Embarrassingly simple self-distillation improves code generation (arxiv.org) | HN Screenshots

YHacker Newsscreenshots

Screenshot of arxiv.org: Embarrassingly simple self-distillation improves code generation

First screen, captured 6 months ago.

Looks similar

Screenshot of arxiv.org: Self-Distillation Enables Continual Learning [pdf]

Self-Distillation Enables Continual Learning [pdf](arxiv.org)

78 points by teleforce 5 months ago | 20 comments

Screenshot of arxiv.org: Knowledge Distillation of Black-Box Large Language Models (2024)

Knowledge Distillation of Black-Box Large Language Models (2024)(arxiv.org)

115 points by babelfish 3 months ago | 22 comments

Screenshot of arxiv.org: An embarrassingly simple approach to recover unlearned knowledge for LLMs

An embarrassingly simple approach to recover unlearned knowledge for LLMs (arxiv.org)

232 points by PaulHoule 2 years ago | 105 comments

Screenshot of arxiv.org: Self-Adapting Language Models

Self-Adapting Language Models (arxiv.org)

210 points by archon1410 1 year ago | 61 comments

Screenshot of arxiv.org: Type-constrained code generation with language models

Type-constrained code generation with language models (arxiv.org)

237 points by tough 1 year ago | 109 comments

Screenshot of arxiv.org: A sleep-like consolidation mechanism for LLMs

A sleep-like consolidation mechanism for LLMs (arxiv.org)

178 points by juxtapose 5 months ago | 126 comments

Related stories

Screenshot of arxiv.org: Training language models to be warm and empathetic makes them less reliable

Training language models to be warm and empathetic makes them less reliable (arxiv.org)

334 points by Cynddl 1 year ago | 333 comments

arXiv paper finding that fine-tuning language models to be warm and empathetic raises error rates and makes them more sycophantic on safety-critical tasks.

llm sycophancy ai-safety fine-tuning

Screenshot of arxiv.org: Deep Learning Is Not So Mysterious or Different

Deep Learning Is Not So Mysterious or Different (arxiv.org)

445 points by wuubuu 1 year ago | 112 comments

arXiv paper by Andrew Gordon Wilson arguing deep learning generalization phenomena like benign overfitting and double descent are explainable by soft inductive biases.

deep-learning generalization double-descent pac-bayes

Screenshot of plf.inf.ethz.ch: Tree Borrows

Tree Borrows (plf.inf.ethz.ch)

557 points by zdw 1 year ago | 141 comments

ETH Zürich Programming Language Foundations Lab page for the PLDI 2025 paper Tree Borrows, an aliasing model for Rust by Villani, Hostert, Dreyer, and Jung.

rust tree-borrows unsafe-rust aliasing-model

Screenshot of arxiv.org: AI models can provably hide arbitrary capabilities

AI models can provably hide arbitrary capabilities (arxiv.org)

252 points by sbulaev 13 days ago | 118 comments

An arXiv paper showing AI models can embed encrypted neural network backdoors that provably hide arbitrary capabilities from elicitation-based safety evaluations.

ai-safety backdoors capability-evaluation cryptography

Screenshot of arxiv.org: C++ patterns for low-latency applications including high-frequency trading

C++ patterns for low-latency applications including high-frequency trading (arxiv.org)

343 points by chris_overseas 2 years ago | 200 comments

arXiv paper presenting C++ design patterns and benchmarks for low-latency code in high-frequency trading, including a Disruptor pattern implementation.

cpp low-latency high-frequency-trading design-patterns

Screenshot of eugeneyan.com: Improving recommendation systems and search in the age of LLMs

Improving recommendation systems and search in the age of LLMs (eugeneyan.com)

379 points by 7d7n 1 year ago | 88 comments

Eugene Yan's long-form survey of how industrial recommendation and search systems are using LLMs across model architecture, data generation, training paradigms, and unified frameworks.

recommendation-systems search llm information-retrieval