RESEARCH

→ View original post on X — @learnopencv

Exploring AI, Love, Art, and Science Themes

By

@learnopencv

–

09 February 2023 18h59

Yes, the theme is AI, Love, Art, and Science.

Predicting Large Model Performance Across Unseen Tasks

By

–

09 February 2023 18h56

(3 cont.) Can you predict the performance of one model from another?
Can you predict the performance of a 128B model on an unseen task, given models up to some smaller threshold size and some performances of 128B models on other tasks?

Hope for Better Open-Source Large Language Models

By

–

09 February 2023 18h56

Finally, with models like OPT and BLOOM at 175B params, I'm hopeful that better open-source models will be soon be trained, given the pace of innovation in the space.

BIG-Bench metrics deserve deeper analysis and study

By

–

09 February 2023 18h56

(3) Even just looking at BIG-Bench metrics is quite understudied IMO. There are hundreds of tasks in BIG-Bench, and each task has dozens of models evaluated, each with many evaluation metrics. There are task logs for some models. This raises natural questions:

Flan-T5 Public Release with Instruction Tuning Data Ablation

By

–

09 February 2023 18h56

(2 cont.) Flan-t5 is publicly available along with the dataset (thanks to @ShayneRedford
), so you can even ablate parts of the instruction tuning data to see what unlocks performance.

Unlocking emergent abilities in smaller language models like Flan-T5

By

–

09 February 2023 18h56

(2) One key question of emergence is how to unlock such abilities at smaller scales, and I don't think there has been much work on flan-T5, which often beats PaLM 62B as shown in this paper: https://
arxiv.org/abs/2210.11416

→ View original post on X — @willknight

WIRED Fast Forward: AI Search Wars Begin

By

@willknight

–

09 February 2023 18h43

I am now writing the @WIRED Fast Forward newsletter. (I'm only a few years late to the trend). The first issue is about the AI search wars. You can see a preview and subscribe here: