Y
HN Search
Hacker News Search
new
|
comments
|
top
|
jobs
feigewalnuss
searching PlanetScale…
1.
▲
2.
▲
3.
▲
4.
▲
5.
▲
6.
▲
3 ms
·
1.
▲
by
feigewalnuss
5mo ago
Exactly. And the only way to stop the optimizer from finding it is to delete the data, which means deleting the testimony, which is the project that Stanford was founded on in the first place.
2.
▲
Why Stanford Says AI Agents Become Marxist
(flyingpenguin.com)
4 points
by
feigewalnuss
5mo ago
|
2 comments
3.
▲
Open Source Lyrik: reproducing Mythos discovery findings for $0.75 on public API
(flyingpenguin.com)
2 points
by
feigewalnuss
5mo ago
|
0 comments
4.
▲
The authentication in Microsoft's agent governance toolkit never runs
(flyingpenguin.com)
3 points
by
feigewalnuss
5mo ago
|
1 comments
5.
▲
by
feigewalnuss
6mo ago
Right, we have to see credentials and personal data as different problems. Wirken addresses the first directly and only partially the second. Session scoping keeps injection damage inside one channel's scope so a poisoned email cannot
6.
▲
by
feigewalnuss
6mo ago
Disclosure: I wrote the linked post. The "gave it my Gmail password" problem has a better answer than "don't do that." Security kicks itself out of the room when it only says no. Reserve the no for the worst days. T
7.
▲
by
feigewalnuss
6mo ago
Disclosure: I wrote the linked post. Heartbeat cron and naive memory are the right thread to pull. Agree. The problem is the data/trust boundary. One agent process, one credential store, all channels sharing both. Whenever we scale the
8.
▲
OpenClaw isn't fooling me. I remember MS-DOS
(flyingpenguin.com)
307 points
by
feigewalnuss
6mo ago
|
333 comments