Hacker News Sleeper AI agents and how Anthropic detects them [video]
https://www.youtube.com/watch?v=Z3WMt_ncgUI
#ycombinator #Anthropic #AI_Safety #Alignment #Sleeper_Agents #AI_alignment
#alignment
3 posts3 participants0 posts today
Apple study shows how aligning Language Models can be improved with checklists (instead of rewards)
9to5Mac · Apple study: LLMs also benefit from an old productivity trick - 9to5MacAn open-source LLM saw big performance improvements after being told by Apple researchers to check its own work by using one simple productivity trick.
Replied in thread
@DM_Zeppelin @Craig @enerhpozyks My introduction to the game was with AD&D (and Basic/Expert).
We eagerly set alignment when creating characters, and never referenced it again. (we were teenagers, so "I'll do what I want!" in Cartman's voice).
Replied in thread
@DM_Zeppelin @enerhpozyks Thinking more about alignment, I see it more as a plot-hook than a rule per se. "It's not a rule, but a tool." It doesn't determine a player's choices, but can determine the results of their choices (a paladin violates their alignment-guided mandates and thus loses their power, etc.)
When I referenced saying to players, "You're acting outside of your alignment" it was more a warning of potential consequences than a mandate. #ttrpg #dnd #alignment
Is there a reason why D&D seems to be the only ttrpg in existence (correct me if I'm wrong) that uses Alignment?
The only time as a DM I've ever told a player, "You're acting outside of your alignment" (or any comment similar to that) is at my teen's D&D group where evil alignments are not allowed.
Should Alignment be one of those big D&D rules that we just ignore altogether?
“World 
+ Brain 
= Consciousness 
?” – Part 1
#Zoomposium with #GeorgNorthoff & #PhilippKlar
How does spontaneous #brainactivity (“#scalefreedynamics” / #fractalpatterns) adapt to #environmentalstimuli?
The Temporo-spatial Theory of Consciousness (#TTC) calls this #alignment – the key to conscious #experience.
Blog: https://philosophies.de/index.php/2025/06/09/welt-gehirn-bewusstsein/
Podcast: https://open.spotify.com/episode/6gQ02XCD77liFbC5Da3nhD?si=uBeWuLyBQcmiqo8j1CX87w
New video! Repairing the Panasonic JU-253-043P floppy disk drive from my Amiga 600.
YouTube: https://youtu.be/jDxcNkC2J8w
PeerTube: https://makertube.net/w/5JSaGyYfmVYdxxEBQFxzPa
How to Contact Heaven—A Guide to Spiritual and Psychological Alignment
https://michaeltheanon.blogspot.com/2025/08/how-to-contact-heavena-guide-to.html
michaeltheanon.blogspot.comHow to Contact Heaven—A Guide to Spiritual and Psychological AlignmentHow to Contact Heaven—A Guide to Spiritual and Psychological Alignment This is a path of radical honesty, psychological alignment, and pure,...
I made this photograph of an #alignment of #Megalithic monuments, a stone circle, a passage tomb, and the grave of Queen #Maeve atop #Knocknarea seen from #Carrowmore, #Sligo, IE, on April 25, 2007 using old technology, a Canon PowerShot A710 IS. Today, I used #gimp 3 to convert it to black and white using different settings than last time I posted it.
A story about Queen Maeve is at: https://celt.ucc.ie/published/T301012/index.html
Continued thread
#USpol #Alignment of the press in the #USA is progressing:
*#CBS bribes #Trump with 16 million #USD and cancelation of Trump opponent's show*
"The Late Show with #StephenColbert" was the show with the highest ratings
1)
Der Head of the #FCC, a Trump acolyte, approves the sale of the parent company of #CBS,
#Nickelodeon & the movie studio #ParamountPictures to #Skydance.2)
The Independent · The Late Show with Stephen Colbert will end next year
You've heard of the alignment problem in AI? This is where it will really start to be an issue:
LLMs Can Now Self-Evolve At Test Time Using Reinforcement Learning | by Dr. Ashish Bamania | Jul, 2025 | AI Advances https://ai.gopubby.com/llms-can-now-self-evolve-at-test-time-using-reinforcement-learning-e769ee6d3f86
AI Advances · LLMs Can Now Self-Evolve At Test Time Using Reinforcement Learning
#DH2025: Another #alignment project in #CLS. Anastasia Glawion and Dhara Lechner of DHSS FAU Erlangen-Nürnberg present their work on "Tracing #Transformation: editorial shifts in the Grimm brothers’ tales".
#FairyTales #German-language #fiction
Replied in thread
@GottaLaff
#USpol #History #Totalitarianism
"Trump...is going on offense against 3 members of the Corporation 4 Public Broadcasting who have not left their posts"
"the "Gleichschaltung der Presse" (#Alignment of the media in the #ThirdReich) exercised by the #Gestapo + lifted to new levels by Xi through the Great Firewall of #China and the censorship department, has already begun happening in #Russia, #Hungary, and in the #US (#BillionaireMedia,) as well as elsewhere"
MastodonHistoPol (#HP) 🏴 🇺🇸 🏴 (@HistoPol@mastodon.social)@VeroniqueB99 @ar1
I only found this:
https://www.nytimes.com/2024/11/06/technology/trump-election-denial.html
Not the thing.
However, the "Gleichschaltung der Presse" (#Alignment of the media in the #ThirdReich) exercised by the #Gestapo and lifted to new levels by Xi through the Great Firewall of #China and the censorship department, has already begun happening in #Russia, #Hungary, and in the #US (#BillionaireMedia,) as well as elsewhere.
@Beachbum@mastodon.sdf.org
Gerade ein Kundengespräch gehabt. Der Kunde war "not amused". Idee, für den nächsten OKR Alignment Workshop bauen wir ein Siedler von Catan Spiel. Vielleicht ist das eine gute Idee. Jetzt baue ich mal so ein Kärtchen.
The #AI #alignment of GIGO with FAFO was a huge success
A wild journey into tech delusion where aligning an #LLM somehow requires running #JavaScript. 
Who would’ve thought enabling a web script could solve "systemic misalignment"? 
Please hold while we recalibrate reality. 
https://www.systemicmisalignment.com/ #techdelusion #alignment #systemicmisalignment #recalibratereality #HackerNews #ngated
www.systemicmisalignment.comSystemic MisalignmentResearch platform for analyzing systemic misalignment in AI alignment methods
#Development #Techniques
Safe align your content · The CSS ‘safe’ keyword is worth knowing https://ilo.im/164jgu
_____
#CSS #Layout #Container #Alignment #WebDev #Frontend #CssGrid #CssFlexbox
ilo.imSafe align your content
More from 
CSS by T. Afif 
Continued thread
One of the cogent warnings Daniel raised is, that #AI already deceive the users.
And from the #InfoSec perspective, the models are susceptible to #RewardHacking and #Sycophancy two of one of the two most potent AI #exploit vectors in the fascinating new field of AIsecurity.
#AIalignment #AIsecurity #alignment