades everything, the pcb itself, the epoxy holding the traces and pads down, and all the components, and means higher chance of some failure happening sooner than it would have. But it's sold as a real rpi unit for real rpi price or a little less to beat real ones.
That is the fraud they speak of.
You should be free to do this to your unit if you like of course, and rpi has infinite other possible ways they could have chosen to address the problem and didn't have to do it this way. I'm only saying there is actually a problem.
I certainly don’t say this with the implication that I would do what RPi did if I was in their shoes or that I agree with it, but the answer to your question is “you probably can upgrade it yourself but you just haven’t figured out how to.”
Just like the idea that you “can’t” rip a Blu-ray Disc was only a matter of time and energy to thwart.
I also don’t think this device is one that has ever been sold with the expectation that it would be upgraded. It is (or was) a cheap single board computer.
Like it’s kind of funny to me how this is becoming a controversy among people who probably didn’t even think raspberry pi memory could be replaced in any semblance of an economically sensible way as of yesterday.
The same argument has been used against most data center operations from the beginning. You can criticise specific bad actors and bad practices (like Musk and his blatantly illegal gas turbines) without morally condemning an entire field of technology through the lens of religious environmentalism. There is nothing about the technology that prevents data centers from running 100% renewably.
There should be no plan of building new data centers up to that time. I am not against new technology at all, only against indiscriminate use.
It's like cars and motorbikes. People are illuded that they drive good tech, but they are strongly contributing to pollution. This is a topic that should join all people.
reply