2
points
seletskiy
Joined 209 karma
Want to chat?
Reach me out at s.seletskiy [at] gmail.com
---
meet.hn/city/40.1777112,44.5126233/Yerevan
Socials:
- github.com/seletskiy
- linkedin.com/in/seletskiy
- t.me/seletskiy
Interests:
Hacking, Mentorship, Philosophy, Programming, Research, Robotics, Science, Social Impact, Space Tech, Biotech
---
- seletskiy parentDrop me a line if you would like to talk. Email is in the profile.
- I kind of see your point, but only in the context of working on time-sensitive task which others rely upon. But if it is hobby/educational project, what is wrong doing things by yourself? And resort to decomposing existing solution if you can't figure out why yours is not working?
There's nothing better for understanding something rather than trying to do that "something" from scratch yourself.
- I would say that AI is not to blame here. It just accelerated existing process, but didn't initiate it. We (as a society) started to value quantity over quality some time ago, and, apparently, no-one care enough to change it.
Why tighten the bolts on the airplane's door yourself if you can just outsource it somewhere cheaper (see Boeing crisis)?
Why design and test hundreds of physical and easy-to-use knobs in the car if you can just plug a touchscreen (see Tesla)?
Why write a couple of lines of code if you can just include an `is-odd` library (see bloated npm ecosystem)?
Why figure out how to solve a problem on your own if you can just copy-paste answer from somewhere else (see StackOverflow)?
Why invest time and effort into making a good TV if you can just strap Android OS on a questionable hardware (look in your own house)?
Why run and manage your project on a baremetal server if you can just rent Amazon DynamoDB (see your company)?
Why spend months to find and hire one good engineer if you can just hire ten mediocre ones (see any other company)?
Why spend years educating to identify a tumor on a MRI scans if you can just feed it to a machine learning algorithm (see your hospital)?
What more could I name?
In my take, which you can say is pessimistic, we already passed the peak of civilization as we know it. If we continue business as usual, things will continue to detiorate, more software will fail, more planes will crash, more people will be unemployed, more wars would be started. Yes, decent engineers (or any other decent specialists) will be likely a winners in a short term, but how the future would unfold when there will be less and less of them is a question I leave for the reader.
- To tptacek and other guys who seem to have unwavering trust in OCRs/LLMs, as well as to opposite party who think that technology is not there yet — you are all partially right, but somehow fail to hear each other while also spending time on baseless arguing instead of factual examples and attempts to find common truth.
Can it be used to greatly simplify efforts by getting through boilerplate? — Yes.
Should the result be reviewed and proof-read by human? — Also yes.
---
Here subtle one: https://catalog.archives.gov/id/34384201?objectPage=40
Here is (one of) transcripts made by `o1-pro`:
I'm not native english speaker, but even I can read where it is wrong. I'll leave it to be an excercise for the reader to find out mistakes, but it is certainly not a Teapot trial.(2) …and I don’t know whether it can be reset for a date in December or not. Cornell seemed anxious that it should not come up too close to Christmas, and of course new suspicion [would be aroused?] [about?] him. I will take this up with the Judge as soon as I can get rid of the brief. Meanwhile I would like to know whether there is anything else in which I can be useful to you, since it behooves me in ways of uncomfortable relations with the present management. Are you going East in December? Has any word come from Hagerman? Were there any noteworthy developments at the hearings on the [Teapot?] trial? I have no inclination yet whether Wheeler will be wanted in Washington, but the chances are that he will not. With regards to all the brethren and [flock?], I am very sincerely yours, George A. H. FraserSomehow GPT-4o performs better on this example and fails only on "New Mexican practise" part.
- I am primarily interested in applied projects that make a difference and aim to tangibly reduce human and animal suffering: med R&D, eco, civil infra, pharma or space, but may consider other projects as well.
Location: Armenia (UTC+4) [temporary]
Remote: Yes (flexible working hours including US timezone)
Willing to relocate: Yes, preferably to country with high HDI
Technologies: technology agnostic. Boring and incomplete list of things I have worked with: frontend (React, Vue, Vite/Webpack, Typescript), backend (Go, Node, Python, Java, PHP, Erlang, Clojure), databases (PostgreSQL, MySQL/MariaDB, Redis, Memcached, ES, MongoDB), ops (Linux, *sh, K8s, Docker, VMs, IPMI, Zabbix, Grafana, Prometheus, Nginx, HAProxy, AWS, GCP, baremetal, on-call), ML (coding, [cross-]validating models, deploying, simulated envs), networks (all OSI levels), low-level (C, Ziglang, CUDA, assembly), hardware (MCUs, AVR, I²C, SPI, prototyping).
Résumé/CV: I have ~15 YoE and worked with a number of startups and projects, both as independent contractor and team leader. Check my GitHub to get rough overview: https://github.com/seletskiy and https://github.com/reconquest. Please contact me for more details.
Email: s.seletskiy@gmail.com
- I wrote a bash script [1] with similar functions for my piano, which produces ogg recording and raw midi recording, as well as splitting realtime rendered midi stream (via fluidsynth) into two audiostreams (I've mostly used it for Zoom calls with a piano teacher).
[1] https://github.com/seletskiy/dotfiles/blob/master/bin/piano
15+ YoE. Architect, lead engineer.Location: Armenia Remote: Yes, but flexible Willing to relocate: Yes, to a country with higher HDI Technologies: technology agnostic, everything from microelectronics to Web/GUI frontends Résumé/CV: on request, prefer 1:1 video chat Email: s.seletskiy@gmail.comI would like to join biotech, med, eco, civil infra or aerospace projects that aim to tangibly reduce human and animal suffering and figuring out internal workings of Universe we live in.
Excel in small pragmatic teams with no bureaucracy. Able to provide optimized and highly reliable software.
Not interested in blockchain or GenAI projects.
- I will never understand people who say that mortality is what gives life a meaning. It is exactly opposite. If I can't observe effects of my actions (and most likely "I" would not be able to do so after death), then it does not matter for me what I do during life, since outcome is all the same.
There should be no death. For whatever reason, it is incredibly hard to find people thinking the same, despite, paradoxially no one wants to die.
Can we chat? My e-mail is in the profile.
- Hey. I know _exactly_ what you are talking about.
I've been thinking about that for a quite awhile now, and I guess I have some perspective, but don't have exact answers and I don't think I can fit it into a HN message nor have enough motivation to do so.
_However_, I'll be more than glad to chat personally, maybe we can figure something out together. So, if you¹ want, drop me an e-mail. Address in profile.
--
¹) You, the topic starter or anyone reading this message who finds it relevant.
- Hey. It's nothing very fancy. About 150 lines of C/CUDA code with no deps, including args parsing and logging.
The code runs at steady rate of 18.00-18.40 GH/s at cloud GPU. In fact it's not hashes-per-second, but actually messages-per-second checked.
It launches a 64⁶ kernels in a loop, where each launch checks first two bytes of the SHA of a message concatenated with unique 6-byte nonce per kernel + 6-byte incremental nonce for each launch. There is only one block, so SHA algorithm is heavily trimmed. Also, most of the message is hard-coded, so pre-calculated SHA state is used; it's 10 loops less than needed to encode whole block. Since we only 2 bytes of the hash to check, last two loops also unrolled by hand to exclude parts that we wouldn't need. All code also in big-endian since SHA is, so message hardcoded in big-endian as well.
Base64-encoding is pretty time-consuming, so I've optimized it a bit by reordering the alphabet to be in ASCII-ascending order. I've got to the point where single binary-op optimization can earn 100-500 MH/s speed-up, and I don't really know what else here is remaining.
I don't have RTX4090, so instead I just rented 4090 GPU to run code on. It's less than $0.3 per hour.
I've tried GPT-4 to get some directions for optimization, but every single proposal was useless or straight wrong.
I by no means a C/GPU programmer, so probably it can be optimized much more by someone who more knowledgeable of CUDA.
GPU's are ridiculously fast. It freaks me out that I can compute >18,000,000,000 non-trivial function calls per second.
Anyways, if you want to chat, my e-mail is in the profile.
- Interestingly, I have the ability to increase my resting pulse rate and then decrease it back to normal, without using any specific imagination techniques. I believe this is closely linked to the release of epinephrine into the bloodstream. To achieve this, I simply recall the sensation that arises when my body has the actual epinephrine response. This either tricks my body into responding as if there were an actual epinephrine response, or perhaps a small amount of epinephrine is actually released - I'm not entirely sure. Apart from increased pulse, my pupils also dilate significantly for a short time, which is also can be linked to adrenaline response.
I've had the opportunity to test this in a clinical setting under an ECG. Not only did it increase my pulse, but it also caused the QRS complex to invert. Upon seeing this, the doctors advised me not to continue with this practice. However, I didn't experience any negative effects from this experiment. On the flip side, I haven't found any practical use for this ability either.
I would be intrigued to connect with someone else who has a similar capability. It seems that those internal "feelings" is as close to direct control as we can get.
As of pain, I have similar experience as well, except instead of detaching myself from the pain, I "look" as closely as I can. At a certain point pain decomposes to what it really is — electrical impulses, and from this point it literally starting to feel as electricity going through your body, quite the same feeling as if you accidentally grab both pins of the electrical plug, albeit not as intense.
- Hey, I've realized the same thing (that my workflows are stack-based) awhile ago, but didn't get to the point of writing a tool yet. Dare to share?
Also, which approach you use to efficiently store and re-store relevant context information? I often find that intricate but important details are lost during context switch.
- I guess it is a matter of experience and practice, as with anything else, really.
There is a very subtle mental state between "I am too aware that I am dreaming" and "there is no awareness of dreaming at all". I don't really know how to put it in words, but it seems that you BOTH need to supress part of the brain that wakes you up and prevent loosing awareness at the same time. It is very apparent during "ordinary" falling asleep, when you suddenly catch yourself that you are "seeing pictures" (hypnagogia phase) and become fully awake again. The trick is continue "falling asleep" without loosing awareness. Same applies during the lucid dream. You need to maintain balance.
You basically need to continue experimenting to notice those subtle changes to know which mental state would wake you up and which would not, so you'll get more precise control.
There are other factors at play for sure, like if you will manage to get your lucid dream right after first deep phase of sleep, your body would likely be not rested enough to quickly reach wakefulness, so you'll have more time.
I don't think that lucid dreaming is easy, and it was never easy for me. It was actually pretty hard work. It was almost impossible to get lucid dream if I was already mentally exhausted during the day. As soon as I stopped to practice, lucid dreams stopped too.
- It is a common problem when you starting to get lucid dreams. There are number of techniques to combat that too, but it will generally go away after you become used to it. You just get too excited when you get lucid dream.
Maybe sound absurd, one technique that worked for me is to just "start spinning" around in your dream, so you can't focus on any specific part of the dream for a long time. Continuous attention to one detail in a dream somehow starts to breaking it apart.
- Yeah. Read "Exploring the World of Lucid Dreaming" book by Stephen LaBerge. It proposes a number of techniques, some of them works better than other for different people. If you are not predisposed to lucid dreaming (some people are), then the main obstacle is will or motivation to keep practicing.
One that worked for me is having a mechanical counter of some sort (like lap couter or just miniature code lock so you can use it as a counter), which you reset every morning and increment during the day every time after you ask yourself "Am I dreaming?" and doing some basic "reality check" (e.g. trying to breathe in through your nose when you pinch it, or checking clocks twice in a succession and validating that it makes sense, or just looking at your hands). The goal is to consistently do a number of reality checks during your waking life, which inevitably will increase the chance that you will do it in your dream. That will likely would cause you to wake up immediately for couple of times, but then you'll get used to it. It works simply because "if you don't ask yourself that you are dreaming during your waking life, what is the chance that you'll ask yourself it during sleep?".
There are number of other and more advanced techniques that borderline with "magic" to me, like falling asleep without loosing conciousness (and then waking up the same way), but I had this experience only a couple of times I guess.
- With the new advancements in AI, it's expected that there will be more code being generated by programs rather than written by programmers. Making sure these programs deliver accurate and error free code is a difficult task. The challenge here lies in effectively proofreading AIs so that any bugs are caught before they reach end users. Good tools such as unit tests can help but may not be enough to make sure AI generated code is flawless. As the use of AI increases, we must look to develop better methods for ensuring quality assurance on these programs if they are to truly be useful.
- 'seletskiy/godiff' author here.
In my judgment, the problem has been overstated. At the very least, with the 'tea,' it could be easily resolved by contacting me directly via e-mail, as mentioned in my GitHub profile.
I never intended for this code to be used by anyone else (there is no even a README file). But I'm delighted it helped someone attain their goals.
Anyway, I've stated that my code is MIT licensed.
- As another Arch Linux user, I can attest that it is the rock-solid foundation for your computer.
I use double boot to host both Linux and Windows; when I need to use Windows, I just put Linux into hibernation. This greatly extends the amount of time it can go without being rebooted.
Driver and X problems still cropped up occasionally, but things seem much more reliable now than they did a while back.
This is a also the distro I'm familiar with from my time working on servers, where I've used it with ZFS with great success.
- There is octo.nvim [1] which brings rich integration of GitHub's features directly into nvim.
- Also, some obscure yet very helpful ZFS feature is that all your ZFS datasets have hidden (e.g. not visible via `ls -a`) `.zfs` directory in their roots. This directory contains all snapshots of a given ZFS dataset mounted as directories.
For example, if your home is a separate dataset, then `/home/operator/.zfs/` will contain all snapshots automatically mounted for your convenience.
- Shameless plug — https://github.com/reconquest/zeus
Daemon for automatic incremental backup of your ZFS systems to the another (physically connected right now) pool.
- Some time ago I wrote zsh helper [1] that uses jo to quickly construct complex JSON requests, mainly for testing and quering services from console.
Paired with httpie [2] aliases [3] it produces concise APL-like syntax:
Which translates to:POST https://httpbin.org/post test:=j`a=b c=`e=3` l=`*1 2 3``
Or, in other words, sending POST with the following body:http POST https://httpbin.org/post test:="$(jo a=b c="$(jo e=3)" l="$(jo -a 1 2 3)")"
[1]: https://github.com/seletskiy/dotfiles/blob/78ac45c01bdf019ae... [2]: https://httpie.io/ [3]: https://github.com/seletskiy/dotfiles/blob/78ac45c01bdf019ae...{"a":"b","c":{"e":3},"l":[1,2,3]}