3d modeling for use in software (games or similar) imo is fine, but I would love to not see this stuff in civil engineering. People without AI are unable to design buildings that don't collapse (e.g. recent NYC apartment highrise), adding AI into the mix will just make things worse.
Why do you presume it will make it worse instead of better? Humans, as you point out, could not manage to competently review buildings that don't collapse. Why would adding an additional level of checks (human or AI) do worse than not adding one?
Some PM realized that agents will not push back against waterfall style project planning, so now we get an agent for all possible versions of waterfall projects.
fair critique but copperhead works as a feedback loop: design, verify, revise and repeat. closer to an engineering control loop than waterfall with an agent on top :)
When it gets to the end of the design loops, can it run simulations and tests, or is that out of scope i.e., for another product? (I didn't notice testing on a very quick peruse, but that doesn't mean I didn't simply miss it)
copperhead isn't just a wrapper around claude or gpt. it combines model intelligence with its own hardware IR, deterministic engines and continuous verification.
You joke but my coping saw cuts wood much faster than any of my normal saws. I don't know a lot about saws, I presume it's the type of blade on it. And I do indeed have three saws, which is a lot for someone who doesn't know much about saws. One that says it's good for cutting metal, one for wood and one for tight corners in wood (a coping saw).
Your coping saw probably has a pretty thin and narrow blade. That minimizes friction on the non cutting surface, and the amount of material removed because of the narrow blade.
Ram, hdd, ssd, even blu-ray writers are expensive. I purchased mdisc blu-ray writer for $70 in 2021, I see people are trying to sell the same device for $500+ wtf
Is it a Pioneer model? They stopped making their blu-ray drives which are at least perceived to be better quality than the competition, so NOS prices have gone way up.
It is minimal and somewhat neat. It claims to be P2P, but it still needs to bootstrap the connection via central server. I think that always might be a limitation of P2P.
I had a similar idea but for wireguard only, where you would allow people to join a vpn through a "shared" jumpbox. To get started with wg is very simple, the problem is the same you need public IP address to scaffold everything.
Out of the two options I probably would prefer going wg route, just because it generalizes better. But this is a pretty cool demo of webrtc.
Iroh's discovery is genuinely neat, and node-IDs-as-public-keys is similar to bitbang.
A browser can't join a DHT though because of no UDP sockets, so the connecting side would need a native client, which is the thing bitbang avoids. And DHT lets you locate the peer, but you still need somewhere to trade SDP and ICE candidates, so the rendezvous stays either way I think.
What I can do is keep the broker small enough that self-hosting is realistic, which is where it is now -- no accounts, no registry, nothing that grants access, only a simple broker.
If you are running this stuff on a cheap VPS with limited bandwidth, you will not want all of your traffic being routed as ingress and egress as that could easily eat your monthly bandwidth limit depending on the size of files and frequency.
Right, bootstrapping needs a rendezvous point -- that's true of any P2P system, including the wg version. The difference is what the rendezvous knows and costs. Here it holds no accounts, no registry, and nothing that grants access, so it's a phonebook rather than a keyring, and it's out of the data path so it's cheap to run and you can self-host it.
On generalizing: wg wins for "put these machines on one network." WebRTC wins on reach, because the connecting side can be a browser -- a phone, a borrowed laptop, a machine you don't administer -- with nothing installed. A wg jumpbox still needs a client on both ends and a public IP for the box. Different tradeoff rather than a strictly better one.
Fixed IP would be a way to go. Some people pick dynamic dns server so they can periodically update if their IP changes. But IMO it's just too complicated. I don't think there's a good way to go around ISP restrictions, especially in USA.
I host my site on my own home server, but I do have a proxy ec2 server to tunnel public traffic via wireguard back to my home server. This keeps things a bit more protected and my router/home network not directly exposed. I'm also not locked into AWS, I just use them for convenience, but could get any other cheap proxy to run wireguard. No dependency on tailscale either, it's just nicer interface to wireguard. Wireguard config is like 5 lines btw.
This will get worse with new generations. They grow up on tables and phones where file system is a completely foreign concept. You need an app for everything.
This is great! Thanks for the detailed write up and suggestions to improve. Like I said, this is my first time doing it and I have no background in EE or PCB design. If I make something else I'll give your suggestions a try.
I work on software and a bit of hardware for my job, at tonari. It's basically a low latency video "portal" to connect spaces. My most recent hardware project was to add an OLED display + rotary encoder dial to the product. That was a fun project, and it turned out great! It was also my first time doing a 4 layer board and having to worry at least a little bit about EMI and all that. All my previous stuff has been lower speed 2 layer stuff, like a custom mechanical keyboard.
I was actually very surprised myself. As I mentioned in the post, I don't do that type of work so it's all new to me. This got me also interested in CAD. You can design parts and don't need to own the 3d printer, could just get parts manufactured online.
Yeah, and since it seems there's some competition there, the prices are decent. You can get milled metal, 3D prints, populated PCBs... crazy what you can do now and how quickly it can be turned around, if you are willing to pay.
I wish there were more regional places like PCBWay and JLCPCB in US, EU, etc (with similar pricing) so shipping didn't require circumnavigating the globe.
EU has Aisler but what you win in shipping times you lose in production lead time (unless you pay for faster service).
And there are probably others but with even less visibility / willingness to interact with private customers. Hell, I used to live next door to a sales office for a local PCB fab, but they never bothered to answer my inquiry about prototypes.
(That's another thing the proto-friendly companies do right: instant quotes without log-in requirement)
> but with even less visibility / willingness to interact with private customers
It's honestly depressing. I could believe the theories (off vibes, since I'm not in manufacturing, just a hobbyist) that this is one of the reasons why China is overtaking the West in hardware. Manufacturers here just don't want to talk to you at all unless you're willing to buy a million units, it seems. From what I've heard, in China, it's far easier to get started (and that certainly matches my experience with jlc/pcbway).
Like, I get it, profit wise. But it's a bit of a sad state of affairs.
I hope that open models will dominate. The difficult part to reconcile for me is the amount of compute that's required to create and run such models. Small models are fine, I run local llms 27b param on a gpu, but it's not even close to frontier in capability. Who wants to drop $40k+ on hardware to run these things. Companies, maybe/perhapts. On the other hand, to run a DB I can get a server for $3k and handle tons of traffic on it and other things too.
A data center or a cloud? It's not difficult to find a good data center to colo at. The problem is then you have to bring your own hardware, technicians, and sysadmins.
However, if you don't trust cloud providers or inference providers for whatever reason then you probably aren't going to be excited to enter a co-op model where you're still effectively renting access to hardware that you don't directly own. There are already reasonably priced options to rent bare metal from a cloud provider.
The only way I see it working is if it's a bunch of medium to large sized businesses getting together to be able to rent out the spare capacity on hardware that they physically control. So an AWS equivalent where each rack is owned by a different company and retail VMs migrate between them transparently. But I question the overall economics of such an arrangement.
> enter a co-op model where you're still effectively renting access to hardware that you don't directly own
If we're imagining a co-op, then the participants should all be equal owners in an organisation that owns the hardware itself, otherwise it's not much of a co-op really.
But at that point you don't have the same sort of security guarantees about the hardware. This isn't like a farming co-op where a single expensive piece of equipment gets passed around to each participant one week at a time. There needs to be either an economic, security, or other advantage to entering into this arrangement for a medium to large sized player that would otherwise be colocating multiple racks in a privately owned datacenter or else renting bare metal instances in bulk from one of the hyperscalers.
Participants as equal owners doesn't preclude security guarantees.
The advantage is having a voice in setting direction, investment and ultimate shared incremental profit recovery over the rent capture from hyperscalers.
It's an interesting idea really. But it also comes with cooperation, which means trust, and a lot of for-profit enterprises are bad at operating cooperatively outside of 2- or 3- way engagements.
Would actually be a good biz model for the Colo facilities that keep shutting down as everyone moves to the big cloud providers.Now if they can get their hands on enough GPUs and RAM.
Incentives are pretty easy to align, if you have repeat customers and repetition. McDonald's sells me burgers I expect at a reasonable price, because they want me to come back tomorrow and not go across the street to Burger King.
I believe until the hardware designs catch up to be more commodized ala cryto mining evolution from GPUs to ASICS for specfic algos.
Designs (like Google TPUs equivalent) would also need to evolve to be more memory dense to be able to handle them.
Untill then it seems will be system time shares for the larger models , probably with a bring your own model and pay as you go.
> ala cryto mining evolution from GPUs to ASICS for specfic algos
I don't see it happening. A current gen GPU with a huge and fast block of memory isn't a perfect fit for these algorithms but it's relatively close. With cryptocurrency, mass small sha256 hashing was a totally different kind of computation.
> isn't a perfect fit for these algorithms but it's relatively close
I don't think that's true. The best fit out of what's presently available perhaps. Inference is almost entirely memory bandwidth bound at present, to the extent that GPUs with HBM have a massive advantage over those with GDDR. TPUs appear to be a much better overall design.
I expect that a hypothetical advance in fabrication enabling processing elements to be placed directly adjacent to dense RAM on the same silicon (not merely in the same package) would be superior in all regards.
> I expect that a hypothetical advance in fabrication enabling processing elements to be placed directly adjacent to dense RAM on the same silicon (not merely in the same package) would be superior in all regards.
Processing scales better than DRAM does. I think an HBM-like stack where the bottom layer has the math units is probably the ultimate form of that.
And it's possible that flash instead of DRAM is actually the better play, as long as you can hook up enough in parallel. RIP Optane.
There might be a community effort at some point. This happened in chess where the community recreated and then improved on Alpha Zero. You could run small training chunks on your machine. Some people donated thousands of hours of server time.
They never will. The only reason China releases open weights is because they can't compete on frontier models. Whoever has the frontier model has no incentive to give it away for free.
reply