Tag: ai/ml

Search-augmented large language models (LLMs) excel at knowledge-intensive tasks by integrating external retrieval. However, they often over-search – unnecessarily invoking search tool even when it does not improve response quality, which leads to computational inefficiency and hallucinations by incorporating irrelevant context. In this work, we conduct a systematic evaluation of over-searching across multiple dimensions, including …

Read more “Over-Searching in Search-Augmented Large Language Models”

Unified multimodal Large Language Models (LLMs) that can both understand and generate visual content hold immense potential. However, existing open-source models often suffer from a performance trade-off between these capabilities. We present Manzano, a simple and scalable unified framework that substantially reduces this tension by coupling a hybrid image tokenizer with a well-curated training recipe. …

Read more “MANZANO: A Simple and Scalable Unified Multimodal Model with a Hybrid Vision Tokenizer”

Recent advances in test-time alignment methods, such as Best-of-N sampling, offer a simple and effective way to steer language models (LMs) toward preferred behaviors using reward models (RM). However, these approaches can be computationally expensive, especially when applied uniformly across prompts without accounting for differences in alignment difficulty. In this work, we propose a prompt-adaptive …

Read more “AdaBoN: Adaptive Best-of-N Alignment”

Sprostowanie: Publikacja fałszywych twierdzeń nt. Palantira i Rządu Szwajcarskiego przez czasopismo ‘Die Republik’ Wprowadzenie Artykuł opublikowany w grudniu 2025 r. w Republik zwraca uwagę na raport Sztabu Sił Zbrojnych Szwajcarii (Armeestab) z 2024 r., w którym oceniono możliwość wdrożenia oprogramowania opracowanego przez firmę Palantir. Artykuł przedstawia sytuację w sposób fałszywy i wprowadzający w błąd, zgodnie …

Read more “Publikacja fałszywych twierdzeń nt. Palantira i Rządu Szwajcarskiego przez czasopismo Die Republik”

Despite being trained on vast amounts of data, most LLMs are unable to reliably generate well-designed UIs. Designer feedback is essential to improving performance on UI generation; however, we find that existing RLHF methods based on ratings or rankings are not well-aligned with designers’ workflows and ignore the rich rationale used to critique and improve …

Read more “Improving User Interface Generation Models from Designer Feedback”