jclarkcom

I have a node middleman that proxies request from an HTML/JS front end to a native cuda process using web sockets. To support multiple windows, the node process process provides communication between two browser windows. This lets me have render a model using 3JS in one window and a ray traced version in another window.

I've been enjoying the breadth of projects made possible with AI, I've cataloged over 200 of them created in 2025 here: https://jonathanclark.com/posts/ai-coding-million-lines-2025...

A few of my recent favorites: - swim lap counter in html/JS that uses the camera to watch you swim and count laps/timing - video recorder that records your window/desktop and uploads a file to S3 - video conferencing app that allows a 2 year to click on a family member face and initiates a video conference using webRTC, STUN, and browser audio/video capture with automatic bandwidth adjustments (works on all platforms with pure HTML/JS). - CUDA based ray tracer with HTML UI that can trace over 2m rays per second on my laptop for scientific study, allowing real-time display of optical parts. - chat front-end for image models like gemini-pro and openai that take other images and text as references and generate a big library of options to chose from in seconds, I've been using photoshop for decades but I tend to use this more now.

looks that way to me. However, that isn't necessarily a bad thing if a human reviewed and guided AI to get the article they want.

When humans are in the loop everything pretty much becomes stochastic as well. What matters more is the error rate and result correctness. I think this shifts the focus towards test cases, measurement, and outcome.

I implemented an assembler as part of VMware thinapp and this was big performance boost for me but maybe you have a different experience from your efforts?

I was trying to provide a more detailed explanation without typing a lot. I studied this problem a lot as PE at vmware.

More detailed explanation from ChatGPT. As quick estimate you could achieve a >2x speed up using memory mapped files for a typical assembler workload.

https://chatgpt.com/share/68b5e0db-a6d0-8005-9101-d326d2af0a...

Windows uses memory mapped IO for loading all executable processes because it allows you to start executing a process after loading a few pages even if the exe is megabytes. You can use the same to reduce latency for starting to assemble data before the rest of the file loads, the rest can be loaded using more efficienct asynchronous mechanisms. Using for output also means your process doesnt waits on flushes that is also async. And in memory constrained environments the OS doesn’t have to write your data to swap, it can just reload it from the meeting mapped file.

You might look into using memory mapped IO for reading input and writing your output files. This can save some memory allocations and file read and write times. I did this with a project where I got more than 10x speed up. For many cases file IO is going to be your bottleneck.

Very cool. We are launching a sensor that mounts on the inside of your window and measure the sky color for a small cone of the sky and transmits this to our skylight and window fixtures inside (see innerscene.com) so they can replicate exactly the same thing indoors. You could potentially use a computer monitor to do this, but it generally doesn't provide great light due to using RGBs instead of wide-spectrum sources.

One issue with the current code is it doesn’t model clouds, haze, or smoke so the rendered sky can differ from what you see outside (numerous HN comments notice this). You can partially correct for this by using semi-realtime satellite imaging but hard to get super accurate which is what pushed us to develop our own sensor. There are various CCT sensors on the market already but they only measure directional+diffuse+reflected light which is typically ~7500k but the sky color goes up to 40,000k.

Here is a plot showing the color of the sky as it changes during the day from real sensor readings. Each one is 30s apart, so it change change quickly. https://www.innerscene.com/built_pages/cs_specsheet/cct/cct_...

A bit more info as well: https://www.innerscene.com/SpecHelp/CircadianSky/cct/cct.htm...

For broadband IR, using a gas IR heater will give you the cheapest output - followed by an electric heater. Hard to compete with devices design specifically for heating when you are trying to do fancy optics in a compact form factor at the same time.

Yuji is similar to a lot of Chinese brands doing something similar which is a backlit Raileigh Scattering panel. The show images of the sun in their marketing and sharp sunbeams on the wall, but these are complete fiction. The also advertise color tuning, but the only natural color they can produce is a blue because Rayleigh Scattering doesn't allow very good color control. Still, for some applications like wall washing if you don't need a dynamic sky color and you can hide the view of the sun it could be a reasonable option.

We haven't researched plants a ton, but did some test to confirm they can grow under this light. Here is a cool time lapse video showing this in action: https://www.youtube.com/watch?v=2TDIVnXfE9I

UL safety cortication and energy regulations make it difficult to add UV and IR, UV because it can easily damage people's eyes and IR because it introduces too much heat which can be a fire or touch/burn risk. Energy regulations also limit how much wattage you can put into lighting, if most of that energy goes into creating IR (like real sunlight) then you won't comply with those regulations. For now it's better to buy a separate space heater to create IR, then you could also use a cheaper energy source like gas.

Yes, there is some really interesting research on this topic. It turns out that if blue light centered around 490nm is an order of magnitude more effective at treating SAD.

https://www.innerscene.com/papers/lux-vs-wavelength-in-light...

Because we recreate the blue sky there is a lot of light in this spectrum

how does this compare with heat dissipated per square cm when you compare with desktop CPUs?

Yes! We started out building light field displays and originally used digital displays (arrays of micro LCD screens) - the result was super awesome but had a long road to becoming becoming a commercially viable product so pivoted to analog use cases which are a lot more energy efficient meaning we could achieve a lot brighter output.

Thanks for the feedback! We’ll see if we can improve that image.

Thanks. This is our own custom PDF viewer. I made the page by generating images from a PDF and then overlaying transparent text so you can search and copy/paste. This allows us to make the PDF much more interactive (I.e. help topics, forms, etc) and allows us to create URL that link to specific pages or topics. You can actually save the page locally as a single file and open it locally and it still works, similar to a PDF. There is a script that bundles all the content into a single page as base64 data. As HtMl it loads a lot faster than a PDF and generally smaller. 10 years ago this would have been a major undertaking to create but with AI and rich libraries available today it was just a couple of days work.

its tricky to compare because ours are designed to be integrated into a space and create the appearance of window with a big focus on high CCTs (2200k-40,000k) which relocates the color and spectrum of a blue sky using up to 180W. This lamp seems to be focused on lower CCTs (1800k-5000k) at high powers (750w is a lot of heat to dissipate). Both have similar Rf and CRI but we maintain great Rg and CRI across the whole range.

The one you linked is a stand alone lamp and they seem to focus on red light which I don’t quite understand. Red light (or very low CCTs) is used for skin therapy but typically requires direct exposure to the skin because irradiance levels need to be super high for that - from my understanding you wouldn’t have any benefit from a red lamp even at 750w. But blue light does can have lots of health benefits at low levels (search for ipRGCs).

Ah! :) :)

Yes! We just announced and demoed a sensor that does just this at LEDucation last week and will be shipping before the end be of the year. It’s called SkySync

I don’t understand, is this a general jab at Oracle or was there something specific to this topic?

Outside direct sunlight can be 100,000 lux. With one fixture if you are standing in the sunbeam you are more likely to be at 2000 lux which is a long way off. You can add more fixtures to increase the light you are getting. However it turns out that one of the most important parts of natural light is “melanopic lux” which comes from light centered around 490nm (blue), if you get more or this spectrum it can have the same effect as having a lot more light from a low blue Light fixture. (I.e typical 2700k bulb) Some info about this; https://www.innerscene.com/products/circadian-sky/CircadianS... There is now a lot of research that looks into how much of this blue light you need and architects are starting to incorporate this into Building designs.

Yes, the more LEDs you mix the closer you can get to your target SPD but more LEDs also add cost and may not fit mechanically depending on how you are mixing them.

One other variable is energy efficiency, there are a lot energy codes around the world that limit how much energy you are allowed to use to light a space. In California for example it can be 0.6w per sqft. Sunlight is more like 100w/sqft so you end up having to optimize the spectrum for what humans can see and feel. I like to compare it with jpeg compression for light. In jpeg we throw away components you can’t see very well, you can do the same for light for an energy efficiency sake and maintain a close perceptual proxy.

We found 4 chips is a minimum bar for good light and use 5 for virtual sun - more info: https://www.innerscene.com/products/circadian-sky/CircadianS...

You might want to test for mold as well

there are companies that can do custom phosphor formulations for you to target a specific output. The minimum order quantities don’t make it practical for DIY but not too bad for a small startup. Our approach is to mix a bunch of different LEDs together to get the color and spectrum we want. Check out telelumen.com for an example that uses 16 chips. These are designed for researchers

Yes! Good eye. They installed a pretty early version of our product - I think it took 4-5 years for the episode to finally come out

Not a bunker per say but I recently learned that in some states/cities you don’t pay property taxes on square footage below ground so you end up with people building more space below than above ground and Virtual Sun can make it feel like above ground space. Just visited a billionaire project last week where this was the case.

Hmm, never heard that before, I’ll have to research it more. Maybe the UV from sunlight is killing something you are allergic to. We don’t have UV for safety and energy efficiency reasons but you could try buying some strong UV lamps and see if that helps. Ideally you turn them on when you are out of the room so they don’t damage your eyes or cause cancer.

here is an example install, this is not a render (the window is real but the skylight is not) https://www.innerscene.com/gallery/residential-virtual-sun

Here is a video showing the units suspended in the air https://www.dropbox.com/scl/fi/e0p4rsx31cqag42cklc61/A7_free...