The difference is that raw files are scene-referred and jpeg files are output-referred.
To help those who are barely aware of color management and read that as inscrutable jargon, it might be helpful if you explain what those two terms mean and why understanding the difference is important.
When software works with JPEG (or HEIF etc.), it decodes the RGB values for each pixel, and those values refer to the output device (monitor, TV, printer).
In the raw files, the data refers to the light captured from the real scene.
To me this "referred" business is still jargon and still hard to make sense of. It seems arbitrary.
You have per-pixel numbers in R,G,G,B channels in raw and per-pixel R,G,B numbers in jpeg. What do those numbers represent?
In a raw file the numbers measure, roughly speaking, incoming photons from the scene. So it's a digitally encoded representation of a real scene (technically, the scene light + lens + Bayer filter + noise).
In a jpeg file the numbers represent the brightness of pixels on a monitor.
You say that the decoded RGB values in a jpeg are "referred" or "refer" (two different voices) to the output device whereas the raw data "refers" to the light captured from the real scene. But the data in the jpeg also "refers" to the light captured from the real scene. That's why it looks like, for example, the cat. And the raw data, when viewed in a raw viewer, also "refers" to the monitor.
Yes, but in the processing chain, raw data is much closer to the scene than RGB values in jpeg.
That doesn't mean there's no concept of colours in raw - but raw viewers need to perform more transformations before renderable RGB values are produced, and those transformations aren't standardised. So the exact default 'look' of raw images (before any editing) is not well defined.
More inscrutable stuff. If "the exact default look ... is not well defined" doesn't that mean that raw files, prior to interpretation, don't look like anything in particular and don't have any particular colors?
Neither raws nor jpegs have any particular 'look' before they're rendered.
With raw formats, we can define the default look as the look of OOC jpegs or jpegs embedded in raw files. That is, with demosaicking, white balance and picture style applied. But the exact processing is usually a trade secret only known to the camera manufacturers. Adobe attempts to simulate the out of camera look with their camera matching profiles in Lightroom.
At the same time the above makes raw files highly 'malleable'.
So much for the light from the real scene! But the malleability if the raw does seem to be its main selling point. Of course, you can edit jpegs too. It's just that "information" is said to be lost, and "artifacts" are said to be introduced.
There's always a big loss of information when a raw file gets converted/exported into JPEG.
Is there a loss of information, as that term is used in plain English, or is there just a loss of data?
It's a loss of visual information.
One can export a raw file into JPEG or other formats, but the reverse operation isn't possible without losses.
What about TIFF or DNG?
DNG is based on TIFF, and so do many raw formats - they're basically modified and extended TIFF.
You can do conversions between raw and DNG and TIFF with minimal losses. However, when you convert to the standard TIFF, you're applying demosaicking and white balance and you're effectively making it "output-referred". You can't convert from TIFF back to a RAW format without some losses.
you can use 'upstream/downstream' terminology. Just a different view.