HNHacker News
TopNewBestAskShowJobs

heyitsguay

1,081 karma · joined January 5, 2017

submissionscomments
heyitsguay··on Chaos Ink
If you add a `uniform vec3 mousePrevFrame` that tracks the previous mouse position, you can get smoother drawing in main() by replacing your distance function in the line `float dist = distance(mouse.xy, gl_FragCoord.xy)` with a custom

  distanceToLine(mousePrevFrame.xy, mouse.xy, gl_FragCoord.xy) 
defined as

  float distanceToLine(vec2 p1, vec2 p2, vec2 p0) {
    float d12 = distance(p1, p2);
    if (d12 == 0) {
      return distance(p1, p0);
    }
    return abs((p2.y - p1.y) * p0.x - (p2.x - p1.x) * p0.y + p2.x * p1.y - p2.y * p1.x) / d12;
    }
heyitsguay··on Chaos Ink
Amazing! Something I have learned from similar experiments in the past - instead of drawing a circle each frame, try drawing a (rounded) line between the current frame's cursor location and the previous frame's cursor location. Cursor movement looks much smoother that way.
heyitsguay··on D-Wave announces general availability of first quantum computer for business
Do you have any case studies with partners?
heyitsguay··on Water on Mars: discovery of three buried lakes intrigues scientists
Very cool! What's the story there?
heyitsguay··on 'Close to 100% accuracy': Helsinki airport uses sniffer dogs to detect Covid
I wouldn't even say the reported stats are misleading, just that "accuracy" in a headline is too vague to ascertain performance when you're trying to detect (relatively) rare events.
heyitsguay··on Why neural networks struggle with the Game of Life
Depends on how you do your validation, basically. It's easier to give an example if we assume m_1 is your worst model and m_n is your best: if you do a train/eval split, train m_1, eval m_1, make some tweaks to m_1 to improve that eval score and call the tweaked model m_2, etc., you'll implicitly be overfitting your better models to the eval data even though you're never actually training on it. In which case, yes, it is quite possible that given some new unrelated data, m_n will do no better than m_1 or m_2 or whatever.

If you do a train/eval/test data split instead, train+eval m_1 through m_n like before, and then rank them based on performance on the test data only after model selection is finished, you won't have that implicit overfitting to the test data, and m_n stands a better chance of beating m_1 on new datasets.

There's another way things can go wrong - if your train/eval/test data doesn't capture the full range of variability of the data source that will be producing your fresh datasets, then even if you do the train/eval/test split correctly, it's quite possible that m_n will do no better than m_1 when operating on fresh data outside of the train/eval/test distribution.

Basically, it is possible to build a sequence of models that actually improve performance on new data, but there are also lots of ways to get it wrong and produce models that look like they're doing better in a controlled environment, but fail to do better in the wild.

heyitsguay··on 'Close to 100% accuracy': Helsinki airport uses sniffer dogs to detect Covid
Since active covid cases are still pretty rare day-to-day at a population level, "Close to 100% accuracy" is ambiguous, and could be very good or basically useless. For example, if 3% of the people moving through that airport actually have coronavirus, a detector that just reads negative for everything would have 97% accuracy - but a 100% false negative rate.
heyitsguay··on Cloth masks do protect the wearer
I'm curious about this too. I'm no epidemiologist, but as I've kept walking around my low-to-mid-density DC neighborhood I've wondered, if an exposure level likely to cause symptomatic illness is on the order of a minute's close contact with a sick person, then maybe it is good to get a second or two of exposure per week?
heyitsguay··on GPT-3 has no idea what it’s talking about
Really? This seems like the exact sort of thing GPT-3 would be very bad at. The statements aren't the product of an internally consistent logical model, they are statistically plausible word sequences learned in a really clever way from a huge corpus of text.

That said, before dismissing it sight unseen there are effective and quick ways to test its powers with mathematics. You could prime an instance with the axioms defining a group, or a topology, and ask it to produce proofs of varying difficulty from intro abstract algebra or topology textbooks. Whatever the answer it would make an interesting blog post. What happens when you ask it to generate repeated outputs for the same prompt? What happens if you ask it to prove an untrue proposition?

heyitsguay··on Open Problems in Robotics
I know that there's a lot of successful work in specific controlled environments (company X's factory floor), and that a lot of environment understanding/SLAM is broadly unsolved in arbitrary uncontrolled environments, but what about specific uncontrolled environments? What if i want a beer serving robot to learn a consistent, high-accuracy model of just my house, the way it is, and I'm willing to put in some technical work? Can i do any better than completely hand crafting a 3D object model or similar?
heyitsguay··on Amazon Liable for Defective Third-Party Products Rules CA Appellate Court
Better alignment of incentives. The retailer (especially for a huge company like Amazon) has more resources to pursue manufacturers with defective products, they have more information about their supply chain which is useful for issues with international manufacturers, and putting the onus on the retailer encourages the retailer to not allow shoddy or counterfeit products into their marketplace.
heyitsguay··on Atlassian tells employees they can work from home forever
Good point! I can try but it's hit or miss whether it'll get any traction. I guess a couple questions you could help with are (1) are there any good resources for scoping the job market beyond putting "[whatever] jobs stockholm" into search engines, and (2) do you have any do's or don'ts for applying as an American not currently living in Sweden?

Thanks a lot for your perspective!

heyitsguay··on Atlassian tells employees they can work from home forever
This is a total tangent, but I'm an American with a partner from Stockholm, and we're interested in maybe moving over there when that is possible again. Do you have any suggestions or resources for getting to understand the tech scene around Stockholm, especially AI/computer vision stuff? Thanks!
heyitsguay··on NeRF in the Wild: reconstructing 3D scenes from internet photography
That's very cool, but also makes it sound more challenging to integrate into the existing 3D modeling ecosystem vs, say, photogrammetry approaches. Is it possible to generate approximate textured meshes from the color and opacity information?

[Edit] After a little Googling I do see this has been done, using marching cubes (https://www.matthewtancik.com/nerf).

heyitsguay··on Silicon Valley VC says Ghislaine Maxwell attended an investor party in 2011
I'm not sure consent has or needs a technical solution, but that one definitely wouldn't work.
heyitsguay··on OpenNMT: Open-Source Neural Machine Translation with Torch Mathematical Toolkit
Looks cool, but it hasn't been updated in a bit and as they note, the Torch framework is no longer maintained. There is a PyTorch alternative that appears more active: https://github.com/OpenNMT/OpenNMT-py.
heyitsguay··on Show HN: Browse recently expired, pronounceable domain names (Part II)
Grassroots artisan cheese distribution?
heyitsguay··on Holographic optics for thin and lightweight virtual reality
So who's doing interesting applied work in scene understanding for AR?
heyitsguay··on Sightbit deploys AI on beaches to help lifeguards spot distressed swimmers
This makes me wonder - is it possible to make a "smart camera" with an open-source circuit diagram and laymen-interpretable circuit verification indicators? Like suppose you want to sell a CCTV system for this exact situation, that's got a pretrained detection model implemented on some inference circuitry hooked up to a streaming camera. You publish the circuit diagram and any other operating specifications, and you want to convince people that the physical device they have in their hands is impossible (or at least very expensive) to alter, say by adding a hard drive or wireless modem or whatever. Are there sensors you could attach to the circuit such that modifications to what is interconnected would manifest as deviations from published sensor values? Some sort of md5 hash analog for hardware?

That all assumes you've solved the problem of training a good enough detector (presumably from a similar dataset) which has its own difficulties, but OP's question made me wonder about the aspect i described above.

heyitsguay··on A $43,859.00 Refurbished Mac Pro
Seems pretty pricey given the specs. How much would it cost to put together a new Linux box with the same/functionally equivalent components?
heyitsguay··on Responding to the Controversy about YOLOv5
YOLOv5 wasn't a Roboflow project, they just provided an implementation and used the name that the project creator used.
heyitsguay··on YOLOv5: State-of-the-art object detection at 140 FPS
Very cool business model! How long have you been at it? I've been pushing for a while (unsuccessfully, so far) for the NIH to cultivate a team providing such a service to our many biomedical imaging labs. It seems pretty clear to me that this sort of AI hub model is going to win out in at least the medium term versus spending money on lots of small redundant AI teams each dedicated to a single project. What sort of application sectors have you found success with?
heyitsguay··on Deep learning of physical laws from scarce data
Maybe. The challenge you run into for devising radically new theory is that the optimization problem being solved in this work takes place within a broad but still-constrained "theory space". Specifically, from my quick read of it it looks like they're inferring components of a PDE of the form u_t = ϕL, where ϕ is a large dictionary of possible functions of u and its spatial derivatives, and L is a sparse coefficient matrix. This form covers a really broad class of PDEs, but for example, whatever theory ends up unifying QM and gravity probably won't be a fancier PDE of this form that we haven't found yet.
heyitsguay··on Shan, Shui: Procedurally generated Chinese landscape painting
Amazing! I wonder if there is any space for performance optimizations to enable the initial scroll rate to continue indefinitely?
heyitsguay··on Uber is destroying thousands of electric bikes and scooters following Jump sale
Same here! There was a brief golden time in Mt. Pleasant when it was starting to get nice out and there were Jump bikes everywhere. So convenient, so nice for doing socially-distanced wanderings of neighborhoods. What a waste! I hope something like them returns.
heyitsguay··on Jack Dorsey Responds to Trump
Presumably he believes that this sort of moderation is really, really tricky to get right, and from both an ethical and PR perspective it may be best to apply it to only the most egregious and potentially damaging cases.
heyitsguay··on Retrofuturism
Steampunk?
heyitsguay··on All the “reopen $statename” websites were registered on the same day
It doesn't need to be, but if you have your own interests in mind (economic, anti-US, others?), then why not? Since the rise of mass social media we've learned it's pretty easy to mindfuck a bunch of rubes, and if you get caught, the rubes will double down because nobody's ego will ever take the hit of realizing they're the rube.
heyitsguay··on The Wolfram Physics Project
You're in a position to have a different perspective of course, but that doesn't track with what I've seen. It doesn't seem like anyone has been obsessed with him, in this thread or in ones I've seen before here. Seems like people just point out he's a grandiose jerk when he comes out with grandiose, non-peer-reviewed scientific theories. The depth of their emotional involvement doesn't appear to go much deeper than that, it just comes up commonly because (a) it's commonly known but (b) Wolfram has enough clout to semi-commonly publish stuff that trends on HN.
heyitsguay··on Amazon fires two UX designers critical of warehouse working conditions
Nobody's getting fired for suggesting the top marginal tax rate should be lower, or expressing the value of hard work and individual responsibility. The apparent phenomenon you allude to is the result of the normalization of sexist, racist, and xenophobic beliefs within the "conservative" sphere over the past few decades. Those beliefs are incompatible with being a member of a pluralistic workforce.
← PreviousPage 11 of 15Next →