A new family of antibiotics discovered with graph deep learning.
RESEARCH
-
AI-Generated Arguments Influencing Political Opinion Change
By
–
Top AI stories of 2023: Can AI-generated arguments change minds on hot-button political issues? Unfortunately, yes. https://
stanford.io/3TuJsTf -
Constitutional AI Named Most Important AI Innovation 2023
By
–
Thank you @time for recognizing Constitutional AI as one of the most important AI innovations of 2023. CAI helps avoid harmful outputs by adhering to defined principles and behaviors, improving the alignment and transparency of AI systems. Learn more: https://
anthropic.com/index/claudes-
constitution
… -
Google Magic Eraser Enhanced with Generative AI Inpainting
By
–
Magic Eraser now uses gen AI to fill in detail when users remove unwanted objects from photos. Google Research worked on the MaskGIT generative image transformer for inpainting, and improved segmentation to include shadows and objects attached to people. pic.twitter.com/GoVtZTxT0y
— Google AI (@GoogleAI) 21 décembre 2023Magic Eraser now uses gen AI to fill in detail when users remove unwanted objects from photos. Google Research worked on the MaskGIT generative image transformer for inpainting, and improved segmentation to include shadows and objects attached to people.
-
AI Companies’ True Priorities Revealed Through Their Evaluation Choices
By
–
If you want to know what an AI company truly cares about, just look at what evals they run. – Lots of academic evaluations against GPT-4 means pressure to train a model more capable than OpenAI.
– Companies that actually care about their consumer product will probably run lots -
Paper Reports Non-Deterministic Logprobs from OAI API Despite Seed Parameter
By
–
Update: But not for logprobs. Paper published today by @maksym_andr reports logprobs returned from OAI API are non-deterministic even with the seed parameter. Not sure how to reconcile this with sampled completions described as deterministic.
-
Most Used Evaluation Metrics for AI Model Assessment
By
–
Most used evaluation metrics We see that most people still care primarily about correctness (and not other guardrails like toxicity or prompt injection) https://
docs.smith.langchain.com/evaluation/eva
luator-implementations
… -
Advanced Retrieval Strategies for AI Systems
By
–
Top Advanced Retrieval Strategies We primarily see people rolling their own retrieval, speaking to need to customize retrieval for each use case. Of general strategies, we see: 1. Self Query https://
python.langchain.com/docs/modules/d
ata_connection/retrievers/self_query
…
2. Hybrid Search
3. Contextual Compression -
Testing LLM Applications: Metrics and Evaluation Methods
By
–
How are people testing? Testing LLM Apps is hard. How are people doing it? 1. We see that 83% of test runs have some form of feedback, suggesting that most people are finding some metrics to eval (rather than just eyeball) 2. We see an average of 2.3 feedback per run,
-
LangChain State of AI 2023: Real Usage Data Analysis
By
–
LangChain State of AI 2023 What are people building? Which LLMs are they using? Which vectorstores are they using? How are they testing? We turn to anonymized usage stats from LangSmith to answer these questions with real data. Blog: https://
blog.langchain.dev/langchain-stat
e-of-ai-2023/
… Recap