dots studio@rednote explore broader intelligences to make AI a more natural companion to humanity
- Tweets 9
- Following 1
- Followers 178
- Likes 1
Pinned Tweet
Drop an agent into the unknown—no rules, goals, or instructions.
On ARC-AGI-3, frontier LLMs with CoT score <1%. One TELL agent reaches 43.9%.
TELL explores, tests, and learns a world model of the unknown.
An early spark of lifelong learning in AI.
Project page in first reply.
1
1
4
1,751
What’s next:
TELL is an early signal, not the endpoint. The next step is faster hypothesis formation, more reliable verification, and more reusable world models.
Code, report, and replays in the link below.
1
1
113
Drop an agent into the unknown—no rules, goals, or instructions.
On ARC-AGI-3, frontier LLMs with CoT score <1%. One TELL agent reaches 43.9%.
TELL explores, tests, and learns a world model of the unknown.
An early spark of lifelong learning in AI.
Project page in first reply.
1
1
4
1,751
🔗 Resources
GitHub: github.com/rednote-hilab/dot…
Hugging Face: huggingface.co/rednote-hilab…
Demo: huggingface.co/spaces/rednot…
Try it out and let us know what you think
Introducing dots.vlm1, our first vision-language model with a 1.2B NaViT encoder (from scratch) and DeepSeek V3 LLM.
📊 Near-SOTA on MMMU, MathVista, OCR-Reasoning
🧠 Strong on AIME, LiveCodeBench, GPQA
🔗 github.com/rednote-hilab/dot…
Demo: huggingface.co/spaces/rednot…
#rednote #hilab
6
972
Introducing dots.vlm1, our first vision-language model with a 1.2B NaViT encoder (from scratch) and DeepSeek V3 LLM.
📊 Near-SOTA on MMMU, MathVista, OCR-Reasoning
🧠 Strong on AIME, LiveCodeBench, GPQA
🔗 github.com/rednote-hilab/dot…
Demo: huggingface.co/spaces/rednot…
#rednote #hilab
2
5
1,674