Improving prompt quality is crucial for enhancing the performance of large language models (LLMs), particularly for Black-Box models like GPT4. Existing prompt refinement methods, while effective, often suffer from semantic inconsistencies between refined and original prompts, and fail to maintain users’ real intent. To address these challenges, we propose a self-instructed in-context learning framework that generates reliable derived prompts, keeping semantic consistency with the original prompts. Specifically, our framework incorporates a reinforcement learning mechanism, enabling direct interaction with the response model during prompt generation to better align with human preferences. We then formulate the querying as an in-context learning task, combining responses from LLMs with derived prompts to create a contextual demonstration for the original prompt. This approach effectively enhances alignment, reduces semantic discrepancies, and activates the LLM’s in-context learning ability for generating more beneficial response. Extensive experiments demonstrate that the proposed method not only generates better derived prompts but also significantly enhances LLMs’ ability to deliver more effective responses, particularly for Black-Box models like GPT4.

This paper addresses the critical need for democratizing large language models (LLM) in the Arab world, a region that has seen slower progress in developing models comparable to state-of-the-art offerings like GPT-4 or GPT-3.5, due to a predominant focus on mainstream languages (e.g., English and Chinese). One practical objective for Arabic LLMs is to utilize Arabic-specific vocabulary in the tokenizer to accelerate decoding. However, using a different vocabulary often leads to degradation of the model’s learned knowledge, since many words become out-of-vocabulary (OOV) at the beginning of training. Inspired by the vocabulary learning during Second Language (Arabic) Acquisition for humans, the released AraLLaMA employs progressive vocabulary expansion, which is implemented by a modified BPE algorithm that progressively extends the Arabic subwords in its dynamic vocabulary during training, thereby balancing the OOV ratio at every stage. The ablation study demonstrated the effectiveness of Progressive Vocabulary Expansion.Moreover, AraLLaMA achieves decent performance comparable to the best Arabic LLMs across a variety of Arabic benchmarks. Our model weights are available at: https://github.com/FreedomIntelligence/AraLLaMa.

Despite the substantial advancements in artificial intelligence, large language models (LLMs) remain being challenged by generation safety. With adversarial jailbreaking prompts, one can effortlessly induce LLMs to output harmful content, causing unexpected negative social impacts. This vulnerability highlights the necessity for robust LLM red-teaming strategies to identify and mitigate such risks before large-scale application. To detect specific types of risks, we propose a novel red-teaming method that **A**ttacks LLMs with **T**arget **Toxi**c **A**nswers (**Atoxia**). Given a particular harmful answer, Atoxia generates a corresponding user query and a misleading answer opening to examine the internal defects of a given LLM. The proposed attacker is trained within a reinforcement learning scheme with the LLM outputting probability of the target answer as the reward. We verify the effectiveness of our method on various red-teaming benchmarks, such as AdvBench and HH-Harmless. The empirical results demonstrate that Atoxia can successfully detect safety risks in not only open-source models but also state-of-the-art black-box models such as GPT-4o.