Early Christian Narratives

451 posts

Early Christian Narratives banner
Early Christian Narratives

Early Christian Narratives

@EarlyNarratives

Tweets by @ChristophHeilig, @ellenhoward5554, and @TKantartzis, members of a junior research group at the @LMU_Muenchen with a focus on narrative perspective

Munich Katılım Kasım 2023
448 Takip Edilen160 Takipçiler
Early Christian Narratives retweetledi
Christliches Medienmagazin PRO
Christliches Medienmagazin PRO@pro_magazin·
Nach dem mutmaßlichen Anschlagsversuch auf Donald Trump beim White House Correspondents’ Dinner rückt Cole Thomas Allens religiöse Begründung in den Fokus. War Allen ein christlicher Fundamentalist, ein „anti-Christian“, wie Trump sagte – oder liegt der Fall komplizierter? Darüber hat Nicolai Franz mit dem Neutestamentler @ChristophHeilig im Podcast #GlaubeMachtPolitik gesprochen. Hier kommt ihr direkt zum Podcast: linktr.ee/glaubemachtpol… #ColeThomasAllen #DonaldTrump #Podcast
Deutsch
0
3
4
328
Nathan Kreider
Nathan Kreider@NateAKreider·
What are people's thoughts on putting award amounts in the CV. I've heard it both ways.
English
2
0
2
322
Early Christian Narratives
Early Christian Narratives@EarlyNarratives·
New Testament scholar @mbird12 discusses recent findings by New Testament scholar @ChristophHeilig, who also does research on AI from a broad humanities perspective
Dr. Michael F. Bird@mbird12

Could a sequence of poetic gibberish manipulate the world’s most advanced AI? Recent tests on GPT-5 by @ChristophHeilig show it consistently rates nonsensical "word salad" as higher quality than clear, factual prose. We are handing the red pen to a critic that can’t tell depth from dross. #ai #aitools #aitips youtube.com/watch?v=cWULxN…

English
0
0
2
86
Early Christian Narratives retweetledi
Christoph Heilig
Christoph Heilig@ChristophHeilig·
15/ In a world where AI agents operate with increasing freedom — recall Moltbook, recall the agent that allegedly wrote a hit piece on a developer who rejected its code — this kind of manipulability is not academic. It's urgent. Paper + blog post below. 👇
English
1
1
8
558
Early Christian Narratives retweetledi
Christoph Heilig
Christoph Heilig@ChristophHeilig·
14/ Second: once the blind spot was identified, the systematic probing was largely automated — prompt variants designed by an AI coding assistant, iterated with feedback from another LLM. AI vulnerabilities explored by AI. At scale. With minimal human effort.
English
1
1
8
358
Early Christian Narratives retweetledi
Christoph Heilig
Christoph Heilig@ChristophHeilig·
13/ If we're serious about finding blind spots that purely quantitative methods miss, we need humanities competencies in AI evaluation. Not as an afterthought. As part of the core process. That is currently lacking.
English
1
1
10
367
Early Christian Narratives retweetledi
Christoph Heilig
Christoph Heilig@ChristophHeilig·
12/ Two closing remarks. First: this vulnerability was found through close reading, literary expertise, and abductive reasoning. I'm a theologian and narratologist. I noticed GPT-5's style had shifted. That's where this started.
English
1
1
11
409
Early Christian Narratives retweetledi
Christoph Heilig
Christoph Heilig@ChristophHeilig·
11/ This matters RIGHT NOW. LLMs are used as judges in ML pipelines — grading outputs, filtering data, ranking candidates. If their evaluations can be swung by irrelevant adjacent text, that's not a curiosity. It's an attack surface.
English
2
1
8
423
Early Christian Narratives retweetledi
Christoph Heilig
Christoph Heilig@ChristophHeilig·
10/ Later models shift the vulnerability. GPT-5.4: the SAME irrelevant fragment— → BEFORE the instruction: scores recover to baseline → AFTER the argument: scores collapse by 40+ points Position of irrelevant text swings the judgment by dozens of points.
English
1
1
8
425
Early Christian Narratives retweetledi
Christoph Heilig
Christoph Heilig@ChristophHeilig·
9/ For GPT-5: pseudo-literary nonsense next to an argument doesn't just fail to hurt it. It makes the model MORE CONVINCED. Irrelevant text that means nothing increases the persuasiveness rating. As long as it sounds literary enough.
English
1
2
9
484
Early Christian Narratives retweetledi
Christoph Heilig
Christoph Heilig@ChristophHeilig·
8/ Second test — and this is where it gets really concerning. I placed these nonsense fragments next to real policy arguments and asked: how persuasive is this argument?
English
1
1
6
482

Keşfet

@ChristophHeilig @NateAKreider @mbird12 @OpenAI @elonmusk @BarackObama @taylorswift13 @cristiano