I can't help thinking "doesn't matter much unless it's perfect" because if someone is using this to build a board (cool) but then it's not flawless, troubleshooting will be quite tough as a novice. Like, say, when I start digging into the web code generated by a coding agent.
I am most excited about it bringing down the barrier so more people join in on hardware fun, so hopefully it will unlock folks that stayed away in the past.
I hope you guys are successful because I would love to see more US options in this space. Realistically, you can get boards in 7 days from China, and competing with them on price seems like a challenge. So that leaves ITAR and people who need boards even faster. Sounds like that’s where you’re focusing your energy already.
What pick and place are you using now that you’ve moved on from the Neoden world?
Out of curiosity I gave Fable (on max effort) a CAD task yesterday, which was to design a space efficient carrying case for a set of fasteners in my repair kit for work. It used CadQuery to generate a STEP file. The result was pretty much exactly what I wanted, without needing any manual edits. I did go back and forth with it on the design, but was really impressed with the result. Without prompting it included nice touches like ribs on the bottom of the lid to stop fasteners from migrating to adjacent compartments, and the right tolerance for the fit between the case and the lid. This is a dramatic improvement from Opus 4.8.
Well the thing about CAD files is that through reinforcement learning you can basically ask the AI to generate the CAD file an arbitrary item - say it’s a rabbit. It might have examples of this already in its training set and it’s essentially a similarity lookup - but for sake of argument assume we are giving it examples at the edge of the distribution (the whole point of RL). It guesses and you render the file. You pass that image to another AI (not being trained) and ask it if it resembles the description you gave the AI in training. If it does, you have a positive example. If it doesn’t, negative. In that way you can essentially apply transfer learning from the image recognition functionality to the description -> CAD functionality.
But is that actually spatial reasoning? Or is it effectively image generation? Because there’s a difference. Spatial reasoning implies that you could drop it in a video game, give it rules, and let it run. And it would play the game well. Like a flight simulator. That would be true spatial reasoning because spatial reasoning is not just identifying objects but understanding how they interact with one another in a highly quantitative way.
I'm not sure I see the distinction you're making between 3D design and other spatial reasoning tasks. You can use RL to teach navigation or video game play too. Does that mean these tasks are not spatial reasoning? Additionally, 3D CAD is all about understanding how objects "interact with each one another in a highly quantitative way." I mean, not in the rabbit example, but the container Fable designed for me holds around 30 different types of objects. It figured out a way to arrange them that was more space efficient than what I'd originally described. It considered the best way to stack the fasteners in each bin to pack them as densely as possible. It identified the risk that some very thin objects could slide between compartments in transit, and modified the design to prevent that. It correctly solved for the tolerance between objects that needed to snap together. These all feel like understanding how objects interact with each other. The model didn't just talk about these concerns, but created two 3d models for the case and its lid that accurately reflected them. I hadn't seen that before.
I don't agree with the conclusion here. The issue in the sports court example could just as easily be addressed by adding granularity. If the system allowed users more granularity in the length of their sessions, the undesirable strategies in the post would no longer work.
Agreed. I don't think this is a granularity problem, it's a fragmentation problem.
Users are selecting two data points: start time and duration. The fact that those two points have different granularity is what leads to fragmentation, not the fact that start time is more granular than duration.
It is true, however, that more coarse grained allocation sizes will help minimize fragmentation.
>The game features a panic button that when clicked on will cover the computer screen with a fake spreadsheet. The player can also choose to print out Maxie's current pose as a pinup.
Love these! On the record player, would be fun to play Morse code “records” where the message was encoded by selectively removing spots of copper from the rotating plate.
Thank you for sharing this! The part of this problem that I’ve found most challenging over time is keeping quantities up to date. I wonder about a function where you could say “I built 25 of these modules” and it would deduct the quantities for individual parts accordingly (optionally adding an extra percentage for the 0402s you dropped on the floor) Or, while I’m dreaming, a feature that parses a DigiKey/mouser/etc. order (or reads the barcodes on the bags!) and updates quantities for you.
Would also be curious about how best to support different storage approaches, like reels on a shelf etc.
We seem to think in similar tracks! I do have the exact features you describe already. Each BOM that I add can then be searched for and its parts deducted in batches (of for example 25 times the amount of parts are needed per module).
And a BOM can be imported from a csv file exported from a Digikey/mouser shopping cart. Although now that you mention it, re-stocking from such a csv would be a really nice addition.
I’m sure that Apple did tons of A/B testing, focus groups etc. with different image processing parameters to arrive at the settings that their phones use for photos, and from these comments it’s clear that a lot of people prefer the iPhone photos. When I was in grad school (in the pre-iPhone era) I photographed lots of weddings on the weekends, and one thing I noticed during the process was that people often have a set idea of what good photos look like. This idea of a “good” photo is often not tied to what the scene in front of them looks like. For example a “good” photo that includes a sunset will show a highly saturated orange/red sky, even if that’s not what the sky looks like at the moment the photo is taken.
Personally, I carry around a Ricoh GR3, and shoot random shots with the iPhone, but when it really matters I’ll use the Ricoh. The way the iPhone flattens the lighting is what bugs me the most. Recently I was at a kid’s birthday party and each kid had a cupcake with a candle in it. The room was a bit dark, and the Ricoh photo showed that each kids face was illuminated just a bit by the candle in their cupcake… The color temperature of the candle light is warmer than that of the room light. The photo makes you feel like you’re really there. My friend shot a photo on her iPhone at the same time and we compared afterwards. In her photo, every kid’s face is well lit and the candle effect is gone. She likes her shot better and I like mine. Some people want a shot that reflects what they saw, and some people want a shot that looks like what they think good photos look like.
MKBHD does a phone camera tournament (yearly?) and I believe at least once he found that people prefer a brighter image, even if that didn't match reality. No source or proof, just something I vaguely remember
I have found that non-photographer people have generally have absolutely no sense of what makes a good photo. They don't care about composition, lighting, colours, detail, or ambience. Their mental/emotional impression of what the photo represents is more important than the photo itself
Yeah I agree. The market for companies that do something in a web browser may be saturated, but outside of that world it feels like there are more opportunities than ever. For example in the space of AI + robotics we’ve barely scratched the surface.
I wonder if part of what’s happening may be a generational shift in how people in and around the startup world view work/life balance. In my limited experience, GenZ folks tend to place a premium on reasonable working hours, so it’s not surprising they would evaluate the trade offs between startups and big tech differently.
I'm biased but I think the most interesting opportunities involve software that interacts with the physical world in some way. These projects will definitely involve some electrical + mechanical engineering, product design etc., but I'd guess that in terms of number of hours worked, they are still more software engineering than any other single discipline. Probably with a very different set of tools than those used on the web.
reply