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

When claude code became quite good, I became quite excited. I wanted to make sure that I consumed my full quota ( 5-hourly and weekly ). I was working on four to five different project at the same time. And it was excruciatingly painful after the initial rush wore off.

I went back to working on one project at a time to regain my sanity.


Hi,

First of all, congrats on the launch. I have worked with many such products and this is something that is required. There is no one solution fits all in this domain. Each vendor has their quirks that. So this is overall a great problem to tackle and kudos to you for taking this up.

Could you elaborate a bit more on how this is different from Renode?

Also you are entering into a market that has well established legacy players. Most of the firmware is developed in China. How do you intend to make a space for yourself in such a scenario.


Thank you, and for sure market is very fragmented. Which is exactly why we are starting small and modelling deep with stm32 as our workhorse here. The quirks we are finding them everyday! I have seen mcu controller behaviour from neither errata nor datasheet and thats the beauty of field testing.

I think our biggest difference to renode is the silicon validation and our commitment to keeping our simulation honest. I have struggled to find simulators that people actually put the measured differences and that is the direction we will go. We haven’t put our silicon validation benchmarks public yet but we will in the coming months!

As for legacy players, they don’t fill the niche enough. Thats why we exist and as a startup can fill this niche for the people that need it, whether its china, or anywhere else.


How would a "9-billion-parameter model" replicate on IoT devices that have typically no more than 2-4 MB of RAM or 8-16 MB of flash?


Maybe the idea is not to have the whole model but the executer so … by copying an API key?



You should read the report, it is interesting technically if you ignore the fear-mongering. It distributes the model's layers over multiple compromised devices.


And considering typical WAN latencies of 20+ ms, what’s the expected inference rate, 1 token per month?

Maybe I did something wrong or unaware about the tricks to make it fast, but last time I tried to play with exo and distribute processing over a few home LAN computers to see if I can run models that don’t fit my 3090, the penalties were brutal.


0.5 tok/s I think is their claim. Basically skynet


As a fellow Indian growing up in 90s, I feel you. I had my fair share of beatings from teachers in school.

There were cases where parents who found it challenging to control their kids, would encourage teachers to use physical force on them. Parents deemed it was school's responsibility to instill discipline in their children.

As far as I can recall, use of violence to control kids was pretty common back in the day, both at home and at school. I don't see that happening nowadays, though.


Thankfully, My parents who were both teachers specifically told my teachers not to hurt me physically and my teachers actually listensed to them.


what happened in 2018/19?


In 2014, he fired his PR person when she asked for a raise (she was very good at her job).

In Spring of 2018, investigative reporters at the LA Times found that Elon acted like a tyrant removing yellow hi-vis safety tape at Tesla factories because he didn't like the aesthetic, which lead to more workplace accidents. This was when Elon started getting extremely adversarial with the press.

https://www.latimes.com/business/autos/la-fi-hy-tesla-factor...

A few months later, during the Thai cave rescue, he flaunted some kind of coffin made from spare rocket parts as his plan to save children who were trapped in a Thai cave, and when one of the actual rescuers Vernon Unsworth told him the cave was too narrow for that to be of use, Elon called him a "pedo" (note, Elon also begged to go to Jeffrey Epstein's parties).

It all went downhill from there.


> one of the actual rescuers

There was about 10,000 people involved in that rescue. There was no single "rescuer" or authority.


Which is why I said "one of". And by "one of" I mean one of the leaders of the whole operation, someone who lives in Chiang Rai and has extensive knowledge of the cave complex, who coordinated Thai government working with the British Cave Rescue Council.


You posit as if Vernon was some sort of official, he wasn’t. It was collective effort by many many people.

The reason Musk got upset was the interview with Vernon where he got mean about Musk’s efforts. And since freedom of speech, Musk got mean in return.

Vernon tried to sue to no avail which all you need to know a out him.

Painting Musk a villain here is insane. He tried to help, instead he got slandered and sued.


Vernon wasn't an official, but he's the first guy the officials called to help and come up with a plan. Elon Musk wasn't trying to help, he was trying to get publicity. Had anyone listened to Musk, the children would have died. I think someone who doesn't know what they are doing wedging themselves in a life or death situation while arguing with the experts with the knowledge and experience wasting their time with cockamamie plans is a villain in my eyes. Elon Musk is basically Syndrome in the Incredibles movie: angry and jealous that he isn't the center of attention.

And if it isn't clear today, Musk is clearly a villain. The Thai cave rescue was when he showed his real face to the world by being a total and utter jerk to the real heroes.


> he's the first guy the officials called to help and come up with a plan

Not true, but that doesn't matter. Tons of people offered help, were invited, had plans, etc. Why Vernon was such an asshole to Musk tho?


Elon was not invited, and he still insisted on showing up. That makes Elon the asshole.


Except public Twitter exchange with PM or something like that


People started realizing that it was all an illusion and that, surprise, surprise, yet another billionaire egomaniac actually has the moral fortitude of a maggot.


It would be great if these posting could have tags like below prefixed at the top by the poster.

[HUMAN-CODE] - code written by humans

[AGENT-CODE] - code written by Claude/Codex/Other agents out there

[HYBRID] - Combination of Humans/Agents

I would prefer to go through projects where people are writing code themselves. And this kind of tagging would help me filter through easily.


Please, yes. Half of the fun of being on HN was "Show HN: I made some random thing", which typically amounts to "look at what prompt I wrote" nowadays. I use agents every day, I'm not against it, but it completely changes the lens through which I view a project


But how would you know if someone was just lying when they added "[HUMAN-CODE]"?


People don't say when they vibe code things but it's still very obvious.


Vibe code, maybe.

But a lot of developers use AI as a pair programmer, and I doubt you'd be able to tell the difference.


Just assume all are [AGENT-CODE] unless otherwise specified.

There's no reason to write code any more by hand, except as a hobby/fun/art or to comply with some requirement.


Indeed as it helps to give more meaningful feedback. Somebody in [HUMAN-CODE] might appreciate suggestions on the code but not care as much about business whereas [AGENT-CODE] probably prefers the other way around.


Good idea. I've edited my post to include that.

Frankly, it's surprising that we'd even need to stipulate something like this. It would never occur to me to post to a "show us what you're cooking" thread with: "Look what I asked DoorDash to bring me!", but I guess that's where we are now...


AI is you being the director. Hey, I directed this movie != I acted in it, I edit it, I control the camera, I wrote the music. It equals I told others what to do and my judgements and decisions affected the final result. Same with making with AI. You direct. If you have taste and talent you direct a good result. If you have no taste nor talent you make slop. This is no different than in movies.

https://www.imdb.com/title/tt1213644/


Excessive groundwater extraction and urban development could be reasons for this. But these are common practice in almost all of the modern world, so why is this only showing up in Mexico City.

There must be other contributing factors too.


Mexico city was built on top a dried lake. It is really fascinating stuff, I would highly recommend to look it up. Ancient farming technology, floating islands, any nerd is gonna love it.


It’s not only Mexico City. This is happening in Jakarta, the Central Valley of California, Beijing, really anywhere with excessive groundwater extraction.

It is just most people pay attention to CDMX because it is a very large 20M+ city with a lot of American and European tourists, and it is happening quickly to the point where you can see it with the naked eye.


Mexico City (Tenochtitlan) used to be a city build on a lake, partly on an island, partly on some kind of static rafts / floatting gardens (chinampas).


You can see evidence of subsidence in the rest of the developed world as well: SAR interferometry is sensitive enough to pick up the minimal amount of building movement from building Crossrail in London. It's just this article focuses on Mexico City, where it's dramatic


I am building https://framebench.dev/ in my free time. It comprises of a bunch of utilities aimed at CAN BUS related workflows. Some things that are supported as of now:

- Easily parsing UDS, ISO-TP protocol CAN frames

- Parsing CAN-DBC files

- Building CAN Frame Payloads based on messages in a DBC file.


When it comes to writing small scripts for various utility tasks, I now rely heavily on LLMs.

Additionally, lets say I would like to write something new that I don't know anything about, I now go first to a LLM, or two of them, to get an idea of what would be the best approach. It saves a lot of time that would otherwise go in research and discovery.

But I still don't trust it enough to write production level code. When I am writing something that I know will be released in a product, I steer clear of LLMs.

All of this when I am doing my day job. For my side projects, I try not to use LLMs.


I am building https://framebench.dev/ in my free time. It comprises of a bunch of utilities aimed at common workflows found in automotive industry. Some highlights:

- Easily parsing UDS, ISO-TP protocol CAN frames

- Parsing CAN-DBC files

- Building CAN Frame Payloads based on messages in a DBC file.

This is really early stage. Any thoughts are welcome.


Consider applying for YC's Winter 2027 batch! Applications are open till November 2.

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

Search: