As ChatGPT and systems like it improve, I worry we’ll forget what hallucination looks like. We’ll forget it’s there, lurking in the distributional tails. And we will commit. And upon our servers will be errors. And we will post. And upon our timelines there will be dunks.
ETHICS
-
Hedonic Adaptation: Why Problem-Solving Beats Passive Consumption
By
–
Exactly, just consuming is prone to hedonic adaptation quickly. While solving problems stays challenging –> fun
-
Google’s AI Content Detection and the Importance of Human Writing
By
–
Google is trying to detect AI content versus human-generated content and I expect they'll get good at it in the next year or two. Best to work some human writing into anything you post imo.
-
AI assistance in writing: content creation with human oversight
By
–
An AI helped me write it more quickly and gave me a few extra ideas. But the overall content and prompt examples are very much mine and every word has been touched and edited.
-
Major ML Conferences Need More Transparency in Policy Changes
By
–
I feel like, in the last couple of years, NeurIPS/ICML/ICLR have made a number of changes, often drastic. Change can be good. But I wish these changes were made with a bit more transparency, rationale, and potentially input from the community.
-
AI Tools: Potential for Time Loss and Valuable Lessons Learned
By
–
There is huge potential to lose yourself in this. And I've burned hundreds of hours experimenting with ChatGPT, and hundreds more making images with the various art AI tools. Hopefully my loss can at least be someone else's gain.
-
Testing GraphViz Reformatting Outside AI Training Data
By
–
(*SPOILERS for Glass Onion – I needed a movie that was outside the training window*) @elzr pointed out Riley's GraphViz experiments (
https://
x.com/goodside/statu
s/1561549768987496449?s=20
…) as a reformatting, and I wanted to doublecheck it doesn't only regurgitate from things it knows (for useful generality) -
AI Explainability: From Engineering to Fundamental Scientific Principles
By
–
5/Been Kim discusses AI explainability. AI has taken an engineering-centric approach, where researchers devise techniques via trial and error, and she urges developing fundamental scientific principles that make explanations more trustworthy and accurate.
-
Key Directions for AI: Multimodality, Safety, Data-Centric Approaches, and Evaluation
By
–
4/Douwe Kiela points out key directions: Multimodality, grounding, and interaction so AI understands us better; alignment, attribution, and uncertainty to make models safer; data-centric AI to improve scaling; and better ways to evaluate AI models.
-
AI Personal Timeline: Tracking Goals While Preserving Privacy
By
–
3/Alon Halevy envisions AI capturing massive amounts of data from your daily life — photos, browsing, purchases — and fusing it into a personal timeline that helps you track and achieve your goals, while preserving privacy.