Your GenAI model might be underperforming, and you wouldn’t know. Outdated eval methods miss hidden failure modes in agentic systems. Rubric-based evaluation fixes that. What it is & why it matters https://
snorkel.ai/blog/data-qual
ity-and-rubrics-how-to-build-trust-in-your-models/
…
MACHINE LEARNING
-
Improving GenAI Model Evaluation: Rubric-Based Methods for Agentic Systems
By
–
-
Neo’s Agentic ML Engineer Tool Launch
By
–
check out Neo’s agentic ML engineer https://t.co/DRFRsweSUb
— Yohei (@yoheinakajima) 30 juillet 2025check out Neo’s agentic ML engineer
-
Neo: First Autonomous ML Engineer Agent Released
By
–
The first autonomous ML Engineer Agent has been released by @withneo! Neo is a multi-agent system that thinks, learns and builds like a real engineer.
— 🚨 AI News | TestingCatalog (@testingcatalog) 30 juillet 2025
– Perform analysis
– Finetune Llama and Gemma LLMs
– Prepare a visualisation
– Prepare training and evaluation pipelines https://t.co/1vvK4ddytw pic.twitter.com/PXN8WnRvYCThe first autonomous ML Engineer Agent has been released by @withneo
! Neo is a multi-agent system that thinks, learns and builds like a real engineer. – Perform analysis
– Finetune Llama and Gemma LLMs
– Prepare a visualisation
– Prepare training and evaluation pipelines -
MIT Research: How Language Models Track Dynamic Scenarios
By
–
How do language models track dynamic scenarios, like completing code or guessing your responses? MIT research finds that LMs don't track "state changes" step by step — they use mathematical shortcuts that we can control to boost LMs' prediction skills: https://
bit.ly/4lJfYfK -
Pelican Benchmark: Five-Attempt Model Selection Process
By
–
I've been meaning to do a version of the pelican benchmark where each model gets five attempts and then something picks the "best" one which moves on to the next round
-
Training AI Models for Under One Dollar Now Possible
By
–
Yep! And you can train a model for literally $1 or less
-
Early LSTM Language Model Architecture Recognition from NLP Community
By
–
Special thanks to the NLP community for the recognition. This was my earliest work in "large" language model at the time , quoted from the paper "Our stacking LSTM models have 4 layers, each with 1000 cells, and 1000-dimensional embeddings" .
-
Early seq2seq internship advances in machine learning history
By
–
Congrats! I have fond memories from the early (& simpler?) days, and your internship pushing seq2seq forward.
-
OpenPipe’s RULER Enables LLM Training Without Labeled Data
By
–
Under the hood, this is powered by @OpenPipeAI
's ART library. Their RULER technique made this possible, as it doesn't require labeled output data. With no need for labeled outputs, we can just sample a LLM to generate scenarios to RL against. Absolutely crazy!