Bart Wronski 🇺🇦🇵🇸
@bartwr.bsky.social
Engineering, Computer Graphics, Art, DSP, ML
Culture, Techno, Industrial, and Electronic Music.
Research Scientist at NVIDIA.
Ex Google Research, Ex games (Sony, Ubisoft, CD Projekt).
Politically leftist. He/they.
https://linktr.ee/bartwronsk
Culture, Techno, Industrial, and Electronic Music.
Research Scientist at NVIDIA.
Ex Google Research, Ex games (Sony, Ubisoft, CD Projekt).
Politically leftist. He/they.
https://linktr.ee/bartwronsk
Obsidian web page summarizer and link clipper? Procedural MIDI melody/bassline/chord generator? Controller for home automations? A service monitor/runner for all of the above? All a few hours of work. And I get a chance to play with web fullstack and learn it through practical use, zero time wasted.
December 1, 2025 at 3:20 AM
Obsidian web page summarizer and link clipper? Procedural MIDI melody/bassline/chord generator? Controller for home automations? A service monitor/runner for all of the above? All a few hours of work. And I get a chance to play with web fullstack and learn it through practical use, zero time wasted.
Maybe I am spoiled with the Siggraph quality of submissions. But seeing a one-page "paper" with unfinished sentences and LaTeX errors, and reviews that were literally 4 half-sentences in each section, made me lose faith in the academic/publishing process in CV/ML.
November 29, 2025 at 7:02 PM
Maybe I am spoiled with the Siggraph quality of submissions. But seeing a one-page "paper" with unfinished sentences and LaTeX errors, and reviews that were literally 4 half-sentences in each section, made me lose faith in the academic/publishing process in CV/ML.
Solution 2: Every author has to do n_submissions / acceptance_rate * reviewers_per_paper / authors_per_paper reviews.
Solution 3: Not reviewing or low quality (flagged by ACs) = cannot submit next year.
Solution 4: Great reviews by students etc. that find unique insights = free conference entry. 2/2
Solution 3: Not reviewing or low quality (flagged by ACs) = cannot submit next year.
Solution 4: Great reviews by students etc. that find unique insights = free conference entry. 2/2
November 29, 2025 at 6:55 PM
Solution 2: Every author has to do n_submissions / acceptance_rate * reviewers_per_paper / authors_per_paper reviews.
Solution 3: Not reviewing or low quality (flagged by ACs) = cannot submit next year.
Solution 4: Great reviews by students etc. that find unique insights = free conference entry. 2/2
Solution 3: Not reviewing or low quality (flagged by ACs) = cannot submit next year.
Solution 4: Great reviews by students etc. that find unique insights = free conference entry. 2/2
I reviewed for vision conferences only twice. The quality of submissions *and* other reviews was depressing (I even got a best reviewer gift card lol) and decided not to do it again (I don't submit to those, so I hope it's fair).
Solution 1: Limit to max 3 submissions per person, including PIs. 1/
Solution 1: Limit to max 3 submissions per person, including PIs. 1/
November 29, 2025 at 6:55 PM
I reviewed for vision conferences only twice. The quality of submissions *and* other reviews was depressing (I even got a best reviewer gift card lol) and decided not to do it again (I don't submit to those, so I hope it's fair).
Solution 1: Limit to max 3 submissions per person, including PIs. 1/
Solution 1: Limit to max 3 submissions per person, including PIs. 1/
I like the idea of conference workshops to discuss promising not yet fleshed out ideas. But this is the opposite of benchmark-driven review process (and extensive evaluations needed) and should be reserved only for truly novel ideas.
November 29, 2025 at 4:22 PM
I like the idea of conference workshops to discuss promising not yet fleshed out ideas. But this is the opposite of benchmark-driven review process (and extensive evaluations needed) and should be reserved only for truly novel ideas.
I agree it's more rewarding, this is why it should be reserved for the best work, not incremental filler. IMO the bar is too low right now, most of published papers are garbage "we tweaked something and beat a benchmark by a statistical error and don't even do statistical hypothesis testing".
November 29, 2025 at 4:20 PM
I agree it's more rewarding, this is why it should be reserved for the best work, not incremental filler. IMO the bar is too low right now, most of published papers are garbage "we tweaked something and beat a benchmark by a statistical error and don't even do statistical hypothesis testing".
This is the other way around, tail wagging the dog. Dissemination of ideas already happens with arxiv and "preprints". By the conference, papers are already outdated or used by others. :) Rubber stamping should be only for the best work. My solution: stop publishing. Only have "test of time" awards.
November 29, 2025 at 3:38 PM
This is the other way around, tail wagging the dog. Dissemination of ideas already happens with arxiv and "preprints". By the conference, papers are already outdated or used by others. :) Rubber stamping should be only for the best work. My solution: stop publishing. Only have "test of time" awards.
And this is how peer performance reviews worked at Google and I hated it. Someone picks you = you kind of have to write a review and if you write anything negative = you will be hated by that person and their manager. The only honest reviews were manager asking for it verbally in private.
November 29, 2025 at 5:26 AM
And this is how peer performance reviews worked at Google and I hated it. Someone picks you = you kind of have to write a review and if you write anything negative = you will be hated by that person and their manager. The only honest reviews were manager asking for it verbally in private.
I would prefer not to participate in such a process. Reason: let's say a friend (or a coworker, boss, advisor...) designates me to review a paper. What options do I have when the paper is terrible? Withdraw from the review (red flag), write a dishonest one (nope), or offend someone / risk my career.
November 29, 2025 at 5:10 AM
I would prefer not to participate in such a process. Reason: let's say a friend (or a coworker, boss, advisor...) designates me to review a paper. What options do I have when the paper is terrible? Withdraw from the review (red flag), write a dishonest one (nope), or offend someone / risk my career.
Of course, I agree 100%. There are use cases, it's a tool, it's more realistic/accurate.
...But some will be confused with relative desaturation. That their "light is multiplicative" assumptions are violated. They could prefer the one with less color loss in their artwork when illuminated.
...But some will be confused with relative desaturation. That their "light is multiplicative" assumptions are violated. They could prefer the one with less color loss in their artwork when illuminated.
November 24, 2025 at 9:29 AM
Of course, I agree 100%. There are use cases, it's a tool, it's more realistic/accurate.
...But some will be confused with relative desaturation. That their "light is multiplicative" assumptions are violated. They could prefer the one with less color loss in their artwork when illuminated.
...But some will be confused with relative desaturation. That their "light is multiplicative" assumptions are violated. They could prefer the one with less color loss in their artwork when illuminated.
Your spectral results look so much more realistic... by reminding me of pains (and the "impossibility" to some extent) of white-balancing photos taken under very "peaky" warm light.
I hope it's not controversial to say that it might be one of those cases where artists prefer the "wrong" results? :)
I hope it's not controversial to say that it might be one of those cases where artists prefer the "wrong" results? :)
November 24, 2025 at 5:42 AM
Your spectral results look so much more realistic... by reminding me of pains (and the "impossibility" to some extent) of white-balancing photos taken under very "peaky" warm light.
I hope it's not controversial to say that it might be one of those cases where artists prefer the "wrong" results? :)
I hope it's not controversial to say that it might be one of those cases where artists prefer the "wrong" results? :)
they are able to reclaim it. It's morally right and lawful. But the ship has sailed in 2014,the West gave up,including current allies, and it's not going to happen in the next 10y. Only after the fall of Russian rotten empire. Being upset at people who are allies but gave up on Crimea is foolish.2/2
November 21, 2025 at 8:28 PM
they are able to reclaim it. It's morally right and lawful. But the ship has sailed in 2014,the West gave up,including current allies, and it's not going to happen in the next 10y. Only after the fall of Russian rotten empire. Being upset at people who are allies but gave up on Crimea is foolish.2/2
Goalpost was about Ukraine, Russia/USSR/Eastern Europe. He used to be wrong, he apologized, admitted to a mistake, compensated for it financially, and now explains that Russia is a fascist imperialist right-wing empire. So now it's about Crimea?
Myself, I believe Crimea is 100% Ukrainian and wish 1/
Myself, I believe Crimea is 100% Ukrainian and wish 1/
November 21, 2025 at 8:28 PM
Goalpost was about Ukraine, Russia/USSR/Eastern Europe. He used to be wrong, he apologized, admitted to a mistake, compensated for it financially, and now explains that Russia is a fascist imperialist right-wing empire. So now it's about Crimea?
Myself, I believe Crimea is 100% Ukrainian and wish 1/
Myself, I believe Crimea is 100% Ukrainian and wish 1/
Oh nice, a moving goalpost 🙃in the style of "but has he condemned khhhhhammaaas?"
You might know Kamil and his pro-Ukrainian fight on social media, and this other message is as always 100% on point
You might know Kamil and his pro-Ukrainian fight on social media, and this other message is as always 100% on point
November 21, 2025 at 6:47 PM
Oh nice, a moving goalpost 🙃in the style of "but has he condemned khhhhhammaaas?"
You might know Kamil and his pro-Ukrainian fight on social media, and this other message is as always 100% on point
You might know Kamil and his pro-Ukrainian fight on social media, and this other message is as always 100% on point
Which ones and from when? Are you a troll? He supports Ukraine including financially and done more for them than you
November 21, 2025 at 3:39 PM
Which ones and from when? Are you a troll? He supports Ukraine including financially and done more for them than you
Cybotron in the trailer 🤖
November 21, 2025 at 2:00 PM
Cybotron in the trailer 🤖
Obviously they will use racist Sinophobic geopolitical "national interest" arguments or some AI doomer bs, but make no mistake. Who controls (trains etc.) the AI models will be the one effectively making decision and protecting the corporate (or personal!) own self-interest against the humanity.
November 21, 2025 at 2:11 AM
Obviously they will use racist Sinophobic geopolitical "national interest" arguments or some AI doomer bs, but make no mistake. Who controls (trains etc.) the AI models will be the one effectively making decision and protecting the corporate (or personal!) own self-interest against the humanity.
The whole "this is only ephebophilia and ok" which used to be a defining libertarian talking point that never worked, it rightfully disgusts the majority.
Do they think it will suddenly work now - whether on normie parents or siblings of young girls, or on pizzagate-chasing QAnon idiots?
Do they think it will suddenly work now - whether on normie parents or siblings of young girls, or on pizzagate-chasing QAnon idiots?
November 15, 2025 at 8:00 PM
The whole "this is only ephebophilia and ok" which used to be a defining libertarian talking point that never worked, it rightfully disgusts the majority.
Do they think it will suddenly work now - whether on normie parents or siblings of young girls, or on pizzagate-chasing QAnon idiots?
Do they think it will suddenly work now - whether on normie parents or siblings of young girls, or on pizzagate-chasing QAnon idiots?
Functional, imperative, objective, declarative (especially with DSLs). Console apps, web, GUIs. Plotting and data exploration or writing well-tested production reference numerical implementations. Can use numba to get C++-like performance on specific tasks.
Write in the style adequate to the task.
Write in the style adequate to the task.
November 14, 2025 at 12:49 AM
Functional, imperative, objective, declarative (especially with DSLs). Console apps, web, GUIs. Plotting and data exploration or writing well-tested production reference numerical implementations. Can use numba to get C++-like performance on specific tasks.
Write in the style adequate to the task.
Write in the style adequate to the task.
I agree it's a different category. But I got to a strong personal preference of using only a single language for everything, even if it leads to compromises. Context switching got too annoying for me. And the beauty of Python is that it can be sloppy one-liners or pretty, structured, and typed.
November 14, 2025 at 12:49 AM
I agree it's a different category. But I got to a strong personal preference of using only a single language for everything, even if it leads to compromises. Context switching got too annoying for me. And the beauty of Python is that it can be sloppy one-liners or pretty, structured, and typed.