A quiet AI release that I think a lot of people would actually use is the addition of image editing right inside of ChatGPT. If you click on a DALL-E image, you can highlight the areas to edit & use words to describe the changes. It works well but has the usual inconsistencies.
GENERATIVE AI
-
Dana-Farber Tests GPT-4 in Medicine, Finds Falsehoods and Costs
By
–
Dana-Farber Cancer Institute finds main GPT-4 concerns include falsehoods, high costs The renowned medical institute trialed letting 12,500 staff members use a private instance of GPT-4, a first-of-its-kind implementation in medicine. https://
zdnet.com/article/dana-f
arber-cancer-institute-finds-main-gpt-4-concerns-include-falsehoods-high-costs/
… @OpenAI -
Activation Trade-offs: Current vs Future Token Optimization
By
–
I guess there is a tradeoff between how each activation is used, whether it is doing something that's optimal for a current or future token. i'm using information and store both pretty loosely. 🙂
-
HyperWrite Model Training Progress Update
By
–
I'll definitely try some things over the next few days. Been heads down training models internally for HyperWrite, so it might be a few days.
-
Adaptive RAG Implementation with Cohere Command-R+
By
–
Adaptive RAG w/ Cohere's new Command-R+ Adaptive-RAG (
@SoyeongJeong97 et al) is a recent paper that combines (1) query analysis and (2) iterative answer construction to seamlessly handle queries of differing complexity. We took at stab at implementing these ideas from scratch -
Transformers Token Processing: Information Storage Nuances
By
–
yes! there is a lot of nuance. My preferred way to put it is "transformers don't pre-store information for future tokens at the expense of the current token" (very much)
-
Record Manager: Maintaining Updated Vector Embeddings in RAG
By
–
Upserting documents to vector database is the key of building a RAG solution.
— FlowiseAI (@FlowiseAI) 4 avril 2024
❓But.. how to maintain an up-to-date vector embeddings with updated documents?
🗄️Record Manager! Document chunks will be hashed, stored and compared to prevent duplication
Available in 1.6.3! pic.twitter.com/dQzkTrn9SlUpserting documents to vector database is the key of building a RAG solution. But.. how to maintain an up-to-date vector embeddings with updated documents? Record Manager! Document chunks will be hashed, stored and compared to prevent duplication Available in 1.6.3!
-
Enterprise AI Applications: LLMs Customization and Inference Deep Dive
By
–
At this #GoogleCloudNext session, experts will delve into the complexities of building enterprise #AI applications, including customization, evaluation, and inference of large language models (#LLMs). Don’t miss this deep dive, register now. https://
nvda.ws/3PM0xW4 -
Paper on Pre-caching and Thinking Ahead in Language Models
By
–
oh and here's a link to the paper. includes a lot of theorems, experiments on a synthetic problem to explicitly test pre-caching (one sort of "thinking ahead") and experiments on language.
-
Language Models Predict Next-Next Tokens From Hidden States
By
–
and confusingly, we can predict the next-next token from hidden states at a given timestep. it just turns out that this is seems to be mostly a property of language — for example if you say "New" then it's somewhat likely that "City" is two tokens away