Data Machina #223
State of AI 2023. Large Multimodal Models. GenAI Algos at Deepmind. CogVML. vec2txt. Transformers for Time Series. Docker GenAI.
State of AI. It’s really difficult to keep up with the ton of new stuff happening in AI everyday. And it’s hugely time consuming to filter out so much bs, hype, marketing fluff, and fake research. GenAI ≠ to all AI…Many people have assumed that other than GenAI, there’s nothing interesting in DL/ML/ DS happening. I hope that’s not your view! Today I’m sharing two great reports on the state of AI that cover many aspects beyond GenAI. Enjoy!
6th Annual State of AI, 2023. IMO This is one of the best AI annual reports, produced by Nathan Benaich and his team. The report is a compilation of the most interesting things Nathan et al. have seen with a goal of triggering an informed conversation about the state of AI and its implication for the future. The report covers 5 sections:
Research & Technology breakthroughs and their capabilities
Commercial apps for AI and its business impact
Politics & Regulation of AI
AI Safety
Predictions 12-months ahead on what’s expected to happen in AI
The report is really comprehensive, and comes with great insights on topics like: the battle to beat proprietary models with smaller models, the scaling limits of current AI approaches, or the inevitable growth of multimodality.
Click here to read Sate of AI Report, 2023 (163 slides)
The Kaggle AI Report 2023. Perhaps, Kaggle has been a bit late to GenAI. And has remained the last bastion of “traditional or classic ML & Data Science.” In the meantime, others like Hugging Face have ridden the crazy wave of transformers and GenAI.
This report is great collection of essays written and submitted by the Kaggle community as part of a competition. The report is broken down into the seven sections listed below:
Generative AI
Text data
Image / video data
Tabular / time series data
Kaggle competitions
AI ethics
AI essays
All of the sections contain very interesting notebooks (make sure access them.) I’d say the report is obviously a sort of “Kaggle centric view of AI.” I miss some exciting stuff happening in several Dl/ML areas, like for example how transformers are now applied to time-series forecasting (see the special AI/ DL ResearchDocs section on time-series that I added in the links below.)
Click here to read Kaggle AI Report, 2023 (pdf, 71 slides + notebooks)
Have a nice week.
10 Link-o-Troned
the ML Pythonista
Deep & Other Learning Bits
[free course] Uni of Geneva Deep Learning (1097 slides, vids)
Batch Calibration for In-Context Learning & Prompt Engineering
AI/ DL ResearchDocs (Special on TS)
data v-i-s-i-o-n-s
MLOps Untangled
AI startups -> radar
ML Datasets & Stuff
Google SANPO: Multi-attribute Dataset for Urban Understanding
UltraFeedback Dataset: 64K Prompts for Human Preference Models
Postscript, etc
Tips? Suggestions? Feedback? email Carlos
Curated by @ds_ldn in the middle of the night.