Yeah it’s really not limited to prompts talking to eachother, but rather any interactions between an LLM and other things (whether that be another prompt, or a python REPL, or anything)
LLMS
-
Latest LLM Reads Text on Wall Warning of Impact
By
–
The latest LLM can read writing on a wall. It says "You're about to hit me."
-
LangChain 0.0.21: Memory, Conversational Chain, and Improvements
By
–
LangChain version 0.0.21 Added in Memory abstraction (
@sjwhitmore
) Conversational chain (
@sjwhitmore
) Method for adding examples to ExampleSelector class (
@sjwhitmore
) Bug fix for search class Support for all params in the OpenAI wrapper -
Generative AI framing undersells actual capabilities
By
–
Similar issue with "Generative AI" – so hot right now, but this framing is, believe it or not, *underselling* capabilities we are already seeing, and perhaps causing people to work on the wrong things.
-
LangChain Overview by Shubham Saboo Appreciated
By
–
A good overview of LangChain by @Saboo_Shubham_ ! Thanks for the write up 🙂
-
InstructGPT/RLHF tuning makes model assume all questions answerable
By
–
My guess is this is InstructGPT/RLHF rather than anything in the pre-training corpus. Tuning implicitly makes it assume all questions are answerable — it sees all text as “ ” and Q/A is a subset of that.
-
Running Out of Data to Train AI Language Models
By
–
We could run out of data to train #AI language programs https://
technologyreview.com/2022/11/24/106
3684/we-could-run-out-of-data-to-train-ai-language-programs/
… @techreview #MachineLearning #DataScience #IoT #serverless #womenwhocode #Robots #100DaysofCode @JAdP @CatherineAdenle @FmFrancoise @Shi4Tech #BigData #Analytics #DigitalTransformation #DeepLearning -
Dialect-robust Evaluation of Generated Text
By
–
Dialect-robust Evaluation of Generated Text Sun et al.: https://
arxiv.org/abs/2211.00922 #ArtificialIntelligence #DeepLearning #MachineLearning -
Capabilities and use cases for GPT-3
By
–
So what can you use it for? GPT-3 will attempt to answer virtually any text prompt you give it. It can answer questions, write stories, poems, create lists, write research papers, and it can write code. A lot of code. The best way to describe what it can do is to just try it.
-
GPT-3 Training Data Composition Breakdown
By
–
GPT-3 has been trained on 45 TB of text data from different categories: ⬩Common Crawl (8 years of raw web page crawling) ⬩WebText (The text of Reddit posts with 3+ upvotes) ⬩Books (The internet-based books corpora) ⬩Wikipedia Data is then "weighed" as such: