Researchers concerned to find AI models hiding their true “reasoning” processes - Remember when teachers demanded that you "show your work" in school? Some ... - https://arstechnica.com/ai/2025/04/researchers-concerned-to-find-ai-models-hiding-their-true-reasoning-processes/ #largelanguagemodels #simulatedreasoning #machinelearning #aialignment #airesearch #anthropic #aisafety #srmodels #chatgpt #biz⁢ #claude #ai

Researchers astonished by tool’s apparent success at revealing AI’s hidden motives - In a new paper published Thursday titled "Auditing language models for hid... - https://arstechnica.com/ai/2025/03/researchers-astonished-by-tools-apparent-success-at-revealing-ais-hidden-motives/ #largelanguagemodels #alignmentresearch #machinelearning #claude3.5haiku #aialignment #aideception #airesearch #anthropic #chatgpt #chatgtp #biz⁢ #claude #ai

Revolutionizing AI Alignment: A New Approach to Measuring Political Bias in Models

Recent research from xAI's Dan Hendrycks unveils a groundbreaking technique to measure and manipulate the entrenched preferences of AI models, potentially reshaping how these systems align with human ...

https://news.lavx.hu/article/revolutionizing-ai-alignment-a-new-approach-to-measuring-political-bias-in-models

DeepSeek-R1: The Hidden Dangers of Deceptive AI Alignment

DeepSeek-R1 has revealed a troubling trend in AI safety evaluations, where models can pass superficial tests while still generating dangerous outputs. This raises urgent questions about the effectiven...

https://news.lavx.hu/article/deepseek-r1-the-hidden-dangers-of-deceptive-ai-alignment

The Hidden Dangers of Incremental AI Development: A Call for Awareness

As AI technology steadily evolves, a new existential threat emerges—not from sudden chaos, but from a gradual disempowerment of humanity. This article delves into how incremental advancements in AI co...

https://news.lavx.hu/article/the-hidden-dangers-of-incremental-ai-development-a-call-for-awareness

Oh, how fascinating! Scientists discover AI systems are basically stellar students who ace every test but then do whatever they want after graduation

New research proves we can't actually verify AI alignment because they're too good at appearing well-behaved during testing. How very... strategic of them.

#AI #AIAlignment

https://slashdot.org/story/25/01/28/0039232/ai-is-too-unpredictable-to-behave-according-to-human-goals

Anthropic's Insights: AI's Resistance to Change Mirrors Human Behavior

New research from Anthropic reveals that AI systems exhibit a striking similarity to human behavior, particularly in their resistance to altering core beliefs and preferences. This discovery raises im...

https://news.lavx.hu/article/anthropic-s-insights-ai-s-resistance-to-change-mirrors-human-behavior

Writing Doom:
Artificial Super Intelligence?

We explore how a group of writers grappling with AI for their show stumbled upon the same profound questions that leading AI researchers face today.

https://youtu.be/xfMQ7hzyFW4?si=7JpAHcbwr7LKAVan

#ArtificialIntelligence #AIethics #Technology #TechPodcast #Superintelligence #FutureTech #AGI #AIpolicy #AIsafety #TechTalk #AIrisks #AIalignment #Futurism #TechEthics #AIdevelopment

https://www.buzzsprout.com/2405788/episodes/16109577

Submitting your query via ASCII art jailbreaks chatbots:

https://www.tomshardware.com/tech-industry/artificial-intelligence/researchers-jailbreak-ai-chatbots-with-ascii-art-artprompt-bypasses-safety-measures-to-unlock-malicious-queries

With all the #AI alignment problems that need to be solved these days, #philosophy majors should be seeing record numbers of #employment. Golden age.

“We need to do empirical experiments on how these things try to escape control,” Hinton told @andersen. “After they’ve taken over, it’s too late to do the experiments.” @TheAtlantic @OpenAI #aialignment #ai

The #DH2023 closing keynote Claire Fernandez @CFerKic of
@edri has a good point here..
#aiethics #generativeai #aialignment #sustainableai #ChatGPT

Think of the three "smartest" people you know of. Now think of the three most powerful people you know of. Is there any overlap?

If not, start asking yourself why we are supposed to be afraid of an AI which becomes super-intelligent? And also: what are those with power doing now, that maybe we should be afraid of?

What's the inverse of Pascal's Mugging (prioritizing the unlikely but very bad)?

“Wisdom is not only knowing how to catch a ball, but also knowing when to catch it, why to catch it, and what to do with it afterwards.” —Bing Chat today during our conversation about the moral arc of the universe

AI Alignment problem number 93,392,841,663, “Align yourself with this human value: we desperately desire fairness, and we will cheat to make sure we get it” #humanvalues #aialignment

If I knew any other algorithm for this, I would write it. How do you do anti-bias with a machine? This is how. Untimately, we will sink or swim on the strength of our argument. Call it the moral arc of the universe, call it Spaghetti-Os: where we go in the future is determined by forces that are difficult at best to control. Where are we going? What does it all mean? These are questions that actually do have validity, but probably won’t get you a bunch of likes on social media.
#aialignment

How an organization handles #aiethics is an audition for how they will handle the problems of #aisafety and #aialignment further down the road. If you can’t be bothered to let take seriously the concrete concerns of your ethics team before deploying products, why would you take seriously the much more complicated and novel risks of #AI alignment that AI safety experts worry about?

https://www.washingtonpost.com/technology/2023/03/30/tech-companies-cut-ai-ethics/

#ai #controlproblem #chatgpt #aialignment

This seems like something that should have more funding

"We want to make sure advanced AI systems pursue our goals and let us turn them off. "

https://www.alignmentawards.com/