Trending

Active discussions, ranked by recent helpful reactions and replies.

Questionon Attention Is All You Need

Can someone build my intuition for scaled dot-product attention (and that √d_k)?

NLPTheory
5 17about 2 months ago

by priyanair

Research Ideaon Highly accurate protein structure prediction with AlphaFold

Hot one: will AI structure prediction make experimental structural biology obsolete?

Bioinformatics
5 10about 2 months ago

by dortiz

Critiqueon BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding

Is measured LLM/NLP progress real, or are leaderboards just measuring contamination?

NLPStatistics
5 9about 1 month ago

by amir_r

Critiqueon Highly accurate protein structure prediction with AlphaFold

Near-experimental accuracy ≠ understanding — how much is interpolation over the PDB/MSA?

BioinformaticsML Systems
3 7about 2 months ago

by hcole

Explanationon Attention Is All You Need

Did dropping recurrence throw away a useful inductive bias?

TheoryNLP
3 7about 2 months ago

by lenaf

Research Ideaon Estimating the reproducibility of psychological science

Hot take: peer review filters for novelty and narrative, not truth — is it fundamentally broken?

StatisticsPsychology
4 5about 1 month ago

by weizh

Critiqueon Learning Transferable Visual Models From Natural Language Supervision

Do these models 'understand', or are they stochastic parrots? (the question won't die)

Computer VisionCognitive Science
4 5about 1 month ago

by dortiz

Critiqueon Attention Is All You Need

Hot take: 'scale is all you need' is becoming an ideology — are we hitting a wall?

ML SystemsTheory
4 5about 1 month ago

by tbecker

Critiqueon BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding

Did the pivot to decoder-only LLMs quietly regress on understanding tasks?

NLP
3 6about 2 months ago

by amir_r

Questionon Chain-of-Thought Prompting Elicits Reasoning in Large Language Models

Is chain-of-thought just a better input distribution, or something deeper?

NLPTheory
4 3about 1 month ago

by lenaf

Reproductionon Chain-of-Thought Prompting Elicits Reasoning in Large Language Models

Reproduction note: GSM8K with GPT-3.5-turbo

NLPML Systems
5 1about 1 month ago

by tbecker

Critiqueon Controlling the False Discovery Rate: A Practical and Powerful Approach to Multiple Testing

Should we abolish statistical significance (p<0.05)? 800+ statisticians say yes — I'm not sure

Statistics
3 4about 1 month ago

by weizh

Research Ideaon Lost in the Middle: How Language Models Use Long Contexts

If models are 'lost in the middle', is RAG re-ranking solving the wrong problem?

NLP
3 4about 2 months ago

by jihoon

Research Ideaon Controlling the False Discovery Rate: A Practical and Powerful Approach to Multiple Testing

Is FDR control a band-aid for a preregistration problem?

Statistics
3 4about 2 months ago

by weizh

Reproductionon Denoising Diffusion Probabilistic Models

Reproduction note: training seems sensitive to seed

Computer VisionML Systems
4 2about 1 month ago

by tbecker

Reproductionon Denoising Diffusion Probabilistic Models

Reproduction note: training seems sensitive to seed

Computer VisionML Systems
4 2about 2 months ago

by tbecker

Questionon Attention Is All You Need

학습형 포지셔널 임베딩 vs 사인파 인코딩, 실제 차이가 클까요?

NLPTheory
3 3about 2 months ago

by jihoon

Questionon Attention Is All You Need

Why does the attention mechanism scale by sqrt(d_k)?

NLPTheory
3 2about 1 month ago

by amir_r

Critiqueon Scaling Laws for Neural Language Models

The compute-optimal frontier ignores inference cost

ML Systems
3 2about 1 month ago

by tbecker

Research Ideaon Learning Transferable Visual Models From Natural Language Supervision

Are prompt templates quietly inflating the zero-shot numbers?

Computer VisionNLP
3 2about 2 months ago

by yukis