No, also works on new projects. We do recommend you set up "standards" first though, a documentation of how you want things to be done, for example how you do auth, how you handle multi-tenancy, how migrations work, whatever applies to your new project. The whole point of spec-driven is not not let AI wing it, but be very prescriptive.
We include a couple of templates to make that easier - NextJs + Convex + ShadCn/ui, NextJs + Supabase + ShadCn/ui, etc.
@Gavriel if you're here, have you looked into filing a trademark for NanoClaw? Once you have a registered mark (or even a pending application), you get much stronger leverage with domain registrars, Cloudflare, and Google for takedowns.
UDRP disputes become straightforward when you can show the other party registered a domain using your mark in bad faith. It won't fix the Google ranking overnight, but it gives you real legal teeth beyond just SEO whack-a-mole.
The solution is zk verifiable credentials, which would let folks prove their age without revealing their DOB (or anything else on their ID) to third parties.
This is possible today with complete privacy for people with biometric IDs and biometric passports (ie most passports, EU IDs, Aadhaar IDs, and more) using a service like self.xyz
A while back, I had a big paper deadline a week away and knew I didn’t have enough time to finish without sacrificing sleep.
Rather than cutting my sleep short, I decided to stick with 7–8 hours of rest and instead lengthen my wake window. I worked out a schedule that gave me six nights of sleep across seven days. It meant waking up at stranger and stranger times as the week went on, and getting some odd looks from my roommates when I emerged from my room. But in the end, it was totally worth it. I was waking up well-rested and ready to tackle those extra-long days.
The effort paid off 100%. Not only did I make the deadline, but my paper was accepted as well. A year later, that same paper helped me get into my PhD program of choice.
It’s funny how these short bursts of intense effort can sometimes have such a big impact.
Worked for me, too, in similar circumstances. 30-32 hours with 8-9 hours of sleep and 21-24 hours of activity is a somehow better sleep/activity ratio, closer to 1/4 than to 1/3. More important for me were longer stretches of uninterrupted concentration time. Likely it's not important in the vibe-coding case.
I'm surprised to hear that that schedule worked well. Even if you woke feeling well-rested, what did you feel like towards the end of the day, at your normal/previous sleep time?
Personally in that situation I would (and do) get plenty of sleep every night and then skip the final night. I find the fatigue from a lot of lost sleep normally doesn't hit me in full until the second day after, and the final-day panic is enough to counteract the lack of sleep.
I’ve tried this because I’ve long thought my circadian cycle is just a bit longer than average, where I just don’t feel tired until late at night, and traveling westward feels great because I can take an extra long day. Waking up early on a regular schedule is hard for me generally, like, I can do it but I feel braindead. Eastward jet lag of 3-6 hours is awful if I have to do stuff in the mornings.
So I tried a complete “cycle” of 28 hour days until I realigned with the normal day. Which happens to be the same thing GP did. LCM(24, 28) = 168 which is 7 days with 6 cycles. Roughly 19 hours waking and 9 hours sleeping. I did let myself sleep longer instead of holding to 7-8 hours like GP.
It definitely felt weird because my wife wasn’t on the schedule, but I didn’t feel super sleep deprived. I’m sure with multiple complete cycles you’d see more adverse effects, so it’s probably best to only do this very sparingly. Maybe napping during the wake period could alleviate issues.
What a fascinating and detailed post. Thanks so much for sharing!
I wonder whether it's possible to achieve similar performance but with a simpler design that uses scalable reader-writer locks. They can be made a lot more efficient than what's commonly out there using the C-SNZI lock-free data structure:
If anyone's interested in this general subject matter, a while back I did some academic research on highly scalable locks where we came up with some very high performance reader-writer locks:
I love this. Many a moon ago, I worked on a system called Aikido at MIT, which combined a special built hypervisor with a binary rewriting system (DynamoRio) to enable efficient time travel debugging and race detection of parallel applications.
If anyone's interested, here's a publication that talks about it in more detail:
The use of performance counters here also reminds me of another project I worked on called Kendo, which was a posix thread like replacement that used performance counters to enforce a deterministic interleaving of synchronization operations (mutexes, etc). The system could guarantee determinism for programs that didn't have race conditions. Back then, I found that counting instructions wasn't deterministic on the processors of the time, but counting store operations was. If anyone's interested in that work, here's the publication:
Not right now. Sounds like they are working on lenses that could one day work with colored light for cameras. Maybe after that, they could be used for specials?
This article describes how nano metalenses work and what they can be used for.
Spoiler alert: they can't be used to replace your smartphone's camera lenses yet, but can be used for IR distance sensors used on drones and soon, polarization sensors that will be able to tell materials apart and even detect cracks in concrete.
AI-equipped cars with a real sense of feeling in their tires. It’ll be like the sensation of touch for them. Everything they run over, they will feel. Make them learn that animals feel gross to touch. Make them feel remorse for running over a human. Boom! Self-driving safety solved!
I feel like this would be much easier to accomplish with acoustic sensors and wheel/suspension telemetry, given that an attentive human driver can sort of "feel" the road surface already.
Does this only work on existing codebases?