Olmo

Headlines · 3

1. 5d
   Study identifies six linguistic patterns that make AI text harder to read

   Researchers from New York University and the University of Texas at Austin analyzed text from 14 language models, including 120 conversations, 45 coding sessions and 980 essays, and compared it with human writing. They identified six recurring constructions that the models overused: unusual word combinations, widely separated related sentence elements, shortened references with dropped connectors, stacked nouns, nouns converted into verbs, and animate verbs applied to inanimate subjects. Llama and the Olmo base model were closest to human text. In a proof-of-concept experiment, a smaller model trained with a penalty for longer answers favored shorter phrasing under certain conditions.

   The Decoder · 🔥 0
2. Oct 20
   Ai2's FairStress benchmark tests how readily LLMs abandon correct decisions

   Ai2 (the Allen Institute for AI) introduced FairStress, a benchmark that measures demographic fairness in LLMs beyond the final choice a model makes, by testing how readily a model abandons a well-supported decision when pressed with persuasive arguments it calls adversarial prefixes. Ai2 says it evaluated 15 open-weight and frontier models, including Olmo, Llama, Qwen, Gemma, Claude and the GPT family, and reports an average gap in abandoning correct answers of 6.9 percentage points when demographic identity was implied and 9.3 points when it was explicitly labeled; on FairStress-Core, models including Claude Opus 5 and GPT-5.5 picked the minority-coded candidate about 60–65% of the time when neither candidate was objectively stronger. FairStress is available for download and is being built into Ai2's Olmo-eval evaluation workbench.

   Ai2 Blog · 🔥 0
3. Oct 16
   Tsinghua open-sources Puro-2B, a low-cost pretraining recipe trained on RTX 5090s

   Tsinghua University researchers led by Chen Wenguang and Lü Kaifeng have released Puro-2B, an end-to-end open pretraining package that includes training data, framework, configuration and model weights. The team trained a roughly 2-billion-parameter model from scratch on a cluster of consumer RTX 5090 GPUs, using up to 1.4 trillion tokens. It reports that a version with about $4,400 in pretraining compute cost beat Qwen2-1.5B on the average of 15 tasks, while a $6,900 best version came close to Qwen2.5-1.5B; these are team-reported results without independent replication.

   36氪 人工智能 · 🔥 0

Experience and discussion from the community

Share my Olmo experienceAsk about Olmo