Autoregressive protein language models (pLMs) have become transformative tools for designing functional proteins with remarkable diversity, demonstrating success in creating enzyme families like lysozymes and carbonic anhydrases. These models generate protein sequences by sampling from learned probability distributions, uncovering intrinsic patterns within training datasets. Despite their ability to explore high-quality subspaces of the sequence landscape,…
The rapid development of Large Language Models (LLMs) has transformed natural language processing (NLP). Proprietary models like GPT-4 and Claude 3 have set high standards in terms of performance but often come with drawbacks such as high costs, limited accessibility, and opaque methodologies. Meanwhile, many so-called open-source models fail to fully embody the ideals of…
Natural Language processing uses large language models (LLMs) to enable applications such as language translation, sentiment analysis, speech recognition, and text summarization. These models depend on human feedback-based supervised data, but relying on unsupervised data becomes necessary as they surpass human capabilities. However, the issue of alignment arises as the models get more complex and…
Large Language Models (LLMs) and neural architectures have significantly advanced capabilities, particularly in processing longer contexts. These improvements have profound implications for various applications. Enhanced context handling enables models to generate more accurate and contextually relevant responses by utilizing comprehensive information. The expanded context capacity has significantly strengthened in-context learning capabilities, allowing models to utilize…
Large Language Models (LLMs) play a vital role in many AI applications, ranging from text summarization to conversational AI. However, evaluating these models effectively remains a significant challenge. Human evaluations, while reliable, often suffer from inconsistency, high costs, and long turnaround times. Automated evaluation tools, particularly those that are closed-source, frequently lack transparency and fail…
Theory of Mind (ToM) is a foundational element of human social intelligence, enabling individuals to interpret and predict the mental states, intentions, and beliefs of others. This cognitive ability is essential for effective communication and collaboration, serving as a pillar for complex social interactions. Developing systems that emulate this reasoning in AI is crucial for…
Reasoning systems such as o1 from OpenAI were recently introduced to solve complex tasks using slow-thinking processes. However, it is clear that large language models have limitations, as they cannot plan, break down problems, improve ideas, summarize, or rethink due to their training and methods. While these tools try to enhance reasoning, they depend on…
The evaluation of LLMs in medical tasks has traditionally relied on multiple-choice question benchmarks. However, these benchmarks are limited in scope, often yielding saturated results with repeated high performance from LLMs, and do not accurately reflect real-world clinical scenarios. Clinical reasoning, the cognitive process physicians use to analyze and synthesize medical data for diagnosis and…