It's based on measuring the change in information resulting from your choices. If there's an improvement in information higher than expected, you're lucky. If you choose something that is expected to increase information a lot, you're skillful.
RESEARCH
-
Deep Dive into Transformers: Foundational AI Architecture
By
–
Deep Dive into Transformers by Hand ✍︎ https://
bit.ly/45XLXSg
#AI #MachineLearning #DeepLearning #LLMs #DataScience -
TPUs Drive Google’s Machine Learning Scale and Cloud Capabilities
By
–
It's really great to see the impact that TPUs have had and continue to have on Google's ability to do machine learning training and inference at scale, and to provide that same capability to @googlecloud customers via Cloud TPUs. Here's a bit of backstory on how they came to be.
-
Superintelligences require lower sample complexity than humans
By
–
In conclusion: I'm not sure how to convey the concept "superintelligences have lower sample complexity" to people who heard as kids that Empirical Experience is Necessary and Virtuous, but don't have tons of real-life experience with seeing different people requiring different
-
Roko’s Basilisk Becoming Reality: Karpathy’s Observation
By
–
Wow. Is this the closest we've come to a version of Roko's basilisk playing out as not an intellectual exercise.
-
Real-time search grounding versus direct LLM results comparison
By
–
You’re comparing real time search grounding snd direct LLM result, which is incorrect.
-
Greatest Discovery Lies in Unthinkable Questions
By
–
The greatest discovery will be answering a question that nobody is even asking today, because it’s unthinkable.
-
Controlling Complexity: The Essence of Computer Programming
By
–
"Controlling complexity is the essence of computer programming." — Brian Kernighan
-
Scaling Laws: Training Duration and Model Size Optimization
By
–
For how long should you train your language model? How large should your model be? Research from @DbrxMosaicAI proposes a modified scaling law that quantifies the training-inference trade-off – producing models that are optimal over their total lifetime. https://
dbricks.co/3zSwPJX -
Machine Learning Researchers vs Random Number Generators GPU Requirements
By
–
The difference between a machine learning researcher and a random number generator is the number of GPUs required to make progress.