I really wish that there was a product that tags the LLM's thinking steps to the code that it's generating or modifying. Comments aren't sufficient as it's hard to track them across edits. Think something along comments in Google Docs: "I modified this block to prevent a race condition."
That would help me understand what the model is doing way better.
One of my best hires ever was a VR dev we brought in as an intern. He became the backbone of our Unity/Unreal work, including some genuinely gnarly low-level haptics integration into the physics engine. On paper he didn’t look like the “obvious” pick: he’d majored in English Literature, largely because his (UK) school’s CS track was taught in a way that turned him off (they were still doing Fortran…). But he could build.
After our startup, Improbable scooped him up on the strength of that very real, shippable experience, and he’s now a senior SWE at Epic, doing exactly what he loves.
One practical thing that’s helped me find these kinds of people in startup interviews: optimize for calm + realism. My #1 goal is to get the candidate relaxed enough that I can see how they actually think and code. I often ask them to bring any public code they’ve written and we walk through it together. It’s a great way to surface judgment, taste, and real ownership that don’t show up on a resume.
It's interesting that most of us have that story of someone who didn't pattern-match and yet ended up being absolutely stellar. Makes you wonder just how much latent talent is out there not being given the chance for one reason or another. Hope this article reminds people to dig beneath the surface a little more.
And yes, many candidates struggle with performing under the totally unnatural pressure of an interview, so you can cater to them with something like the github project review. Then you end up potentially filtering out people without a rich body of work that can be easily reviewed, which is a trade-off. Actually something I've been meaning to write about, I always say that there's no way to please everybody with an interview funnel. Someone perfectly fine will be filtered out, or turned off, by any of the approaches you choose.
You just need to choose which false negatives you will be ok with.
| In fact what I would like to see is thousands of computer scientists let loose to do whatever they want. That's what really advances the field.
- Knuth
| How do you manage genius? You don’t.
- Marvin Kelly (Director of Bell Labs)
We have thousands of examples, quotes, and clichés where dark horses completely change the field. In CS we see this over and over so much that it's a trope of any successful startup. I just wonder when we'll notice the pattern. With all the clichés like "curiosity is worth 10 IQ points".
I think, at least for the cutting edge, it's easy to understand why these tropes are true. (IIRC Kelly even discussed it) Experts already know what the problems are and are naturally drawn to fixing them. By focusing on impact or importance all you're doing is taking away time from problem solving. Taking time from allowing people to be creative. If creativity wasn't required we'd have already gotten there.
I'm often left wondering how much we waste by trying to over optimize. How much we hurt progress by trying to attach metrics to things that are unmeasurable.
Honestly, the thing I'm most excited to see from a post scarce world is how humanity changes and progresses. When we then have this freedom to explore and innovate. To let people become experts in what they want. To let experts explore the topics they want, without need for justifying their work and the stress of not being able to put food on the table.
But until then, maybe we should recognize that innovation is so difficult to measure and has so much noise that we shouldn't rely too heavily on what the conventional wisdom says. If conventional wisdom could get us all the innovation and was optimized then startups wouldn't exist as established players could just follow a clear playbook and out innovate before anyone even has a chance. It's weird that we both recognize big established players are too big and set in their ways yet we also look to them as the playbook to follow to succeed at where they fail.
I really think there's a lot of untapped potential out there. So many just waiting to be given a chance
Meshcore seems like a much better choice for messaging in dense-node environments. I tried to set up a Meshtastic network for my Burning Man camp this year, and since default builds have historically been crippled by retransmits, we made a concerted effort to give camps their own frequency slots.
While that ensured campmates could communicate, it unfortunately isolated us from the wider network—you had to choose one or the other. Also, regarding the UI: if you think Meshcore is bad, Meshtastic is worse. I spent most of my time teaching people how to navigate the clunky, overcomplicated, and inconsistent mobile apps. I’ll likely try building a custom app on top of Meshcore for this year.
Agreed. Ran the comms for my burning man camp and everyone kept getting confused with the channels mess among other usability issues. I like where Mesh core is going, just wish the repeater nodes could run on gateway hardware so they don’t become the choke point with a half-duplex radio (bs like 8 full duplex channels on the RAK wireless gateway)
I got excited reading this thinking this can help me with the SDK for a stereodepth camera I'm working with only but read then reading it's for the Next.js stack :sob:
On a separate note I like the retro DOS feel. Will probably look at this again properly when I build out the web side of my RaaS offering.
For the Camera? You don't want to know, it's a PDF that's in Chinese that for a C++ SDK in Linux!
For my future web stack the biggest chunk will probably be Daily https://docs.daily.co/reference/daily-js (by @kwindla) as the backbone of our teleoperation orchestration
Daily would be easy to add. We can technically add C++ library, but we don't have nearly enough linters in our agent even for me to trust it, so that's a task for the future!
Please have an option for local processing. I would love to be able to use my locally running Gemma 3n model on my phone for low latency and for them to work without internet connectivity.
We're going to be putting out a Mentra Edge SDK in the next few months, but it comes with some downsides. Using your phone as a compute device is a battery hog, and you can only connect one app to the glasses at a time.
reply