We’ve published a quantitative case study on prompt engineering for one of our most popular features, Claude’s industry-leading 100K token context window.
LLMS
-
Prompt Optimization Techniques for Improved Claude Recall
By
–
The analysis highlights two prompt optimization techniques you can use to improve Claude's recall over long context documents: reference quotes, and examples of correctly answered questions.
-
Claude 2 vs Perplexity: Which AI Model Do You Use?
By
–
Do you use Claude 2 or the default perplexity model?
-
LLM Reversal Curse: Asymmetric Relationship Learning Failures
By
–
The Reversal Curse: LLMs trained on "A is B" fail to learn "B is A" Berglund et al.: https://
arxiv.org/abs/2309.12288
v1
… #ArtificialIntelligence #LLM #MachineLearning -
LLMs Dramatically Improve Coding Skills and Productivity
By
–
LLMs can dramatically improve your coding skills This diagram illustrates how LLMs can help update your mental models of the data, revise your intent, and check results. In the next 3-5 years, LLMs are going 2-10x our understanding of the world and productivity.
-
Stanford CS224N: New 2023 Deep Learning NLP Lectures Available
By
–
[New Lectures] Stanford CS224N: Natural Language Processing with Deep Learning Stanford NLP course is arguably one of the best Deep NLP courses on the web. The new lectures from 2023 iteration are public now. The course covers fundamental techniques and topics related to deep
-
Detecting Additional Forces Beyond Token Prediction in Fine-tuned LLMs
By
–
Many LLMs have already been RLHFed and finetuned into activities other than "predict the next token a human would write". This being the case, how would you tell if the output was being driven by some extra force above that and all the finetuning?
-
The Inscrutability of LLM Low-Level Floats and Knowledge Gaps
By
–
You can! But in the case of LLMs, we don't. We speak of the inscrutability of the low-level floats because we have no higher-level knowledge to speak of instead.
-
Pre-training evidence debated; reversal learning skepticism noted
By
–
evidence for it in pre-training is just experiment 2 on celebrity parents — seems plausible but less clear than the SFT case imho. idk; LLMs generalize in pre-training in ways that become more impressive if reversal learning is known impossible so my prior is low