Hugging Face's logo
Activity Feed

AI & ML interests

RLHF, Model Evaluation, Benchmarks, Data Labeling, Human Feedback, Computer Vision, Image Generation, Video Generation, LLMs, Translations

Recent Activity
Rapidata-Bot  new activity about 6 hours ago
Rapidata/text-to-image-human-feedback-benchmark:Sync Benchmark.ai Image Benchmark
Rapidata-Bot  new activity about 10 hours ago
Rapidata/text-to-svg-human-feedback-benchmark:Sync SVG Benchmark
Rapidata-Bot  new activity about 12 hours ago
Rapidata/image-to-video-physics-human-feedback-benchmark:Sync Physics
View all activity

Articles

Rapidata 's collections 13

Text & LLM: human evaluation
Human judgements of LLM output: translation quality across 10+ languages and multilingual humour.
Video generation: human preferences & benchmarks
Human judgements of text-to-video and image-to-video models. Start with the consolidated 823k-vote set.
Benchmark Release Data
The vote data behind every live benchmark on benchmark.ai. One dataset per benchmark.
Text-to-Image Model Releases
One dataset per launch. A new image model ships, we put it in front of thousands of people, and publish how it actually did against other SOTA models.
Foundational Preference Sets
The 2M+ annotations our first benchmarks were built on: Flux, SD3, Midjourney and DALL·E 3, split into preference, coherence and prompt alignment.
Image generation: human preferences & benchmarks
Text-to-image, SVG and face-generation preference data behind benchmark.ai/image and benchmark.ai/svg.
Video Model Releases
One dataset per launch, text-to-video and image-to-video. Veo, Sora, Kling, Pika, Seedance, ranked by real humans.
Research & Human Behaviour
What people think when no model is involved: where the future sits in your head, which sounds feel round, how our crowd compares to Prolific.
Text & LLM: human evaluation
Human judgements of LLM output: translation quality across 10+ languages and multilingual humour.
Image generation: human preferences & benchmarks
Text-to-image, SVG and face-generation preference data behind benchmark.ai/image and benchmark.ai/svg.
Video generation: human preferences & benchmarks
Human judgements of text-to-video and image-to-video models. Start with the consolidated 823k-vote set.
Benchmark Release Data
The vote data behind every live benchmark on benchmark.ai. One dataset per benchmark.
Text-to-Image Model Releases
One dataset per launch. A new image model ships, we put it in front of thousands of people, and publish how it actually did against other SOTA models.
Video Model Releases
One dataset per launch, text-to-video and image-to-video. Veo, Sora, Kling, Pika, Seedance, ranked by real humans.
Foundational Preference Sets
The 2M+ annotations our first benchmarks were built on: Flux, SD3, Midjourney and DALL·E 3, split into preference, coherence and prompt alignment.
Research & Human Behaviour
What people think when no model is involved: where the future sits in your head, which sounds feel round, how our crowd compares to Prolific.