898 karma · joined May 24, 2024
meet.hn/city/us-Boston
I don’t really think I was moralizing… just commenting on the funny juxtaposition of the language and the context - or on the comedy of the language specifically when not considering the whole context. I was not saying DARPA should or should not be doing this - though I’ll grant that what I wrote could be read as an implicit criticism, even though it was not my intention.
> I don't see UX experts hand-wringing about the effects of advertising when they're designing their products.
Plenty do. Plenty don’t. Similarly, plenty of machine learning engineers might choose not to work on, say, a predictive algorithm for facial recognition or a product recommender system because they don’t feel like being a part of that system. Some people don’t have that luxury, or don’t care. It’s fine either way, though I of course encourage anyone to do some reflection on the social implications of their engineering projects from time to time. Hamming, who worked on everything from the ABomb to telephones to the foundations of computer programming (and everything in between) strongly recommends this, and I agree. Working on weapons might be necessary, but you still need to reflect and make a conscious decision about it.
> It's not about "feelings" […] It comes from both directions, designing the machine to be manageable and training the human to manage in exceptional circumstances and _knowing the limits_ of both.
Of course, totally understand that. That doesn’t mean we can’t find humor in decontextualizing the language! Or in thinking about how science always must struggle with euphemism for the purposes of concision.
There is something deeply, darkly comedic (depressing?) about the qualitative language here. Primarily the way it simultaneously intersects with modern discourse around wellness, anxiety, and mental health in such a banal manner at the same time as the latent/implicit violence of action (given that the obvious subtext is operating semi-autonomous killing machines).
I’ve had some trickiness trying to get them to work when some of the targets are continuous and some categorical, but regardless just a really cool method… really nailed it on the name imo!
Capital is never going away I don’t think, but that doesn’t mean you have to be resigned to its inexorable subsumption of all productive potentials for value extractions… just means you need to keep finding ways to leverage your own knowledge of its behavior and response modes to make positive change (eg start working on demand forecasting in p2p battery storage networks, or utility scale deployment controls, etc etc).
> Imagine if instead of an LLM the billions of people instead simulated a human brain. Would that human brain experience consciousness? Of course it would, otherwise they're not simulating the whole brain.
However, I don’t really buy “of course it would,” or in another words the materialist premise - maybe yes, maybe no, but I don’t think there’s anything definitive on the matter of materialism in philosophy of mind. as much as I wish I was fully a materialist, I can never fully internalize how sentience can uh emerge from matter… in other words, to some extent I feel that my own sentience is fundamentally incompatible with everything I know about science, which uh sucks, because I definitely don’t believe in dualism!
Impossible to answer.
Btw I mostly think it’s reasonable to think that there might be consciousness, phenomenology etc are possible in silicon, but it’s tricky and unverifiable ofc.
It would be great to chat some time!
I’ll shoot you an email.
The lack of a regular HDMI output is mildly annoying but not really a problem. Audio configuration is sometimes problematic, usually not…
If a client wants to use their own Pis, getting them provisioned with our software isn’t always the easiest if the customer is techno-phobic (though that’s partly on us - RPi usage is relatively infrequent for our customers so we haven’t put the time/energy into docs and into baking an image etc).
I love Pi’s, but they just that extra bit more finicky than BrightSigns which are hyper optimized for our use case and prevalent in our customers’ equipment rooms already.
Having a separate box is just good separation of concerns - you can hook it up to whatever kind of projector you like (one projector might cost as much as $100k!), or you might need to use an analog display device (eg a CRT monitor) which certainly won’t have any USB/SD compatibility, in which case you will need some sort of hardware to convert signals appropriately. The separation of concerns just gives you much more flexibility.
Additionally, as mentioned before, you can network the boxes, which lets you do things like creating multi-channel synchronized video art installations.
Most BrightSigns also have GPIO pins on them, so I’ve even done things where I’ve synchronized kinetic art to the video playback.
You can write entire custom applications for your brightsigns (or plugins to the BrightAuthor configuration engine) - it’s just its own self contained platform, so there are a lot of benefits to having it be agnostic to the display.
Your main “cheap” alternative to a BrightSign is a Raspberry Pi, which is definitely cheaper, but has its own host of issues to deal with.
If you are ever interested in working on some mediasync-related codebases hit me up! We hire devs to do freelance contracts fairly often.
You work on BMS stuff? That’s cool- a little bit outside my domain (I do energy modeling research for buildings) but have been to some fun talks semi-recently about BMs/BAS/telemetry in buildings etc. The whole landscape seems like a real mess there.
FYI that podcast I linked has some interesting discussion about some issues with PTP over NTP- worth listening to for sure.
Here’s a great podcast on the topic which you will surely like!
https://signalsandthreads.com/clock-synchronization/
And a related HN thread in case you missed it:
Separately, it seems like it would be incredibly useful to use your models in various embodied carbon estimation tooling and other decarbonization research streams. Have you thought about partnering with any academic researchers on this? If you are interested, let me know, as I can definitely connect you with a bunch of researchers who would be interested!
Not sure I follow your usage of discontinuous: clearly the load inputs are continuous variables (albeit discretely sampled in time and magnitude, but I don’t see the significance of either of those sampling frequencies, each of which can be chosen to be effectively arbitrarily small). Also don’t understand the relevance of continuity at all since you didn’t reference it in the post I replied to.
> A few null or unknown inputs that you didn’t even consider will break it.
Null handling should be assumed to be a trivial problem here (ie just use the last 7 valid values for the time step of interest), and I’m not sure what “unknown” values would be here, can you give an example? I think it’s safe to assume that the inputs have already been normalized to some meaningful scale as an implementation detail of the load sensors. Even if the telemetry scheme changes because the instrumentation changes or the infrastructure changes, so long as the instrumentation produces values at the same scale and the actuators can still respond to the outputs, the core kernel of sense/compute/actuate can still be left unchanged.
I’m assuming that you are citing the failure to respond to null or unknown conditions as a lack of adaptation capabilities, especially as you are referencing its inability to last 100 years - though good luck finding any piece of software that can last 100 years, especially more robust machine learning algorithms that have to deal with problems like data distribution drift, hardware changes, etc etc - same goes with wetware (besides turtles?). I’m strawmanning you a little bit on the appeal to time though, since I assume your main point was to just emphasize that “it won’t deal with changing conditions/new scenarios well” which I think is somewhat fair. On the other hand, I think it is safe to say that what I described does at least respond to the world and in turn affect it in a kind of cybernetic feedback loop, which includes adapting to the shifting conditions to meet some desired state. So maybe it would be helpful if you could define more precisely what you mean by adaptation? Not trying to be snarky, just trying to genuinely push you to try to make a clear definition of what you mean by adaptation.
In the example I mentioned, the program clearly is taking inputs (load), storing them, querying for previous values that satisfy certain conditions (the value at a certain timestamp for each of the last 7 days), running computation (computing a mean), and operationalizing the result (pre-emptively scaling) to achieve a goal (avoiding having insufficient capacity) that affects the world (whoever interacting with the system will have different experiences because of the actions taken by the system). That seems like it satisfies your criterion to me.
Alternatively, neural networks are also just doing explicitly what they are designed to do… sure there is a larger computational graph with lots of operations, but it’s all deterministic… backprop is not really much different on a procedural level than the simple fitting algorithm that I outlined above, in as much as it is just a specific well-defined algorithm or sequence of steps designed to compute some parameters from data.
What differentiates a “traditional computer program” from a single matmul? How about a bunch of matmuls? How about a bunch of matmuls with some non-linear arithmetic thrown in between them? And if we are adding in sampling from a distribution but set temperature to zero and use random seeds so that it is fully deterministic?
Obviously things like the diffusivity (so conductivity, mass, density etc) of the ground matter a lot, as does the rate of heat exchange at the surface for it to reject (or absorb) heat to the environment.
Moscow rules and George Smiley’s tradecraft are probably the only real security… ha!
> The craziness is in the architecture that provides cryptographic proof to clients that the server they’re connecting to is running an audited binary and running on secure hardware.
I definitely missed this concept when skimming the links before posting my comment - very very cool!
> open source software requires considerably less trust than closed source. Right?
Of course… but at the same time, I think the difference in the degree of trust I am placing in say, Signal’s end to end encryption and Apple’s (claims of) end-to-end encryption is not as large as it might cursorily seem. Would I be more surprised to read in the news that Apple had secretly embedded some back door than I would be reading in the news that malicious actor managed to push some hidden exploit through to Signal in an otherwise innocent PR? I’m genuinely not sure which would surprise me more, or which event would be more probable, so can I really make any claim as to which is more secure, given the current knowledge I have? Obviously I could think more deeply about this, but superficially, both are requiring pretty large amounts of trust from me - which I don’t think is misplaced in either… though I do personally trust something like signal more at the end of the day based on… what, intuition? A gut feeling?
At the end of the day, it ultimately still boils down to trust though, yes? Trust that they are running the data centers the way they say they are, trust that their supply chain is what they say it is, and so on? At the same time, using some open source piece of software also entails a great amount of trust: I’m not going through the source code of Signal myself, and I’m also not checking that an open source locally served model isn’t sending traffic/telemetry etc back to some remote server via whatever software is running the model… rather, I’m placing my trust in the open source community that others have inspected and tested these things. I’m sure all sorts of shady PRs into important open source code bases are made on the reg after all. So that’s not to say that trusting Apple is necessarily more or less wise than trusting open source software from a security standpoint… my point is just that it seems like they are aspiring to a zero trust architecture, but at the end of the day, it does still require trust that they are operating in good faith vis-a-vis what they are representing in the white papers right? To me, it seems like a relatively safe assumption that they are for a variety of reasons, but nonetheless, it is an assumption right?
As always, it’s not the tool, it’s how you use it… and as always, being creative is always hard, and the challenge is (usually) within yourself and not your tools (setting aside accessibility concerns in which tooling obviously becomes more important; and at least a little small dose of mise-en-scene).