tiny_language's profile picture. TinyLLM: Bringing language models to constrained edge devices for supporting embedded sensing applications

TinyLLM

@tiny_language

TinyLLM: Bringing language models to constrained edge devices for supporting embedded sensing applications

TinyLLM 已轉發

Automating the Search for Artificial Life with Foundation Models arxiv.org/abs/2412.17799


TinyLLM 已轉發

Whoa! In @NEJM_AI: A Multimodal Biomedical Foundation Model Trained from 15 Million Image–Text Pairs… fully open-access foundation model! ai.nejm.org/doi/full/10.10…


TinyLLM 已轉發

150GB of math dataset, holy shit

maharshii's tweet image. 150GB of math dataset, holy shit

TinyLLM 已轉發

🧠💡 Our LLMs just had a ‘memory augmentation’—now they can deliberate like seasoned thinkers! arxiv.org/abs/2412.17747

PfeiffJo's tweet image. 🧠💡 Our LLMs just had a ‘memory augmentation’—now they can deliberate like seasoned thinkers!

arxiv.org/abs/2412.17747

TinyLLM 已轉發

Sloth: scaling law for LLM skills predicts benchmark scores (even from a single model per family). It is based on factor analysis and has family-specific parameters, making it interpretable and accurate! Paper: arxiv.org/abs/2412.06540 GitHub: github.com/felipemaiapolo… 🧵1/8

felipemaiapolo's tweet image. Sloth: scaling law for LLM skills predicts benchmark scores (even from a single model per family). It is based on factor analysis and has family-specific parameters, making it interpretable and accurate!

Paper: arxiv.org/abs/2412.06540
GitHub: github.com/felipemaiapolo…

🧵1/8

TinyLLM 已轉發

Our new paper "Fooling LLM graders into giving better grades through neural activity guided adversarial prompting" expertly lead by @atsushi_y1230 arxiv.org/abs/2412.15275 With the potential rise of automated grading, we examine the fragility of these grading systems to attacks…

Excited to share our latest work, "Fooling LLM graders into giving better grades through neural activity-guided adversarial prompting" (w/ @SuryaGanguli)! We investigate distorting AI decision-making to build fair and robust AI judges/graders.arxiv.org/abs/2412.15275 #AISafety 1/n



TinyLLM 已轉發

I'm more and more confident that tokenization will be gone. Humans don't think in "tokens". Tokens are hardcoded abstractions in LLMs that lead to weird behavior: LLMs can solve PhD-level math questions but cannot answer "Is 9.9 > 9.11?" Meta is shifting LLMs to LCMs (Large…

Yuchenj_UW's tweet image. I'm more and more confident that tokenization will be gone.

Humans don't think in "tokens".

Tokens are hardcoded abstractions in LLMs that lead to weird behavior: LLMs can solve PhD-level math questions but cannot answer "Is 9.9 > 9.11?"

Meta is shifting LLMs to LCMs (Large…

TinyLLM 已轉發

2023: the year prompting engineering was deprecated 2024: the year supervised finetuning was deprecated what will be in 2025?


此帳戶尚無追隨者

United States 趨勢

Loading...

Something went wrong.


Something went wrong.