Unravel the steps of the Data Science Lifecycle! Understanding the problem to deploying solutions, each phase is pivotal. Enhance your data journey with insights from 'The Digital Edge' http://
bit.ly/3u4pILl #DataScience #AI Follow @ingliguori for groundbreaking insights!
AI Dynamics
-
Data Science Lifecycle: From Problem Understanding to Solution Deployment
By
–
-
Scale launches SEAL Leaderboards for frontier LLM evaluation
By
–
New from Scale: SEAL Leaderboards — a new benchmark arena for frontier LLMs – Private, novel assessments that models can’t train on
– ELO-scale rankings (via Bradley-Terry)
– Domain leaderboards (today: coding, math, instruct, Spanish — more soon!) (Links in reply) -
New AI Safety Paper Published in Science
By
–
New AI safety paper in Science with a lot of authors: science.org/doi/10.1126/scie… [Translated from EN to English]
→ View original post on X — @geoffreyhinton, 2024-05-23 21:48 UTC
-
Using ChatGPT Code Interpreter to Implement Hidden-Information Games
By
–
Using ChatGPT's code interpreter to hide choices also lets you play games that need hidden information or simultaneous turns, like hangman, 20 questions, or rock paper scissors, as shown here:
-
Prompting trick: split example responses into half-messages
By
–
This won’t help in zero shot but if you have in-context examples try splitting each example response into two half-messages, so it appears assistant responses are expected come in pairs with a semantically meaningless split.
-
Advantages of letting AI produce longer, self-evaluated responses
By
–
Yes because you can let it talk longer, consider multiple options, self-evaluate, etc. without a wall of text flooding your screen every time. There are more opportunities for it to usefully babble on when you don’t have to read any of it, even to fish out the answer.
-
Prompt technique: hiding chain-of-thought in code comments
By
–
Teaching ChatGPT to “think silently” by hiding planning, chain-of-thought, or other intermediate writing in interpreted Python code blocks as comments or strings. Text remains visible to the model (and the user, by clicking to view code) without cluttering the chat dialog.
-
Using ‘You are GPT-4o’ as a ChatGPT memory prompt
By
–
Adding “You are GPT-4o” to my ChatGPT memory like it’s a Memento chest tattoo
-
ChatGPT 4o can describe images of real people
By
–
New ChatGPT 4o can now describe images of real people, like this one of Barack Obama racing Lightning McQueen in his Mercedes Benz while smoking a cigar, listening to Playboi Carti, getting blocked on X by Paul Skallas, and running over Charli XCX:
-
Karpathy Recommends the Best AI Newsletter Right Now
By
–
+1 to the best AI newsletter atm that I enjoy skimming, great/ambitious work by @swyx & friends: https://buttondown.email/ainews/archive/ "Skimming" because they are very long. Not sure how it is built, sounds like there is a lot of LLM aid going on indexing ~356 Twitters, ~21 Discords, etc.