Debugging its own software by running what it creates and doing some QA. Nifty.
SAFETY
-
AI Safety: Regulating Model Naming Practices
By
–
AI safety demands the creation of a law that AI researchers should not be allowed to name their own models anymore. The only thing worse than a hostile AGI would be a hostile AGI called Phat Goose.
-
AI Detection in Writing: CAPTCHA for Authors and Education
By
–
Just checked my new book and the only times the word "delve" is used is when I am quoting GPT-4. Phew. Its like CAPTCHA for authors (but also prone to false positives, which is why accurate AI detection in school settings in nearly impossible)
-
How AI Makes Smart Guesses and Appears More Perceptive
By
–
Plus, it also shows how easy it is for AIs to make smart guesses and seem more perceptive than they are. Paper: https://
open-eqa.github.io/assets/pdfs/pa
per.pdf
… -
AI Models: The Risk of Mimicking Human Flaws Too Perfectly
By
–
i feel like there's potential for an Eponymous Law when we make AI model humans but forget that we don't actually want it to be completely human. an RLHF application of the Sour Lesson solutions via @jason_lopatecki and
-
Experts advocate for legal safe harbor to evaluate AI tools
By
–
Experts call for legal ‘safe harbor’ so researchers, journalists and artists can evaluate #AI tools
by @sharongoldman @VentureBeat Read more: https://
buff.ly/3Vm2iNp #ArtificialIntelligence cc: @pascal_bornet @yvesmulkers @kuriharan -
AI Chatbots Risk Overwhelming Election Workers
By
–
Important piece by @telliotter foreshadowing the potential for AI to overwhelm election workers. There is huge potential for even relatively dumb chatbots to interfere with politics.
-
OpenAI Issues $10M in Superalignment Fast Grants
By
–
Just issued ~$10M in superalignment fast grants:
-
Conscience Canada: AI Ethics and Digital Consciousness Precedent
By
–
Here’s one precedent I know about https://
consciencecanada.ca/?page_id=482 -
AI Data Extraction Accuracy Requires Human Review
By
–
I included the date mistake (assuming 2019 not 2024) in the video to try to help illustrate that – it's generally great at extracting things correctly if everything is present in the text, but you still have to review what it did because there are many ways things could go wrong