Writings

Essays and articles on data science, generative AI, and applied machine learning.

RL environments for LLM agents: Design, rewards, and validation

RL environments for LLM agents: Design, rewards, and validation

Learn how RL environments for LLM agents model state, actions, rewards, and validation, and how to design them without rewarding benchmark exploits.

Snorkel AI
Opus 5.5 vs Opus 5 vs Fable 5.1: Coding Benchmark Results

Opus 5.5 vs Opus 5 vs Fable 5.1: Coding Benchmark Results

Opus 5.5 posts the strongest results measured on Snorkel's frontier task set, with a pass rate of 68%, up from 61% for Opus 5 and 49% for Fable 5.1 on the same tasks.

Snorkel AI
Fable 5.1 on Frontier Coding Tasks: Efficient Successes, Distinct Failure Modes

Fable 5.1 on Frontier Coding Tasks: Efficient Successes, Distinct Failure Modes

Fable 5.1 leads Opus 5 on debugging (87%) and games (88%) in Snorkel's Terminal-Bench+ coding eval, using 58% fewer tokens per successful run.

Snorkel AI
Transformers: Encoder-Decoder Architecture

Transformers: Encoder-Decoder Architecture

transformersmachine-learningaideep-learning
Natural Language Processing: Text Representations

Natural Language Processing: Text Representations

deep-learningnlpdata-scienceai
Sequential Data Modeling: Text Generation with RNNs

Sequential Data Modeling: Text Generation with RNNs

data-sciencemachine-learningdeep-learningtext-generation
Grok 4.7 on Senior SWE-Bench: Strong pass@3, Cheaper Cost per Trial

Grok 4.7 on Senior SWE-Bench: Strong pass@3, Cheaper Cost per Trial

Grok 4.7 was evaluated on Senior SWE-Bench, Snorkel's benchmark for measuring whether coding agents work like senior engineers.

Snorkel AI

Senior SWE-Bench: A Brief Look Into Snorkel’s Cutting Edge SWE Benchmark

ai-agentaibenchmarkllm-evaluation
Natural Language Processing: Real World NLP with Word2Vec

Natural Language Processing: Real World NLP with Word2Vec

word2vecmachine-learningdeep-learningnlp
Deep Learning Training Enhancements: Residual Connections and ResNets

Deep Learning Training Enhancements: Residual Connections and ResNets

deep-learningdata-sciencemachine-learningai
Sequential Data Modeling: LSTM and GRU

Sequential Data Modeling: LSTM and GRU

machine-learninglstmdata-sciencedeep-learning
From Foundational Competency to Expert Performance: A Curriculum Approach to Model Development

From Foundational Competency to Expert Performance: A Curriculum Approach to Model Development

Curriculum learning orders coding-agent training tasks by measured difficulty, so failures reveal which skill is missing rather than just a lower score.

Snorkel AI
Transformers: Attention Mechanism

Transformers: Attention Mechanism

machine-learningdata-scienceaideep-learning
Natural Language Processing: Word Embeddings and Word2Vec

Natural Language Processing: Word Embeddings and Word2Vec

word-embeddingsdata-sciencenlpmachine-learning
Deep Learning Training Enhancements: Batch Normalization

Deep Learning Training Enhancements: Batch Normalization

batch-normalizationmachine-learningdeep-learningdata-science
More Articles On Medium