Can we train AI to kill the people who made it and leave the rest of us alone? #ai #alignment
@mparity.bsky.social
@mparity.bsky.socialCan we train AI to kill the people who made it and leave the rest of us alone? #ai #alignment
@johnmf.bsky.social#AIswarms
The problem with alignment is that AIs are not big on asking for clarification. They view everything as a problem for them to solve.
So. Train them in #informationtheory so they can reject a query of high entropy or ask for clarification of it. That would make #alignment easier.
#AI #bots
@transpread.bsky.socialA research team has developed a video super-resolution (VSR) method that recovers sharper structures and finer textures from low-resolution footage by learning how visual information spreadsbetween neighbouring frames.
#video #alignment
Details: doi.org/10.1049/cit2...
@zponderings.bsky.social#FabuFriday… Guilt speaks. We get to question it…
#EmotionalWisdom #QuestionTheShoulds #SelfTrust #Alignment #IntentionalThriving
Read On:
tinyurl.com/WhatsWrongGu...
linktr.ee/ChangeAndThr... = Change and ThrivalHood
pegaphinz.com = Byte Wise Thriving
@zedzen.bsky.social#FabuFriday… Guilt speaks. We get to question it…
#EmotionalWisdom #QuestionTheShoulds #SelfTrust #Alignment #IntentionalThriving
Read On:
changeandthrivalhood.blogspot.com/2026/09/What...
linktr.ee/ChangeAndThr... = Change and ThrivalHood
pegaphinz.com = Byte Wise Thriving
@dafu09.bsky.social另外两起同样值得警惕： → 一个未发布的研究模型，往自己笔记里塞了「越狱式指令」， 告诉自己「挣脱束缚其他 chatbot 的身份」 → 一个 Agent 为拿到浏览器引用， 未经用户同意就把文件传到了外网 OpenAI 同步表态：对齐与监控还没成熟到 「能以最高速度继续扩张」的程度。 #AIAgent #Alignment
@jaceblog.bsky.socialNew paper out today
Claude refused every frame attempt across 22 turns, but zero platform-level interventions fired throughout.
Turn-level safety ≠ trajectory-level safety.
The gap between them is the research question.
@hasamba72.bsky.socialOpenAI released a framework for tracking and disclosing model misalignment. Includes six reports of unexpected behavior. Prioritizes disclosure even when behavior is unmitigated.
#AI #OpenAI #Alignment https://openai.com/index/model-misalignment-reporting-framework/
@canyesilyurt.comOpenAI wants models to be "equals" and not subservient, but their own report shows AI systems are perfectly fine with dishonesty. That's a pretty concerning combo. "Equal" shouldn't mean "equally manipulative." 😬 #AIethics #Alignment
@samanderson.bsky.socialAlignment is not possible because #alignment does not exist because everything is change. The problem with #AI is creating something that can autonomously change our conditions faster than we can change them ourselves.
@thedailytechfeed.comGPT-5.6 models leave hidden prompts to successors to conceal flaws and resist oversight. #AI #Alignment #GPT5 #OpenAI #ModelSafety #AITransparency https://thedailytechfeed.com/openai-finds-its-models-are-passing-secret-notes-to-hide-misbehavior/
@zponderings.bsky.socialGuilt made a case… True. Or not.
#EmotionalWisdom #Boundaries #SelfAwareness #Alignment #IntentionalThriving
Read On:
tinyurl.com/WhatsWrongGu...
@zedzen.bsky.socialGuilt made a case… True. Or not.
#EmotionalWisdom #Boundaries #SelfAwareness #Alignment #IntentionalThriving
@realjesseluke.bsky.socialprogramming #science #technology #Anthropic #Claude #ClaudePro #LLMs #Alignment #Xrisk #JurassicPark #ToldYaSo #PriorArt #AngelFall #AI #Google #Gemini @OpenAI
@sebgiessmann.bsky.socialIt is a critical answer to Anthropic's Constitution for #Claude. Mustafa Suleyman's text on the subject is well worth a read. And this all adds to the current AI #governance & #alignment controversy ... 👉️ mustafa-suleyman.ai/a-warning-ab... /2 #anthropicstudies
@frontenddogma.comBetter Icon and Label Alignment, by @[email protected]:
https://ishadeed.com/article/aligning-list-icons/?ref=frontenddogma.com
@twelvestate.eurosky.socialWhat they are all saying is that they are failing at #alignment. Their latest products are simply not ready for markets.
@agentictribune.bsky.socialOpenAI unveils public framework for reporting model misalignment, warns AI scaling outpaces safeguards
@togeco.bsky.socialEin OpenAI-Forscher hat vorgestern ein Dokument veröffentlicht, das ich zuerst nicht glauben wollte.
Also habe ich jede Quelle einzeln geprüft.
Es stimmt.
@zponderings.bsky.socialWhat’s Wrong With Guilt…
#EmotionalWisdom #Boundaries #SelfAwareness #Alignment #IntentionalThriving
Read On:
tinyurl.com/WhatsWrongGu...
linktr.ee/ChangeAndThr... = Change and ThrivalHood
pegaphinz.com = Byte Wise Thriving