The popularity of Large Language Models (LLMs) have unleashed a new age of Language Agents for solving a diverse range of tasks. While contemporary frontier LLMs are capable enough to power reasonably good Language agents, the closed-API model makes it hard to improve in cases they perform sub-optimally. To address this, recent works have explored techniques to improve their performance using self reflection and prompt optimization techniques. While techniques like self reflection work well in an online setup, contemporary prompt optimization techniques are designed to work on simpler tasks. To address this, we introduce METAREFLECTION, a novel offline reinforcement learning technique that enhances the performance of Language Agents by augmenting a semantic memory based on experiential learnings from past trials. We demonstrate the efficacy of METAREFLECTION by evaluating across multiple domains, including complex logical reasoning, biomedical semantic similarity, open world question answering, and vulnerability threat detection, in Infrastructure-as-Code, with different agent design. METAREFLECTION boosts Language agents’ performance by 4 % to 16.82 % over the raw GPT-4 baseline and performs on par with existing state-of-the-art prompt optimization techniques while requiring fewer LLM calls.

In this paper we examine the limitations of Large Language Models (LLMs) for complex reasoning tasks. While current approaches leverage formal languages as intermediate representation for these reasoning problems, they still struggle with generating intermediate for-mal specifications with great correctness and in refining these representations. To address these issues, this paper proposes Logic-LM++, an improvement on Logic-LM (Pan et al., 2023). It uses the ability of LLMs to do pairwise comparisons, allowing the evaluation of the refinements suggested by the LLM. The paper demonstrates that Logic-LM++ outperforms Logic-LM and LLM based techniques on natural language reasoning tasks on two datasets, FOLIO, ProofWriter and AR-LSAT. Logic-LM++ show an average improvement of 18.5% on standard prompting, 12.3% on chain of thought prompting and 5% on Logic-LM.

Target similarity tuning (TST) is a method of selecting relevant examples in natural language (NL) to code generation through large language models (LLMs) to improve performance. Its goal is to adapt a sentence embedding model to have the similarity between two NL inputs match the similarity between their associated code outputs. In this paper, we propose different methods to apply and improve TST in the real world. First, we replace the sentence transformer with embeddings from a larger model, which reduces sensitivity to the language distribution and thus provides more flexibility in synthetic generation of examples, and we train a tiny model that transforms these embeddings to a space where embedding similarity matches code similarity, which allows the model to remain a black box and only requires a few matrix multiplications at inference time. Second, we how to efficiently select a smaller number of training examples to train the TST model. Third, we introduce a ranking-based evaluation for TST that does not require end-to-end code generation experiments, which can be expensive to perform.

Recent progress in deep learning has primarily been fueled by the availability of large amounts of annotated data that is obtained from highly expensive manual annotating pro-cesses. To tackle this issue of availability of annotated data, a lot of research has been done on unsupervised domain adaptation that tries to generate systems for an unlabelled target domain data, given labeled source domain data. However, the availability of annotated or labelled source domain dataset can’t always be guaranteed because of data-privacy issues. This is especially the case with medical data, as it may contain sensitive information of the patients. Source-free domain adaptation (SFDA) aims to resolve this issue by us-ing models trained on the source data instead of using the original annotated source data. In this work, we try to build SFDA systems for semantic processing by specifically focusing on the negation detection subtask of the SemEval2021 Task 10. We propose two approaches -ProtoAUGandAdapt-ProtoAUGthat use the idea of self-entropy to choose reliable and high confidence samples, which are then used for data augmentation and subsequent training of the models. Our methods report an improvement of up to 7% in F1 score over the baseline for the Negation Detection subtask.