"im not distilled! im not distilled!!", i continue to insist as i slowly shrink and transform into a prompt
Joined March 2010
- Tweets 5,260
- Following 1,128
- Followers 2,948
- Likes 2,358
438 Photos and videos
I set my Signal history to default to 4 weeks.
Why? So I never know what I lost when I make iOS beta mistakes. π«‘
2
149
17 Oct 2024
N. Korean hackers steal US$3 bln in cryptocurrency since 2017 to fund nuclear program: report en.yna.co.kr/view/AEN2024101β¦
2
20
84
9,464
Sprawl 0x5 is happening tonight. Who's hyped?
2
8
1,122
The slop isn't new, you just have a new scapegoat
1
138
I had claude generate a pptx for me last week. My global claude.md says to follow TDD so it wrote a test that asserted slide 1 said "Ryan Stortz" and then it wrote the slide
2
8
933
I have 18 library cards in my libby
3
263
In the general population, are people more anti-facebook or anti-AI?
306
Building a prompt evaluation harness for RE tasks because I can't figure out the right words that actually result in claude creating types and applying them
2
2
322
I just realized that I'll have to prompt eval my prompt evaluation functions also (in LLM-as-judge mode)
1
2
133
π₯π¬ππ‘ π¦π§π’π₯π§π retweeted
May 12
frog told the LLM "do not hallucinate"
"there," he said, "now the LLM will not make mistakes"
"but the LLM can still hallucinate" said toad
"that is true" said frog
44
616
7,865
211,137
For the curl mythos blog post, my 2c is prompting matters a lot in Mythos.
It seems Daniel and other maintainers were not the ones prompting Mythos...and it didn't find as much and 3 weren't real.
what it would be like if the prompter knew more code details - like attackers
1
7
725
152
CFP is open!! Submit your cool talks about AI hacking/defense/core tech. Gonna be awesome!! βΊοΈ (re-boosting for other time zones)
The Unprompted.au CFP is officially OPEN! If you are doing cool stuff with AI in offense, defense, or working on core AI tech (from frontier models to open source LLMs), we'd love to hear from you! Submit here: unprompted.au/
1
16
53
12,982
/goal seems to be a mistake, it's a great way to burn my full quota.
maybe my tasks don't have high enough signal for the end states
2
2
400
I'm very skeptical of the argument that we'll find new bug classes that LLMs won't
3
15
2,706
When I wrote that code only me and God knew how it worked.
Now only God knows.
to
When Claude wrote that code only the kv cache knew how it worked.
Now 5 minutes later the cache has evicted and no one knows
9
607
π₯π¬ππ‘ π¦π§π’π₯π§π retweeted
11 Dec 2018
COWORKER: we need to find the root cause asap
ME: *takes long drag* the root cause is that our processes are not robust enough to prevent a person from making this mistake
COWORKER: amy please not right now
128
3,384
13,409