2026-02-16 11:12:58 That's very odd - I agree that a netlist is not the best way for a HUMAN to check a schematic, but it's hard for me to see an AI or any other kind of SOFTWARE doing anything with a PDF of the schematic. 2026-02-16 11:13:24 I absolutely believe that schematics carry visual information that's important to a human designer - just in the visual structure. But a program can't really "see" that. 2026-02-16 11:14:06 I mean, I suppose you could attempt to write software that considered the geometric layout of the components on the visual schematic, but I've never heard of one that does. 2026-02-16 11:15:32 I once found a really nasty intermittent race fault in a guy's bit 50 page schematic. I was the company's new EE manager, and the owner asked me to help him - he'd been fighiting the issue for weeks. We sat down to go through it, and I could tell he was really annoyed - like "How is this guy who knows nothing about this design going to help me?" 2026-02-16 11:15:42 KipIngram, I'm still attempting to find out how accurately an AI can analyse a PDF or whatever of a schematic, my expectation is high 2026-02-16 11:16:30 We were just paging through the schematic, just a few seconds a page. It got to the point where he would page forward every few seconds before I said "next page." But all of a sudden something fired in my head and I said "Wait - go back to that last page." 2026-02-16 11:16:50 Down in the bottom right there was a visual indication in the topology of the circuit on the page of reconvergent fanout. 2026-02-16 11:16:57 KipIngram, based on experiments such as asking "explain how a dual npn transistor self oscillating inverter works" 2026-02-16 11:16:58 So I just decided to look closer there. 2026-02-16 11:17:12 And sure enough - there WAS a race hazard there, and that DID turn out to be the problem. 2026-02-16 11:17:29 It's one of my favorite memories - one of the few times in my career I really looked like a badass. 2026-02-16 11:17:46 That guy followed me around like a puppy the rest of the time I was there - the experience just 100% won him over. 2026-02-16 11:18:01 Ive been very impressed with the level of knowledge of earlier AI's posessed there 2026-02-16 11:18:07 And I would have NEVER found that problem had the design been done in Verilog. 2026-02-16 11:19:02 Reconvergent fanout isn't necessarily "wrong," per se, but it is a common source of timing hazards if you're not really careful. 2026-02-16 11:19:59 So in that case it was the fact that we had a SCHEMATIC that saved the day. 2026-02-16 11:20:50 It was one of those really nasty bugs where the thing would run like a champ for a week or more, and then suddenly glitch. 2026-02-16 11:21:19 eww, theyre the worst, I've seen careers ended because of such problems 2026-02-16 11:21:30 Indeed. 2026-02-16 11:22:23 Anyway, that cemented the value of schematics in my mind. 2026-02-16 11:23:20 Ive been using gschem (part of gEDA) for schematics since about 1997 2026-02-16 11:23:52 Basically he was producing signals A and B in one bit of circuitry. They went off different ways, each going through some (separate) additional processing. Then they came back together - one was used as a clock and th eother as data for a flip flop. 2026-02-16 11:23:58 I know it really well after so long ... it's a bit like the old ORCAD for DOS was 2026-02-16 11:24:01 Well, the data BETTER get there first, with a suitable setup time. 2026-02-16 11:24:05 And now and then it didn't. 2026-02-16 11:25:08 yeah, although I have never worked in that area, I can imagine it's a daily problem in new designs of chips 2026-02-16 11:25:30 I guess I could write a program that could extract reconvergent fanout cases from a netlist, but it's something you can just see on the page with a schematic. 2026-02-16 11:25:53 Ive had enough setup problems just with mcu's assessing external peripherals, as regards setup time etc 2026-02-16 11:26:00 It helped that it was all on one page - probably would be a lot harder to spot if it crossed pages. 2026-02-16 11:51:31 PDF schematics often contain a lot of net/link info anyway 2026-02-16 11:51:38 So maybe the LLM can use that 2026-02-16 11:52:08 when I find the pdf of the schemayic Im interested in I'll find out 2026-02-16 11:52:24 sadly I only have a png picture of it 2026-02-16 11:52:44 Good luck 2026-02-16 11:52:53 and I dont have the rest of the schematic for some reason 2026-02-16 11:53:49 veltas, Im only testing the AI, when it comes to verifying how a circuit works, I always use SPICE and simulate the circuit 2026-02-16 11:54:42 but I have a new and impressive Open Source AI and I want to test the areas I'm interested in 2026-02-16 11:55:00 Understood 2026-02-16 11:55:16 so far it's very impressive with it's answers 2026-02-16 12:51:02 veltas: That's fair - I don't really know much about PDF internals, but sure - it has to represent those lines you see somehow. I just don't think of it as a format that really conveys any specifics about what the components are and so on. I did say above that I could at least imagine trying to write software that could suss out all the same stuff from a netlist, and I guess a PDF would 2026-02-16 12:51:04 ultimately contain similar information. 2026-02-16 12:51:41 But the kind of thing I was talking about above is stuff that a human expert can just "know" with a glance - the entire circuit structure just flows into your head. I wouldn't even know how to start saying how we do it. 2026-02-16 12:52:36 I'm not 100% convinced it's all "algorithmic" at all - I've long felt that there's something more to us than just algorithms. 2026-02-16 12:55:19 Actually schematics often you can click on a net name or component to get a list of info/links 2026-02-16 12:55:55 Personally I find schematics very algorithmic, but I'm not an EE by background 2026-02-16 12:56:06 I just have picked up stuff from my day job 2026-02-16 13:07:42 PDF is primarily a page description format, so it's mostly a sequence of commands like "draw a line from x1, y1 to x2, y2," or "place this JPEG image over there," or "put this string here." Of course, JPEG images can themselves contain lines, and there're also line drawing characters in standard Unicode fonts. So "lines" may be represented differently depending on how PDF was generated. 2026-02-16 13:07:42 AIUI, PDF can also contain arbitrary RDF (subject, relation, object) tuples, that are typically used for such metadata as "(this document) : dc:author : John Smith". I suppose it's possible to have a copy of netlist there as well, but I have no idea whether anyone actually does that. 2026-02-16 13:15:16 I don't assume that at all. At least not in every possible incarnation of what "analyze" might be. Certainly a lot of what we do can be modeled via algorithms. 2026-02-16 13:15:16 But no algorithm has an kind of real "awareness," so there's at least that that we do differently. 2026-02-16 13:15:31 You only have to assume that if you buy into materialism, and I don't. 2026-02-16 13:16:25 I just don't buy the model that consciousness emerges from the material world - I think it's the other way around. Consciousness is fundamental - the physical world is an artifact of its operation. 2026-02-16 13:17:13 It's not correct, but it's almost correct to say that these LLMs are like a high-dimensional linear function 2026-02-16 13:17:17 And science explains the physical world. It has no way of reaching outside of it, because all of our measurements and observations are confined to the physical world. 2026-02-16 13:17:37 And sometimes you get comfy almost-interpolated answers, and sometimes you get extrapolated answers that happen to be right 2026-02-16 13:17:39 Yes, I agree with that. I think it's quite close to correct. 2026-02-16 13:17:48 Sometimes you get very very wrong extrapolated answers 2026-02-16 13:17:58 That's for sure. Astoundingly wrong. 2026-02-16 13:17:59 But at no point is it 'thinking' 2026-02-16 13:18:04 Right. 2026-02-16 13:18:28 But it makes a good show of it, and I think the vast majority of the public will fall for it. 2026-02-16 13:18:33 I prefer to think of it as a nondeterministic algorithm for doing a high dimensionsal linear approximation 2026-02-16 13:19:11 And I don't make a habbit of talking to such algorithms :) 2026-02-16 13:19:13 Even knowing as well as I do that it's not thinking I still sometimes find myself getting annoyed with it the way I would a person interacting with me in whatever way it does. 2026-02-16 13:19:19 It's a really easy trap to fall into. 2026-02-16 13:19:40 Yeah and then sometimes it fails to explain if you should drive or walk to a car wash that's 50 metres away 2026-02-16 13:19:54 And you come crashing back down to reality 2026-02-16 13:20:08 Yes. 2026-02-16 13:20:45 Science is indeed concerned with things that can be measured, predicted and, ultimately, controlled. Whether there are things that aren't like that is debatable, but the thing is, science is the tool to /reason/ about such things. How, then, do you reason about things that are out of the scope of science? 2026-02-16 13:20:46 Google probably seemed 'smart' to some people when it came out, but others understood that Google just knew good algorithms for lookup, aggregation and association of webpages based on their content 2026-02-16 13:22:24 iv4nshm4k0v: I'd fall back on rationalisation and religion without ability to do scientific research 2026-02-16 13:22:57 Which I have to do a lot anyway, not because it's impossible but because I don't have the resources to research everything that's not got a paper 2026-02-16 13:23:34 And lots of papers exist that don't really give any serious confidence in a scientific answer, just are pseudo-scientific 2026-02-16 13:23:36 Yes, at least in the beginning Google applied some kind of eigenvector analysis to the web. They were the first ones to really do it in such a mathematical way, and it worked quite well compared to the other approaches around. 2026-02-16 13:24:02 I think it's similar what's happening with LLMs, but deeper 2026-02-16 13:24:20 Also, if materialism holds, the chances of encountering a serial killer outside your door are independent on what you think they are. If idealism holds, then, well, the more you think about serial killers, the higher are your chances of actually meeting one. 2026-02-16 13:24:47 So I frankly find materialism much less scary than. 2026-02-16 13:25:13 I'm not a materialist but I don't believe in a system where thinking more about serial killers make them appear 2026-02-16 13:25:33 Hey iv4nshm4k0v I did actually install NetBSD on my primary computers 2026-02-16 13:25:59 I've just had a lot of trouble with wifi on my laptop, I have to reset it repeatedly to get it to work at first 2026-02-16 13:26:06 Eventually it gets more stable 2026-02-16 13:26:32 I'll maybe try and fix the driver but I've got better things to do currently, the libera channel for netbsd have been helpful 2026-02-16 13:27:47 Well, I don't believe that either (that thinking about things can make them appear). That's getting a little too "New Age-y" for me. 2026-02-16 13:28:06 Ultimately the world we actually perceive out there DOES follow the laws of physics. 2026-02-16 13:28:28 In exactly the same straightforward way the materialists claim. 2026-02-16 13:29:18 Our conscious will CAN affect the world aorund us, but we do it through our physical actions. 2026-02-16 13:29:57 There has to be a link between our will and the physical world at SOME level, but I think it's very limited and just initiates processes that become our muscular activity, which is then the thing that affects the broader world. 2026-02-16 13:30:48 One of the benefits of being an idealist is that you no longer have to EXPLAIN consciousness - you get to just take it as given, the same way scientists do quantum fields. 2026-02-16 13:31:18 And that doesn't bother me, because we all know consciousness exists - each of us directly senses our own. 2026-02-16 13:31:45 In that sense the existence of consciousness is a whole lot more obvious than the existence of quantum fields. 2026-02-16 13:32:22 And we also all know that "mind" can create the perception of a physical world without any physical substrate at all - it happens every time we dream. 2026-02-16 13:35:23 There are even documented cases - clinical data - of patients with dissociative identity disorder (split personality) have dreams in which multiple of their "personalities" participate and interact. Later when interviewed those different personalities each recall the dream from their own perspective. 2026-02-16 13:35:36 So a SHARED physical world - cooked up purely in mind. 2026-02-16 13:43:54 Of course I assume that consciousness exists, theologically. I don't see it as impediment to looking into how the same can be explained scientifically. In fact, there's a variety of health issues - mental health, specifically - that would benefit greatly from having an actual scientific understanding of what's actually going on. 2026-02-16 13:43:54 As to dreams, I rarely remember mine, but I don't recall ever dreaming up something /entirely new/ - so far, they've been rehashings - even if on occasion rather throughout - of my past experiences. I do find it curious, though, how my dreams, while overriding the visual channel, still allow tactile data to propagate to mind. 2026-02-16 13:44:52 When I was young I often had dreams about things and places that I've never seen etc 2026-02-16 13:45:18 At least not as far as I know, it's interesting to wonder if some of these places I dreamed about I had visited before I had a fully developed mind 2026-02-16 13:46:39 I also now mostly dream about places I've never been or seen, that aren't real, but they're similar to real things at least 2026-02-16 13:53:15 I have good memory. I generally can associate places I see in my dreams with real places I've been to, though the spatial relations will often be all wrong. Like I've said, rehashed. It's of course possible to see in dreams something that was previously seen in a movie or on a painting. 2026-02-16 14:27:54 Like I keep having dreams about large buildings, homes, etc where I'm helping move in or 'expand' into unused rooms. I've never done this in my life but it keeps happening. 2026-02-16 14:28:13 Also rooms with lots of spiders, which I don't enjoy, I get those less now 2026-02-16 14:30:58 After leaving uni I still got dreams about forgetting to turn up to exams, it's good to wake up and realise you're not at uni anymore and don't have exams to do or miss 2026-02-16 15:34:08 Most of my dreams are in entirely unfamiliar places. Usually rather unrealistic places. It's like they're realistic on a small scale, but yet you'd never find the particular arrangement of things in the real world most likely. 2026-02-16 15:34:32 And it's very unusual for any of my dreams to contain anyone I know. It's always strangers. 2026-02-16 15:37:22 And the "plot" is usually terribly chaotic and incoherent. You'd think it would be easy for me to realize I was dreaming, but we seem to have some built in resistance to that. I've long wanted to "lucid dream," but I've only done it once in my life, and then only briefly. 2026-02-16 15:38:00 Once that I know about - I probably have more dreams than I remember. Those I remember are fairly infrequent. 2026-02-16 15:47:25 Around 2012, I've had a dream that suddenly started turning into a nightmare of sorts. I've decided to wake up, and found a cat lying on my neck. My guess is that she pressed my carotid artery with her paw, reducing the brain blood flow and resulting in a nightmare. 2026-02-16 15:47:25 Nowadays my dreams - those I can recall - are much too short to realize I'm dreaming. In the 1990s, lucid dreams were fairly common for me. 2026-02-16 19:06:21 I'm starting to think that if each word only returns one thing to the stack then you wouldn't necessarily need words like swap dup or over 2026-02-16 19:06:40 because you could just line up the stack in the correct order 2026-02-16 19:20:21 Not sure avoiding DUP et al. would be all that practical an approach. Implementing : FOO ( n) compute-address-here DUP + ! ; without DUP would require factoring compute-address-here into an actual word, for instance. IOW, you risk requiring the programmer to do something the machine can do. 2026-02-16 19:20:21 AIUI, most of the common Forth words do leave at most one new value on the stack, but sometimes you need OVER and such to account for how word is called. 2026-02-16 20:00:45 (Well, it should've been SWAP OVER rather than DUP, but the point still stands.)