Based on your analogue they drive the car for 7.5 inches (614.4 Kb by 63360 inches by 20 divided by 103179878.4 Kb) and promise based on that that car travels 20mph which might be true, yes, but the scale disproportion is too considerable to not require tests. This is not maths, this is a real physical device - how would it would behave on larger real data remains to be seen.

106 Gbps

They get to this result on 0.6 MB of data (paper, page 5)

They even say:

Moreover, there is no need to evaluate our design with datasets larger than the ones we have used; we achieve steady state performance with our datasets

This requires an explanation. I do see the need - if you promise 100Gbps you need to process at least a few Tbs.

print("x") is you want to screw your students.

Excel enabled non-programmers to create basically any app as long as they are fine with a cell-based UI. Same with Access and CRUD apps. I know people love to dunk on M$ here, and for good reasons too, but these two programs are probably responsible for a decent chunk or PoC/v1 projects worldwide.

deleted by creator

We actually have a live experience of how that could go down

Another example: latest iteration of Google Captcha. Released with promises to end manually inputting text captchas, the main thing it turned out to check for is whatever you are logged into your google account. If so, you get through automatically, or, at worst have to press a checkbox. If you are not logged in, enjoy selecting fire hydrants and crosswalks.

Yea, it’s strange - top to bottom rows sum up to 94, 100, 92, 94, 100, 100

That is conveniently left out of the speck. Attestation server may require signed binary on a client system, it may require whatever it wants really, because why not? It’s a website who decides to trust attestation server or not.

They aren’t proposing a way for browsers to DRM page contents and prevent modifications from extensions.

And yet, this proposal would make it easier to do so.

Basically, it would allow websites to only serve users who comply with website requirements (i.e., no extensions, no ad blockers, only Chrome-based, whatever) whatever these requirements are.

You (your browser) go to a website, example.com, which requires attestation. So you must go to an attestation server and attest your device/browser combo (by telling the attestation server whatever information it requires). If the attestation server thinks you are trustworthy, it gives you an integrity token that you pass to example.com, and then you can see example.com. The website knows which attestation server issued your integrity token, so you can’t create your own.

So no extra software means no attestation server would attest you; means you can’t see example.com. End of story. It’s the same as the current “your browser is not supported” window, only you can’t get around it by changing the user agent.

As usual with these initiatives, bullshit is spread across different specs - this spec by itself implies that any number of attestation servers can exist, and they can check whatever they want, and no browser should be excluded, etc., etc., but practical implementation would probably check installed extensions, etc.

Research linked in the tweet (direct quotes, page 6) claims that for "GPT-4, the percentage of generations that are directly executable dropped from 52.0% in March to 10.0% in June. " because “they added extra triple quotes before and after the code snippet, rendering the code not executable.” so I wouldn’t listen to this particular paper too much. But yeah OpenAI tinkers with their models, probably trying to run it for cheaper and that results in these changes. They do have versioning but old versions are deprecated and removed often so what could you do?