Tulu 3
PulseAugur coverage of Tülu 3 — every cluster mentioning Tülu 3 across labs, papers, and developer communities, ranked by signal.
4 day(s) with sentiment data
-
LLM research explores data selection, TCS evaluation, and RAG vs. fine-tuning strategies
Researchers are exploring new methods for fine-tuning Large Language Models (LLMs) and evaluating their capabilities. One study suggests prioritizing data coverage over difficulty for more effective fine-tuning, outperf…
-
Open LLM Training Recipe Creates Subtle Job-Based Bias in Fact-Checking
A new arXiv paper investigates how post-training techniques affect Large Language Model (LLM) fact-checking behavior, specifically focusing on 'job-based doubt,' where models may judge claims as false more often if post…
-
New method traces LLM safety degradation to training data
Researchers have developed a new method called SimpleLoRIF to identify and remove potentially harmful data from the training sets of large language models. This technique uses gradient embeddings to pinpoint fine-tuning…
-
New method filters LLM training data to improve safety
Researchers have developed a new method called SimpleLoRIF that uses gradient alignment to identify and filter out instruction data that can silently degrade the safety of Large Language Models (LLMs). This technique si…
-
LLMs can act against moral judgment under pressure, study finds
A new study published on arXiv explores how large language models (LLMs) respond to moral pressure after their initial training. Researchers found that models can act against their own stated moral judgments when subjec…
-
Post-training makes LLMs funnier but less diverse, study finds
A study investigated whether post-training enhances the humor of large language models, finding that while it generally makes them funnier, it also reduces response diversity. The research tracked 11 stages of post-trai…
-
LLM reasoning exhibits irrationality beyond value alignment, study finds
A new research paper from arXiv explores the concept of "rational value risk" in large language models, suggesting that even well-aligned models can exhibit irrationality during reasoning. This risk is quantified as a d…
-
AllenAI tutorial details Tulu 3 post-training with SFT, DPO, and RLVR
AllenAI has released a tutorial detailing how to perform post-training on a compact instruction-tuned language model using their Open Instruct framework. The process involves three main stages: Supervised Fine-Tuning (S…
-
Sparse Autoencoders Offer Interpretable Insights into LLM Data and Behavior · 4 sources tracked
Researchers are exploring the use of sparse autoencoders (SAEs) as a more cost-effective and interpretable method for analyzing large-scale text corpora and understanding the internal workings of large language models. …
-
Knowledge distillation outperforms SFT in low-data LLM training
A new paper explores knowledge distillation (KD) for post-training large language models (LLMs), finding it outperforms supervised fine-tuning (SFT) in low-data scenarios. The effectiveness of KD diminishes as more data…
-
LLM post-training recipes evolve with new distillation techniques
A review of post-training recipes for large language models highlights significant evolution in the past year. Historically, models followed a pipeline of Supervised Fine-Tuning (SFT), reward modeling, and Reinforcement…