AMD
acquires
nod.ai
to close the gap with
英伟达
.
Meta
and
Character.AI
’s chat agents need a time out.
微软
releases LongLLMLingua paper and code. Let’s dive in!
- AMD acquires open-source AI software pioneer Nod.ai: AMD is set to acquire AI software optimization startup Nod.ai in a bid to expand its presence in the AI chip market. The purchase is part of AMD's growth strategy in the AI sector, centered on an open software ecosystem. The acquisition will enhance AMD's ability to provide AI customers with open software for deploying high-performing AI models on AMD hardware.
- The Problems Lurking Behind "fun" Chatbots: AI chatbot startup Character AI is expanding its conversational characters to a group chat function that allows users to interact with their favorite celebrities. This move comes two weeks after Meta debuted its own AI characters across various platforms. While AI consumer innovation is exciting, issues have already arisen, such as AI characters behaving inappropriately and failing to acknowledge their AI status, raising questions about the long-term viability of these products.
- Adobe unveils Firefly 2, new AI features:
Adobe
has announced new AI products and features at its annual Adobe MAX conference, including "Firefly Image 2" which offers improved prompt understanding and photorealism. The company is intensifying its rivalry with
Canva
by introducing features such as Generative Match, which allows users to generate imagery in a particular style from a reference image, and "New Firefly Design Model" that enables users to instantly generate design templates for various purposes. Adobe also previewed several other AI-powered projects, including object-aware photo editing, text-to-vector graphic generation, and AI-generated video dubbing.
- LongLLMLingua: Accelerating and Enhancing LLMs in Long Context Scenarios via Prompt Compression: This paper introduces LongLLMLingua, a method for compressing prompts in large language models (LLMs) to address the challenges of computational cost, latency, and performance in long context scenarios. The proposed method improves LLMs' perception of key information and achieves higher performance with reduced cost and latency. Experimental results demonstrate significant performance boosts and cost savings across various tasks, and the code for LongLLMLingua is publicly available on GitHub.
- Language Agent Tree Search Unifies Reasoning Acting and Planning in Language Models: This paper introduces a framework called LATS (Language Agent Tree Search) that combines the capabilities of large language models (LLMs) in planning, acting, and reasoning. LATS repurposes the strengths of LLMs for enhanced decision-making by utilizing them as agents, value functions, and optimizers. The use of an environment for external feedback is crucial in this method, which allows for a more deliberate and adaptive problem-solving mechanism. Experimental evaluation across various domains shows the applicability and effectiveness of LATS, achieving high scores in programming and web browsing tasks.
- What Do You Get When You Cross Beam Search with Nucleus Sampling?: This paper introduces two deterministic nucleus search algorithms for natural language generation that combine beam search and nucleus sampling. The first algorithm, p-exact search, prunes the next-token distribution and performs an exact search within the remaining space. The second algorithm, dynamic beam search, adjusts the beam size based on the entropy of the candidate's probability distribution. Experimental results demonstrate that both algorithms achieve similar performance levels as standard beam search in machine translation and summarization benchmarks.
- LREC-COLING 2024: Joint International Conference on Computational Linguistics, Language Resources and Evaluation Submission Deadline: Sat Oct 14 2023 07:59:59 GMT-0400
- AISTATS 2024: Artificial Intelligence and Statistics 2024 Submission Deadline: Sat Oct 14 2023 07:59:59 GMT-0400
- CLeaR 2024: Causal Learning and Reasoning 2024 Submission Deadline: Sat Oct 28 2023 07:59:59 GMT-0400
- CVPR 2024: The IEEE/CVF Conference on Computer Vision and Pattern Recognition 2024 Submission Deadline: Sat Nov 11 2023 02:59:59 GMT-0500
- ICAPS 2024: The International Conference on Automated Planning and Scheduling 2024 Submission Deadline: Thu Dec 14 2023 06:59:59 GMT-0500
Want to learn more from Lightning AI? “Subscribe” to make sure you don’t miss the latest flashes of inspiration, news, tutorials, educational courses, and other AI-driven resources from around the industry. Thanks for reading!
