One of the strengths of that original OpenAI group was recruiting. Somehow you managed to corner the market on a ton of the top AI research talent, often with much less money to offer than your competitors. What was the pitch?
The pitch was just come build AGI. And the reason it worked—I cannot overstate how heretical it was at the time to say we’re gonna build AGI. So you filter out 99% of the world, and you only get the really talented, original thinkers. And that’s really powerful. If you’re doing the same thing everybody else is doing, if you’re building, like, the 10,000th photo-sharing app? Really hard to recruit talent.
OpenAI senior executives at the company’s headquarters in San Francisco on March 13, 2023, from left: Sam Altman, chief executive officer; Mira Murati, chief technology officer; Greg Brockman, president; and Ilya Sutskever, chief scientist. Photographer: Jim Wilson/The New York Times
An efficient differentiable mesh-based method that can effectively handle complex 2D and 3D shapes. For instance, it can be used for reconstructing complex shapes from point clouds and multi-view images.
Some smaller open-weights AI language models (such as Llama 3.1 70B, with 70 billion parameters) and various AI image-synthesis models like Flux.1 dev (12 billion parameters) could probably run comfortably on Project DIGITS, but larger open models like Llama 3.1 405B, with 405 billion parameters, may not. Given the recent explosion of smaller AI models, a creative developer could likely run quite a few interesting models on the unit.
DIGITS’ 128GB of unified RAM is notable because a high-power consumer GPU like the RTX 4090 has only 24GB of VRAM. Memory serves as a hard limit on AI model parameter size, and more memory makes room for running larger local AI models.
Spectral sensitivity of eye is influenced by light intensity. And the light intensity determines the level of activity of cones cell and rod cell. This is the main characteristic of human vision. Sensitivity to individual colors, in other words, wavelengths of the light spectrum, is explained by the RGB (red-green-blue) theory. This theory assumed that there are three kinds of cones. It’s selectively sensitive to red (700-630 nm), green (560-500 nm), and blue (490-450 nm) light. And their mutual interaction allow to perceive all colors of the spectrum.
RASTERIZATION Rasterisation (or rasterization) is the task of taking the information described in a vector graphics format OR the vertices of triangles making 3D shapes and converting them into a raster image (a series of pixels, dots or lines, which, when displayed together, create the image which was represented via shapes), or in other words “rasterizing” vectors or 3D models onto a 2D plane for display on a computer screen.
For each triangle of a 3D shape, you project the corners of the triangle on the virtual screen with some math (projective geometry). Then you have the position of the 3 corners of the triangle on the pixel screen. Those 3 points have texture coordinates, so you know where in the texture are the 3 corners. The cost is proportional to the number of triangles, and is only a little bit affected by the screen resolution.
In computer graphics, a raster graphics orbitmap image is a dot matrix data structure that represents a generally rectangular grid of pixels (points of color), viewable via a monitor, paper, or other display medium.
With rasterization, objects on the screen are created from a mesh of virtual triangles, or polygons, that create 3D models of objects. A lot of information is associated with each vertex, including its position in space, as well as information about color, texture and its “normal,” which is used to determine the way the surface of an object is facing.
Computers then convert the triangles of the 3D models into pixels, or dots, on a 2D screen. Each pixel can be assigned an initial color value from the data stored in the triangle vertices.
Further pixel processing or “shading,” including changing pixel color based on how lights in the scene hit the pixel, and applying one or more textures to the pixel, combine to generate the final color applied to a pixel.
The main advantage of rasterization is its speed. However, rasterization is simply the process of computing the mapping from scene geometry to pixels and does not prescribe a particular way to compute the color of those pixels. So it cannot take shading, especially the physical light, into account and it cannot promise to get a photorealistic output. That’s a big limitation of rasterization.
There are also multiple problems:
If you have two triangles one is behind the other, you will draw twice all the pixels. you only keep the pixel from the triangle that is closer to you (Z-buffer), but you still do the work twice.
The borders of your triangles are jagged as it is hard to know if a pixel is in the triangle or out. You can do some smoothing on those, that is anti-aliasing.
You have to handle every triangles (including the ones behind you) and then see that they do not touch the screen at all. (we have techniques to mitigate this where we only look at triangles that are in the field of view)
Transparency is hard to handle (you can’t just do an average of the color of overlapping transparent triangles, you have to do it in the right order)
