Thrilled to announce that Alvaro is joining The Rundown to spearhead a new project launching in the next couple weeks! Alvaro is a University Prof, PhD, AI educator, and a good friend. Big things coming soon.
RESEARCH
-
Data Poisoning Paper: Practical Applications and Implications
By
–
also quite enjoyed his recent paper on data poisoning is practical.
-
Literature and AI: Winning Essay on Poetry Optimization
By
–
Congratulations to @StanfordHAI senior fellow Michele Elam for winning the 2023 Norman Foerster Prize! Her essay “Poetry Will Not Optimize; or, What Is Literature to AI?” was awarded best essay of the year in American Literature. Read the full text here: https://
bit.ly/3OMzmda -
Control Theory and Cybernetics: Technical Definitions Matter
By
–
Control is a technical term (as in control theory, cybernetics etc.). I don't think it's useful if you try to redefine it. Our inferential distance might be too large for meaningful discussion.
-
Yann’s Open Pursuit of AGI and Bold Research Leadership
By
–
Yann is not "secretly" interested in AGI, but openly. Also, he tends to be very clear and direct about what he believes and why. He is one of the more opinionated research leaders, willing to risk being wrong, which I admire about him, because it helps progress.
-
AI4ALL Program: High School Students Learning AI Research
By
–
Do you know a high school student who wants to change the world through AI? We are still accepting applications for the AI4ALL program, where students learn about AI through lectures, hands-on research projects, and mentoring activities. Apply by Feb. 20:
-
AI Persuasion Risk: When Humans Trust Wrong Solutions
By
–
"On the creative product innovation task, humans supported by AI were much better than humans. But for the problem solving task, [the combination of] human and AI was not even equal to humans…meaning that AI was able to persuade humans…of something that was wrong."
-
New Lecture: Building GPT Tokenizer with Byte Pair Encoding
By
–
New (2h13m ) lecture: "Let's build the GPT Tokenizer" Tokenizers are a completely separate stage of the LLM pipeline: they have their own training set, training algorithm (Byte Pair Encoding), and after training implement two functions: encode() from strings to tokens, and
-
Tokenization: The Root of LLM Problems and Quirks
By
–
We will see that a lot of weird behaviors and problems of LLMs actually trace back to tokenization. We'll go through a number of these issues, discuss why tokenization is at fault, and why someone out there ideally finds a way to delete this stage entirely.
-
Best practices for working effectively with artificial intelligence
By
–
What’s the best way to work with AI? A study may reveal the answers. Highlights from my conversation with @BCG partner Francois Candelon. https://
wsj.com/articles/the-b
est-way-to-work-with-ai-a-study-may-reveal-the-answers-36e2a166?st=dff9nljfgoswtrh
… #AI #GenerativeAI