Hacker Newsnew | past | comments | ask | show | jobs | submit | syntaxing's commentslogin

Whoa! I asked about something like this 2 years ago but never got to making anything [1]! Super exciting something like this exists!

[1] https://news.ycombinator.com/item?id=37885584


There was an old legend for an Apple bug (but I can’t exactly remember what). He complained about this macOS bugs for years. Worked for Apple for a couple months, fixed the issue, then quit.

Whoa this is pretty cool. I’m pretty into homelab stuff but self hosting your own gateway/firewall/router can get pretty tiring. I tried OPNSense and it worked great for a while. I tried self hosting unifiOS (I actually don’t get what’s different from the docker one from back then compared to this one) but it gets annoying. Was easier just to buy a cloud ultra for $130.

I used to do a wireless mesh with U6 (wifi 6) and I 100% agree with you. But with WiFi 7, everything changes. WiFi 7 wireless mesh is actually faster than hardwiring (mainly because my second AP is connected to a 1G flex mini). A WiFi 7 mesh can actually exceed 1G (using Wifiman, I get about 1.5G). Meshing with a WiFi 7 AP works great for me too since all my devices are only WiFi 6 compatible (for now). Highly recommend replacing your APs with a U7-<whatever you want>

But my wired network is already 2.5G and some devices are 10G compatible.

Interesting. That’s sounds really promising. Please tell me more. Are there any YouTube videos or resources you can point to?

Not much to it! Buy any Wifi 7 Pro stuff. Don't get the lite, its just not worth it. Especially confusing with Lite because Wifi 7 still works on Wifi 6 stuff but they call it Wifi 7?

Hardwire the one closest to your gateway, remotely adopt the others, and youre good to go. Also, double check your frequencies. 1, 6, 11 for 2.4 Ghz. No DFS for 5 Ghz. Scan the environment and try your best. Wifi 7/ 6 Ghz must be same frequency though which is another plus since no ones really has it yet.


All Qwen models are good at using tools, even the smaller 4B one. The 1.7B one gets confused easily

Thank you

Have you tried using them with something like Claude code or aider?


I’ve used it with Aider (32B and 30B, the previous 30B one, haven’t tried this fully nonthinking one yet) and 4B with home assistant. Both works great in terms of tool calling.

Like what type of tasks/tools are we talking about here, asking questions about the content from (PDF) documents or?

It’s interesting how the Qwen team more or less proved that hybrid reasoning doesn’t work and makes things worse. The fact that this model is almost on par with the bigger model in non thinking mode (old, they released a non hybrid model recently) is crazy.

Qwen3 32B is a hybrid reasoning model and is very good. You have to generate a lot of think tokens for any agentic activity but you will probably run the model locally and it wont be a problem. If you need something quick and simple, /no_think is good enough in my experience. It might also be because its not a moe architecture

Qwen3 32B was a hybrid model that came out in April, but these new Qwen July models have all ditched the hybrid mechanism and are either thinking or non-thinking.

By Qwen3-32B you mean the first released version from late April? I don't think Qwen3-32B-2507 has been released yet.

I agree with GP that since Qwen is now releasing updated Qwen3 version without hybrid reasoning, and experience a significant performance boost in the process, it likely means that the hybrid reasoning experiment was a failure.


Isn't that because all "reasoning" approaches are very much fake? The model cannot internalise the concepts it has to reason about. For instance if you ask it why water feels wet, it is unable to grasp the concept of feeling and sensation of wetness, but will for sure "decompress" learned knowledge of people talking how it is to feel the water.

Everything about LLMs is fake. The "reasoning" trick is still demonstrably useful - the benchmarks consistently show models using that trick performing better at harder code challenges, for example.

I'd argue that what's generally considered "reasoning" isn't actually rooted in understanding either. It's just the process you apply to get to a conclusion

expressed more abstractly: is about drawing logical connections between points and extrapolating from them.

To quote the definition: "the action of thinking about something in a logical, sensible way."

I believe it's rooted in mathematics, not physics. That's probably why there is such a focus on the process instead of the result


What hardware do you train on using axolotl? I use unsloth with Google colab pro

They unfortunately released at an unfortunate time. Bambulab became widely successful at the same time and just ate up the market. Anker printers would have been a hot sell otherwise.

Is there a link to the TestFlight itself?


Wasn’t sure if a random TestFlight link would be safe/wise to share, so shared original source.




Whoa this is really neat. I’ve been trying to get into Meshtastic but it’s hard to convince others when you need special hardware. Would be super neat if Apple did something similar. Shouldn’t be too hard since the AirTags use the same idea?

Would also be neat if there was a way to build a LoRA proxy to extend the range. I might give this a try with my meshtastic devices.


I'm working on a project that uses the same kind of idea as the Bluetooth tracking tags.

It's an Arduino library for mesh networking, that works over BLE and UDP, but it can also link to MQTT.

An MQTT node routes the packets it sees to the appropriate topics, and subscribes to topics for all the channels local nodes want, so you should be able to talk to anyone anywhere via the gateway.

The packet destination addresses are rolling codes, so you can't tell if someone's online just by watching their channel, at least not for more than an hour.

And there's a web app that talks directly to the public MQTT broker, and it can do chat and sensor data.

All payloads are Messagepack to make it easy to add new data types, and all packets are encrypted, authenticated, and timestamped to provide a bit of replay protection.

Everything is purely symmetric crypto, trust is left to a higher layer or something out of band, so you there's no handshakes or connection state management overhead, aside from one announce packet per hour to make the MQTT gateways work.

No LoRa, but the transports are modular and pluggable so you can easily add them. I just only have one LoRa Arduino node here so I haven't bothered writing a driver.

I'm also working on a Python port for easy pip-installable bots and home automation stuff.

https://github.com/EternityForest/LazyMesh#


Super interesting! Leaving the transport layer as modular is definitely a great choice! I like the idea of MQTT because there’s a lot of methods of serving it. I’ve been in the middle of setting up a meshtastic MQTT mode to try it out.


I was originally going to do OpenDHT, but that would have required building and paying to host a proxy backed for the web app.

I wonder what other transports you could do, like 38khz IR through a telescope?


Any line of sight stuff can be tough. Another one is standard 433 radio but difficult since its such a noisy environment.


It depends on the antenna efficiency of course, but I was surprised to discover that BLE modes around 128kbps using coded-PHY have a range extending over 1-1/2 km without a directional antenna. At 2.4ghz its line of sight, of course, but still…


That's extremely surprising to me too. I would have expected BLE to reach a few meters, not kilometers. How can I learn more?


Search for “BLE coded PHY”.

The S8 coded mode with 125kbps rate is the long distance one. Support for it in phones is not widespread, sadly.


Thanks! I didn't realize that was the key phrase.


I think the reason AirTag works is because Apple turns it on-by-default on i-devices and people can't be bothered to go turn it off. For a chat to work on the same scale it would literally need Apple or Google to ship it as enabled-by-default on all phones.


It'd be cool if Meshtastic's UDP mode could run over BLE like this, for local bluetooth clouds linked by just a few LoRa nodes.


Yeah, a BLE first mesh system honestly makes more sense in today's world since it's baked into every phone. In theory, a BLE to LoRA bridge should be doable with the existing meshtastic hardware like Nordic's nRF52840. The biggest caveat is going to be the data rate. Meshtastic is designed for around 200 bps (Long range mode) which vastly pales the ~2Mbps BLE expectation.


FWIW, I've found a T-1000e to be a pretty good way to get people into meshtastic. It's not perfect because it has a weird dongle to charge, but it's pretty cool and I think you can convince people it's a worthwhile thing for emergency recovery.


Heltec MeshPocket is another. A powerbank and LoRa device.


Once you have brought LoRa into the mix, you might as well just ask for p2p cell connectivity. Our phones could totally talk to each other over reasonable distances with no extra infrastructure.


the special hardware's cheap enough that if they can't be bothered, then they're not serious about it.


Consider applying for YC's Fall 2025 batch! Applications are open till Aug 4

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: