kaijia

you get materially better marketing at the cost of accurately describing the math results in the title and abstract.

I am sick of theory papers that adds human interpretation to connect a mathematical result to some fancy terms that everyone seeks after in real-world ML applications. Ironically, chatGPT is the most productive way to strip off those advertisement so the essence is better exposed. https://chatgpt.com/share/6ae78178-c2ec-83ea-9bc3-fd304c64ec...

I've been using xfce for about five years. I just setup my monthly donation last month and saw this good news today:)

I almost did it with the plain claude code on a $20 plan, without any skills, by leveraging my domain expertise with the following prompt:

  Your goal is to create an HTML5 website that is pixel-accurate as the screenshot @Space-Jam.png . You can use the image files in the `img` folder. To achieve pixel-level reproduction, first write auxiliary programs to measure the pixel locations of the objects on this page. Then use your measurements to create the page. Consider using feature detection/matching with per-object (x, y) offsets and a global scale factor shared by all objects as the free parameters. Consider using RANSAC for robust estimation. After obtaining an initial estimation, crop the image of each object and use local sparse optical flow for refinement. Use JPG format for visual inspection of intermediate results. Your webpage should be able to scale with the window size.

Note: the footer text size/locations are off. We can probably fix that by explicitly asking cc to write scripts for text bounding box detection.

More details here: https://gist.github.com/jia-kai/802de63816711d67f0a090fa267a...

Thanks for this fascinating example! Autosketch is still downloadable ( https://winworldpc.com/product/autosketch/30 ). Then you can unzip it, and

  $ strings disk1.img | grep 'game'
  The object of the game is to get four
  Start a new game and place your first

So if ChatGPT cares to analyze all files on the internet, it should know the correct answer...

(edit: formatting)

Grad student unionization is not only about stipends. You can find more info on MITGSU's website: https://mitgsu.org/why

I dislike this line of research. It just demonstrates large models are capable of memorizing a large number of things, without any understanding.

So I tried the first problem in Appendix A on OpenAI playground.

When I use the prompt "# Sketch the graph of the function f(x) = x + |x|", with the model davinci-codex and a larger response length (other parameters as default), the result seems fine: https://pastebin.com/VT8tPbu6

When I change the prompt to "# Sketch the graph of the function f(x) = x + |x| + x*2", it becomes garbage. It changes the prompt to "# Sketch the graph of the function f(x) = x + |x| + x^2 + x^3 + x^4 + x^5 + x^6 + x^7 + x^8 + x^9 + x^10" and then writes new empty comment lines: https://pastebin.com/2bNEuqaH

Recently I looked into this subject and talked with a postdoc researcher. It is interesting to see this discussion on HN.

C elegans is far more complicated than the connectome abstraction used in these projects. There are all kinds of ion channels, including new ones recently discovered (https://pubmed.ncbi.nlm.nih.gov/34388373/ ), that are regulated by genes and other physical/biochemical factors in the body that are very poorly understood. We only have slightly better understanding of its brain development recently ( https://www.nature.com/articles/s41586-021-03778-8 ). The in vivo research on C. elegans neural system mostly uses the concentration of Ca+ as the neuron activity indicator, which lefts out other details.

According to the postdoc that I talked with, we are still very very far from even accurately understanding a single neuron, let alone simulating any neural network completely.

Edit: typo

Why is there no end-to-end encryption for such IoT devices? Does it mean the central service managing all devices have access to all the video streams?