This seems to suggest that humans were using the browser environment by hand and didn't even consider the source or using some traditional or ai tool to help audit the code. I don't know much about pen testing but I have a hard time believing that is how it's done. Maybe it wasn't how good the AI was but how incompetent the human testers were.

If the word 'literally' is used less than once per sentence it's literally written by AI. Humans literally use it as a word to put emphasis on anything, literally.

I have enjoyed creating a snapshot and allowing claude to --allow-dangerously-skip-permissions You probably shouldn't run something like that on your personal desktop or a production system though.

I changed the personality to "Cynical" and added "Consider possible drawbacks and constraints and advise about them." to the custom instructions under personality in ChatGPT. Helps alot.

I run Jellyfin, but I don't expose it to the public internet. A WireGuard server on the router is perfect for accessing LAN resources, and the app makes split tunneling easy on most devices.

The authorâ(TM)s obsession with the "biological vs. artificial" distinction is the exact same logical fallacy as arguing that airplanes can't actually fly because they don't have feathers or flap their wings like birds. The output (flight) is the same; the mechanism (Bernoulli's principle vs. biomechanics) is irrelevant. This piece tries to argue that because the brain separates language and reasoning , a machine that unifies them can't be intelligent. That is peak absurdity. It is effectively the same thing as arguing that an electric vehicle isn't actually a "car" because it lacks a combustion engine and spark plugs; or insisting that a GPS navigation system doesn't "know" where it is because it triangulates signals rather than "remembering" landmarks like a London cabbie; or, to be even more pedantic, claiming that a speedometer isn't measuring "real" speed because it calculates wheel rotation rather than "feeling the wind" in its face. The method of arrival does not invalidate the destination, and calculating a correct answer via token prediction is functionally indistinguishable from "reasoning" to the observer.

Mazda's implementation of AA starts by making a bluetooth connection to the device and starting music playback. It quickly starts connecting wifi and then back to bluetooth a few times causing several skips. Disabling playback doesn't seem to actually work as the music often start playing anywy. I had to fiddle around and disable bluetooth for the profile and finally it connected smoothly. Has worked mostly fine since then, I've had a few connections failures that were resolved by rebooting both the car and the phone. I suspect most Mazda/AA users never get past the part where it has a problem connecting 100% of the time.

We judge people based on their follower count rather than competency now.

To me the initial reaction in a situation like covid should be an overreaction since we couldn't determine the appropriate reaction at that point. It's really easy to forget just how little information we had. It had overwhelmed medical facilities and nobody had any real information yet about the spread to determine what was safe and what wasn't.

I like an afternoon coffee. In the winter it's always hot and unsweetened. In the summer I like to make heavily sweetened iced coffee creations. Otherwise I mostly drink water and unsweetened seltzer where the sugar is flat/zero year-round.

The article says 100 m of tape = 36 PB.
Earth circumference ~ 40,075,000 m.

Scale it up:
Per meter: 36 PB / 100 m = 0.36 PB/m (= 360 TB/m).

All the way around Earth:
0.36 PB/m x 40,075,000 m = 14,427,000 PB.

That's:
14,427 EB (decimal) = 14.4 ZB (decimal, 1 ZB = 1,000 EB)
13.8 ZB (binary-ish comparison using 1024s)
Songs at 10 MB each (as in the article):

36 PB/100 m -> 3.6 billion songs/100 m -> 36 million songs per meter
40,075,000 m -> 1.44 quadrillion songs.

This can help: https://github.com/AlexxIT/go2rtc The reason I'm aware of it is that it's used by URL:https://frigate.video/. Also worth checking out if you use IP cameras.

How could you miss the fact that Chrome is preinstalled on every Android device? Kind of a major omission when discussing it's rise vs. other browsers.

In future iterations the LLM should be able recognize that it needs help to optimize it's chess play and deploy/use the best chess playing software when facing a master. Assuming it has the goal of defeating the opponent and has "reasoning" and agentic abilities it is not much of a jump for it to pull it off.

I have done a bunch of songwriting with AI and early on I noticed a tendency for it to use the words: neon, echoes, glow, and several others in almost every song. I added instructions to avoid those patterns to my chats and it no longer produced those patterns. In that context it might make sense to assume that some of these papers used AI that was instructed to avoid obvious AI patterns and therefore went undetected.