As I understand it from the interface, we can delete memories, to eliminate risk it retains an unproductive memory.

That said, I am not a very advanced or very sophisticated user of these tools, but I don't really see what memory offers, other than the risk of being fooled more effectively.

What am I even doing here?

In fact, I asked it to explain the output, and what I got back was almost literally a horoscope.

It appears that the corroboration mostly references the conversation itself. This is probably because I have memory disabled. That's a horoscope if I ever saw one.…

I tried this prompt and strongly suspect the output (like much of this kind of unrestrained prompting?) is a horoscope.

Should I be embarrassed by how many times `pacman -Qqn | pacman -Syyu -` has fixed things after I have made a dumb, avoidable mistake (e.g., like rebooting in the middle of a `pacman -Syyu` whole-system-upgrade…)

In single-user mode, the fix is easy, obvious, but a bit irritating.

# using Arch install media
$ cryptsetup luksOpen /dev/nvme1n1p2 root
$ mount /dev/mapper/root /mnt
$ mount /dev/nvme1n1p1 /mnt/boot
$ arch-chroot /mnt /bin/zsh

# inside chroot
$ pacman -S pam{,base,openssh,sudo,util-linux}

if __name__ == '__main__':
with open(__file__) as f:
code = f.read().replace(' ' * 4, '\t')
run(main(handle=environ['HANDLE'], password=environ['PASSWORD'], code=code))
### 04/04

for msg in messages:
post = await client.send_post(
client_utils.TextBuilder().text(msg),
reply_to=models.AppBskyFeedPost.ReplyRef(parent=last_ref, root=root_ref) if root_ref else None,
)
last_ref = models.create_strong_ref(post)
root_ref = root_ref or last_ref

### 03/04

async def main(handle, password, code):
parts = [*chunked(code, size=290)]
messages = [
f'{txt}\n### {idx:02d}/{len(parts):02d}'
for idx, txt in enumerate(parts, start=1)
]

client = AsyncClient()
profile = await client.login(handle, password)

root_ref = last_ref = None
### 02/04