Fooling a superintelligence is a more adversarial problem than training a robot. In the latter case intelligence works with you; in the former, against you. "Can learn from" and "cannot distinguish" are different orders of requirement.
SAFETY
-
Reverting GPT-4 increases x-risk, not safety measure
By
–
Reverting to a pre-GPT-4 state heightens x-risk, as we'd lose our only near-AGI system for testing, learning, and potential incremental progress towards AGI. This isn’t a good advice.
-
Emergent Abilities in LLMs: Evidence of In-Context Learning
By
–
Are Emergent Abilities in Large Language Models just In-Context Learning? ***********
Spoiler: YES Through a series of over 1,000 experiments, we provide compelling evidence: http://
arxiv.org/abs/2309.01809 Our results allay safety concerns regarding latent hazardous -
ACCEPT-AI Framework Advances Ethical Pediatric Medical Research
By
–
Children present uniquely complex data quandaries for AI, especially regarding consent and equity. Researchers proposed a new framework called ACCEPT-AI and discussed how it can help ethically advance AI medical research involving pediatric patients.
-
Reducing Claude Hallucinations in Blog and PDF Summarization
By
–
When prompting @AnthropicAI
’s Claude, what do you do to reduce hallucinations of made up stats and fact for summarising blog posts and longer pdfs? -
Autonomous vehicles struggle with coordination in heavy traffic scenarios
By
–
Parece que el futuro en el que todos los coches autónomos funcionen en armonía todavía está lejos de hacerse realidad.
— Juan Merodio (@juanmerodio) 22 septembre 2023
Cuando se juntan demasiados y algo se sale de la ruta establecida llegan los problemas y se monta un atasco monumental. Nadie dijo que fuera fácil, parece que… pic.twitter.com/tvD882rZmhParece que el futuro en el que todos los coches autónomos funcionen en armonía todavía está lejos de hacerse realidad. Cuando se juntan demasiados y algo se sale de la ruta establecida llegan los problemas y se monta un atasco monumental. Nadie dijo que fuera fácil, parece que
-
Tech Billionaires Funding Doomsday Cults AI Apocalypse Narrative
By
–
Definitely should ally with doomsday cults funded by Big Tech billionaires funding the "AI" megacorporations themselves. Y'all are so unserious. https://
washingtonpost.com/technology/202
3/07/05/ai-apocalypse-college-students/
… -
TESCREALism and AI: Utopia, Apocalypse, and Eugenics Concerns
By
–
Yes. I should ally with the Elmos and Altmans and Anthropics and other TESCREALists like Tegmark who cycle between selling utopia and apocalypse. Not to mention the eugenics roots of all this. https://
youtube.com/watch?v=P7XT4T
WLzJw&t=6s
… -
AI Testing Bias: Race, Gender, Culture and Creator Influence
By
–
Continued
"determining whether a test, perhaps quite unintentionally, is partly a reflection of the race, gender,
culture or social class of its creator is a subtle, complex & often politically charged issue…many have concluded that it is wise to stay well clear of this topic" -
Ethical AI: Expert Discussion on Responsible LLM Applications
By
–
Join #AI industry experts @Aishwarya_Sri0 from @Google and @DalianaLiu@predibase for an interactive discussion on ethical concerns when building with #LLMs incl. real world use cases and best practices for building responsible #GenAI applications. https://
pbase.ai/3RweCZo