I'm fine with that if you label those as assumptions. There has been no comparative test of the abilities that are unique to ChatGPT because, by definition, they are only in ChatGPT and no good benchmarks exist. Dismissing a branch of research via assumptions is not scientific.
LLMS
-
Transformers Learn In-Context Through Gradient Descent
By
–
Transformers learn in-context by gradient descent Oswald et al.: https://
arxiv.org/abs/2212.07677 #MachineLearning #DeepLearning #ArtificialIntelligence -
CALM: Dynamic Computational Effort for Language Models
By
–
Presenting Confident Adaptive Language Modeling (CALM), a novel method that allows language models to dynamically modify computational effort when generating text. Learn how CALM can accelerate text generation while preserving output quality → https://t.co/Nm7yyT8sMA pic.twitter.com/MpuaPBzDMU
— Google AI (@GoogleAI) 16 décembre 2022Presenting Confident Adaptive Language Modeling (CALM), a novel method that allows language models to dynamically modify computational effort when generating text. Learn how CALM can accelerate text generation while preserving output quality → https://
goo.gle/3HJKzbM -
GPT SQL Writing Capabilities and UI Transparency for Applications
By
–
Nice work, app shows application to twitter search but the deeper demo is how good GPTs are in writing SQL. Very broadly applicable. wrt UIUX I like that the decoded SQL is available for verification, imo necessary for higher stake applications. https://t.co/70oLMjZj64
— Andrej Karpathy (@karpathy) 16 décembre 2022Nice work, app shows application to twitter search but the deeper demo is how good GPTs are in writing SQL. Very broadly applicable. wrt UIUX I like that the decoded SQL is available for verification, imo necessary for higher stake applications.
-
ROSCOE: New Metrics Suite for Evaluating Step-by-Step Reasoning
By
–
ROSCOE is a first-of-its-kind suite of metrics for scoring step-by-step reasoning. By publishing this study we hope to provide a foundation that enables scalable systematic evaluation and benchmarking of new language models. See the paper on arXiv https://
arxiv.org/abs/2212.07919 -
ChatGPT’s Persuasive Power in Sales and Negotiations
By
–
The potential for ChatGPT and other powerful language tools to persuade people is fascinating and will only improve. And it turns out that some companies are already harnessing the tech for sales and negotiations. My latest story for WIRED:
-
OpenAI Releases New Tokenizer for Accurate Token Usage Estimation
By
–
@OpenAI also released a new tokenizer This makes it possible to more accurately (and quickly) estimate your token usage This is actually extremely helpful for the `collapse` step mentioned above!
-
LangChain Updates: Tokenizer, Document Reduction, Better Embeddings
By
–
TL;DR: The new tokenizer allows easy estimation of token usage The `collapse` step allows for map-reducing over large number of documents Better embeddings allows for better searching over those documents https://
github.com/hwchase17/lang
chain
… -
Collapse Step for Map-Reduce Chain Scaling
By
–
Adding a `collapse` step into the map-reduce chain This recursively collapses documents into smaller and smaller chunks, allowing you to scale to arbitrarily many documents This was a @johnjnay special – see his thread for more
-
OpenAI Releases New Embedding Model for Semantic Search
By
–
@OpenAI released a new embedding model (`text-embedding-ada-002`) This makes it possible to more easily do semantic search over collections of documents This is now the default OpenAI embedding model in LangChain