RESEARCH

Creative Commons AI Prompts for Educators and Students

By

–

05 March 2024 1h15

Here are our prompts for educators and students, released under Creative Commons license. They are built around the pedagogical approaches we discuss in our papers, and have worked well in class. I hope others experiment with them, improve them, & share. https://
moreusefulthings.com

Why GPT-4 Remains Competitive Against Claude 3 Gemini

By

–

05 March 2024 0h26

How if GPT-4 still a top model? Claude 3 and Gemini Advanced do exceed it in some areas, but it is still pretty close Does it mean:
1) GPT-4 is as good as it gets (I suspect not)
2) OpenAI has The Secret
3) Everyone else stopped training when they beat GPT-4, better models soon

Claude 3 vs GPT-4: Benchmark Performance Comparison Analysis

By

–

05 March 2024 0h18

Useful to note that Claude 3 does not definitely surpass GPT-4 on benchmarks; which have improved over time since GPT-4’s release (Claude 3 compared itself only to the initial benchmarks). This matches my personal experience, its a GPT-4 class model, with some special strengths.

→ View original post on X — @plinz

TF/IDF as Relevance Measure in Machine Learning

By

@plinz

–

04 March 2024 23h58

tf/idf is the most straightforward measure of relevance, it's not surprising that this kind of needle is sticking out like a sore thumb

→ View original post on X — @rodneyabrooks

Rodney Brooks Reflects on 2000 Humanoid Robot Research Predictions

By

@rodneyabrooks

–

04 March 2024 22h53

Had forgotten that I had written this back in 2000, soon after my research group had started building humanoid robots (a later company I founded put 1000s of humanoid robots in factories). Trying to decide how well it has stood up over 24 years. https://
ti.me/Wa5ftZ via @TIME

→ View original post on X — @karpathy

Request for Updated Comparison of Latest AI Models Across Organizations

By

@karpathy

–

04 March 2024 21h05

Would be interesting to see this updated with the latest models from all orgs

Claude 3 Achieves 60% on Difficult AI Test Benchmark

By

–

04 March 2024 20h55

This was designed to be a very hard test for AIs, and the questions were kept private, lowering the chance they were in the training data. PhDs with access to the internet got 34% of the questions right outside their specialty, 65%-75% inside. The new Claude 3 gets 60% overall.

→ View original post on X — @karpathy

Claude 3 Tokenization Challenge: Stylistic but Contains Subtle Hallucinations

By

@karpathy

–

04 March 2024 20h14

Claude 3 takes on the Tokenization book chapter challenge 🙂 context: https://t.co/yRaeTbkblY

Definitely looks quite nice, stylistically!

If you look closer there are a number of subtle issues / hallucinations. One example there is a claim that "hello world" tokenizes into 3… https://t.co/mMjK27Uk2y
— Andrej Karpathy (@karpathy) 4 mars 2024

Claude 3 takes on the Tokenization book chapter challenge 🙂 context: https://
x.com/karpathy/statu
s/1760740503614836917
… Definitely looks quite nice, stylistically! If you look closer there are a number of subtle issues / hallucinations. One example there is a claim that "hello world" tokenizes into 3