"In the new study, led by Abra Jeffers of Massachusetts General Hospital, researchers were able to do two additional analyses: one that looked at cardiovascular disease risk in people who use cannabis but had never used tobacco products and a second one that looked at people who used cannabis but had never used tobacco products or e-cigarettes. Without tobacco use, the higher odds of heart attack and stroke persisted for people who used cannabis. For those without tobacco or e-cigarette use, only the higher odds of stroke remained."
I think this is a wonderful idea and begins to create a foundation of data richness and interoperability for a very exciting new approach to PCB design, as I commented elsewhere in this thread
However I do want to mention that I think it might be necessary to be able to "cross-compile" to visual schematic format, and back. Or perhaps there is an open schematic tool that can be extended?
The issue is that I think electrical schematics are significantly more familiar to EE types, contain more legible information. Instead of reinventing the wheel there, it'd be nice to see a system that can switch back and forth between text and visual schematic.
How are schematics described as files currently? Is there an open standard? Can it be converted to atopile format, and back?
I was thinking the same thing (I'm an EE by day). I think this idea is very cool, but since the dawn of time schematics and PCB layout have always been visual because you are actually building a physical thing. Its easy to see hardware bugs in schematics visually. However, it might be very hard to track it in code, unless the code is one day smart enough to find the bugs for you. You have to hold more context in your head that you cant export to your visual senses when its written as code. You cant see the actual circuit flow.
edit: just wanted to double down on this being very cool though. i dont mean to deflate this project and I'm about to design a pcb for a personal project - I might give this a go for fun. the promise is there and asthe project and feature set grows i can see it being the way forward.
I have found taking a 'functional programming' like approach, where you build up your circuit from small blocks with specific functions that are easy to hold in your mind and then building up by combining those. For example instead of sticking a bunch of resistors and caps in your top level file, you can abstract them into their functions like a filter, resistor divider etc. Very curious to get your thoughts on using the tool!
I agree - also an EE - and I think it is similar to 'programming music' packages like Sonic Pi. If you are used to reading and writing standard music coding will be an odd and difficult change.
Right now we have a comfort for seeing the layout and such physically, and since that is how we manufacture it this is going to be an output anyway, but there is some future world where it is all put together automatically within the requirements we specify and we have an entirely new way of designing circuit boards.
Each component would come with not only a footprint but an array of basic design implementations that would mix and match with others.. autorouting on steroids of sorts.
We do agree. We built an early version of a viewer in the project but later moved away from it because it wasn't good enough to interact with. We might come back to it with something that is more targeted at inspecting only sections of the circuit or provide a block diagram level representation. But we don't think that just outputting a schematic the way they look today is the right solution.
It wasn't so much "pc board by code" so much, but it was more a "pc board by CLI" approach.
And by that what my goal was, was to offer primitives and utility functions that would build the board up over time, but it was to be done incrementally in a Lisp REPL.
My use case scenario, was ancient AutoCAD.
Back in the day, while you could hook up a tablet or other pointing device (I'm talking pre-mouse here) to make AutoCAD drawings, a lot of it was done simply through typing in commands, in AutoLISP.
(lineto 100 100) kind of thing. And the drawing would appear on the screen over time, you'd save the data model, and manipulated with the REPL. If you wanted 10 lines:
(dotimes (i 10) (let ((x (* i 10)) (y 100)) (line x y x (+ y 100)))
(HN does not have a paren matching editor, apologies...)
Where it broke down for me was coming up with a graphic rendition using CL.
If emacs had anything reasonable regarding graphic support (its SVG support is Not Good), I'd have done it there.
Rethinking it, it just occurred to me I could have probably gone a good way using ABCL in a Java GUI shell.
But the key point is that I think using a REPL for building up something like a circuit can actually work, actually be efficient for users, especially if it's extensible (i.e. (defun grid ...) ). Especially a hybrid (like clicking on a line pastes an identifier into the REPL).
My experiences with KiCAD drove me down that mad hole.
You might find this a stumbling point for adoption. My first thought is this could be a good tool for me to use, but how can I send a design to someone else to get feedback on if I’m not going to expect them to also also learn how to use this or how can I get my design manufactured if the tools can’t export out the, normally very visual, layout files.
I do think that is fair. The reason that we don't have a visualizer is not that we don't want one, but rather that the value to effort ratio is quite low (and really because the effort is high, not because of low value). I'm sure we'll get there at some point. But at the moment there are more pressing issues for the people who are ok dealing with no visualization (like typing, language server, equations etc...).
I’d definitely consider not letting some basic visual schematic outputs fall too far down the priority list.
A lot of existing EE is basically just the visual diagrams and a lot of existing EE people will expect the tools to give them the visual diagrams they work with somehow… like even just including the compiled schematic as an SVG in with the compiled outputs of netlist, gerber, etc… something that other software can be used to convert to PDF without extra complexity in your tool and can be used to interoperable with the existing talent pool of EE people who have not just no idea how to use your tool… but know how to read typical style circuit diagrams.
Yeah that’s an interesting aspect. We did implement a viewer very early on. And we then removed it from the project.
What we discovered is that:
- making a visual viewer is a non trivial endeavor. It takes a lot of time but the value add is marginal for an average viewer.
- people tend to spend a lot of time making the viewer look good instead of improving the circuit
We think that in the long run, a viewer could be awesome to inspect what is going on or get a general understanding of the circuit. But it’ll be difficult to justify the time spent on it early on in the project.
Worth adding, we have found much higher returns building tools to solve the problems that a visualizer would normally help you with in other ways. For example ERC checking and modular abstraction of circuits. Working at a 'block diagram' level is much more intuitive than at the net level in my experience.
We also think that there might be better and more interesting ways to view your data, for example maybe you want to just see all the power paths through your circuit, or investigate how a signal travels from your input to your adc through filters, protection etc. Often on big designs these things might be strewn across multiple sheets and a bit hard to follow.
I am definitely curious for feedback from people after using it for a while. I felt the same in the beginning, but after writing a fair bit of it, I rarely find myself wanting it. I could imagine for more casual or new users it could be helpful at least for the transition.
As someone who just taught themself Kicad to make a simple MIDI controller PCB usi an STM32F4, I was totally blown away, coming from software, how "manual" everything was, how abstruse and arcane.
It's quite difficult as a beginner to know that a design is "correct", or perhaps "correct enough", with respect to component placement and EMI.
It seems like even top EE who specialize in board design utilize rules of thumb rather than rely on simulation.
I was also blown away that the state of the art autorouter for traces seems to be from the early 2000's -- no recent AI magic going on here.
Where is my "autoplacer"? It seems like an AI trained on millions of schematic/pcb combos, even just gerber files via image ingestion, ought to be able to generate a pretty decent layout + routing given constraints.
Or perhaps I'm spoiled coming from software and web because it's so much further removed from physics. But it's still the case that there are a ton of modular components with "API's" that should have a templating language, so very much bravo to this project.
Yes, it is like 90% rules of thumb and experience. The reason being simulation is expensive: not just in the software itself, but in the time it takes to learn how to use it, set it up for a particular design, and then run through it and understand if it's actually reflecting reality or not. I've been working on some projects where we really care deeply about noise performance, and even the super-expensive packages which exist (which we can't really afford) don't seem to actually be able to answer the questions we have about the design. Autorouting is similar, autoplacing is basically a disaster here.
I am confident a lot of this is an accessibility and UI issue: if you want to disrupt this you could focus on making simulation and auto-placement tools that are cheap and actually usable, and help inform the rules-of-thumb. But that's a very big investment and mostly orthogonal to making a different way to specify your netlist.
We share similar thoughts about the core of the auto-routing/placement/simulation issues being UX! We're a bit more bullish on code being able to fix that as well.
Noticing that these tools are infrequently even configured because it's simply too much of a pain to do so for every new design - we're hoping the expressive, reuse-focussed code-based-approach means people need to capture these things once and then everyone gets to reuse them.
This is all a while away, but after we're able to get design currently captured in schematics working smoothly we certainly plan to tackle auto-routing/layout as well
For the autolayout part, one thing we realized is that it’s very hard for a computer to do a good job at it when there is a lot of implicit requirements that are not baked into the schematics. We are hoping that by capturing those through code, auto layout can be improved.
On the design check, we aren’t doing a lot there for the moment but the comment above also applies: if you have clear requirements in the code, it becomes easier to test if the solution fits those requirement.
> It seems like even top EE who specialize in board design utilize rules of thumb rather than rely on simulation.
Yup, almost 100% this. Most parts of most boards are pretty trivial, and the design decisions you make don't really matter all that much. Unless you're working at a company the size of Apple, it really doesn't make sense to spend several hours of engineering time to figure out if that $0.001 decoupling capacitor is really needed.
And in the end it's often a lot easier to just build the thing and test if it works. We see the same in programming: it's technically possible to mathematically prove that some programs are correct, but in practice you can get 99% of the way there for all programs in a fraction of the time by just writing a bunch of unit tests.
Simulation is definitely done, but it's limited to the really hard stuff like antenna design or high-speed signalling.
I agree. It may be primarily about integrating the new construction process and a proof of concept for more ambitious projects, but my overall impression is definitely that this is a poor showcase for the technology: it feels sort of embarrassingly academic.
Researchers create many “toy” demos. There is often a progression of stages to a full “proof of concept”.
Where is the downside to an intermediate research artifact that demos their current tech level — Which also happens to be a whimsical functional sculpture, adding interest to a grassy area?
And very successfully creating awareness of their efforts?
Nothing remotely “embarrassing” (or “academic”, in the negative use of that word) here.
Enjoyed this but upon reflection seems odd that they would have a word "meat" and yet be unfamiliar with biological systems / consciousness in animals.
That's what seems odd? It's no more odd than aliens speaking English in the first place.
At the risk of stating the obvious, the language here is not only used to transmit the ideas in the story, but convey feeling. With the use of "meat" as a pejorative, you're supposed to understand the feeling of contempt and almost repulsion that these beings have towards us.
Another great interpretation of the story would be via speciesism/anthropocentrism, we consider ourselves better than other animals, but for the aliens, we would all be "meat".
Nit: KataGo does train on handicap games up to 5 handicap stones. In the KataGo write-up "Accelerating Self-Play Learning in Go" from 2020, Appendix D says that KataGo trains on handicap games with up to 3 handicap stones. Looking at the current version of the KataGo code, nowadays it trains with handicap up to 5.
There is no chance that humans can beat the best Go AI anymore, since the paradigm of AlphaZero (which was trained in the absence of human game records, and beat the version which beat Lee Sedol essentially 100-0)
It is unlikely, also, that a committee of players would be significantly better than a single master, due to lack of coherence -- but that's an interesting idea! I wonder if a committee of the top 100 go players playing a game by vote could beat someone in the top 10 more than 20-0 or something; i doubt it -- it might even go the other way (that the single player would win the series)
I don't think this counts as the real "start of the singularity" because Alphazero was not able to (or capable of) altering its own algorithm, but rather just adjusting its weights.
Something more akin to being in the long march toward general AI.
As a personal note the whole issue of large LLM's capacity for intelligence, beauty, humanity, morality, logic, etc etc was softened in my mind and heart by witnessing with rapt attention this epochal shift in computing.
I had held Go up as a paragon of human brilliance and beauty -- to see that standard fall was a complex process of grief and discovery for me, which I feel has better prepared me for understanding and appreciating the emergence of LLMs
This is a different kind of setup. I'm not sure if the idea of 2-3 super GMs able to consult with one another has been tried but given the estimated rating difference I doubt it would matter. The difference is estimated around 800 points, or the difference between a strong untitled player and Magnus.
Like many other crafts and arts where machines can do better, Go has a deeper role in being transformative for the human learning it -- in the case of Go, developing strategic thinking, being able to make decisions balancing long term and short term gain, uniting reasoning and intuition, an arena for exercising emotional equanimity.
Winning, I think, is secondary to this. It's a useful measure of how one has progressed in that transformation, but I think the lessons and principles from Go that I can apply in guiding my my day-to-day life are more valuable.
This is kind of blowing my mind, but I'm looking for people discussing the proposal here and can't quite find the valence that seems most potentially viral or self-perpetuating:
There seems to be a really exciting incentive to share lots of links, to regularly hit a hotkey to add the current page to a link list, because there is a kind of graph traversal thing that can emerge, akin to recommendation engines spotify or youtube use, whereby your (anonymized) "like history" -- might be interesting to include some metadata, including when it was liked, how you arrived there, whatever -- will connect you with other people who have intersecting likes, and then blow open entire other leaves of search trees you didn't know you wanted.
I have this feeling on the net recently where I feel like starved, it just feels so stale and bland, hard to find actually good content, going back to HN or Twitter or Reddit or whatever, these little linear "feeds" with discussions etc.
I want a feeling of opening-up, branching, discovery, excitement.
I feel like if a bunch of people shared their like histories in a pseudonymous fashion, you could see these fascinating interest clusters emerging and if there was a compelling UI for navigating them, it could really be self-perpetuating and awesome.
Sort of like visualizations of LLM embeddings, showing regional clusters of information domains, but with a navigable, social aspect, and where because it's pseudo-anonymous you don't mind running AI recommendation engines on it for you and others.
Does this exist? Should it? I would love something like this!
This has already been demonstrated with AlphaGo Zero in a limited domain; so perhaps a more generalized version of this kind of approach (no human "expert" data) will tick the box.
"In the new study, led by Abra Jeffers of Massachusetts General Hospital, researchers were able to do two additional analyses: one that looked at cardiovascular disease risk in people who use cannabis but had never used tobacco products and a second one that looked at people who used cannabis but had never used tobacco products or e-cigarettes. Without tobacco use, the higher odds of heart attack and stroke persisted for people who used cannabis. For those without tobacco or e-cigarette use, only the higher odds of stroke remained."
Are't e-cigarettes tobacco products?