It’s only 30 pages long, but it’s a concise summary of the overall LLM and Generative AI research trends. Including Gemini to Q*…
What’s surprising is that 8 out of 30 pages are a list of 330 references.
제목: From Google Gemini to OpenAI Q* (Q-Star): A Survey of Reshaping the Generative Artificial Intelligence (AI) Research Landscape
Summary:
This comprehensive research study explored the evolving environment of generative artificial intelligence (AI) with a particular emphasis on the innovative impact of supposed advances toward expert mixing (MoE), multimodal learning, and artificial general intelligence (AGI). The report critically reviewed the current state and future trajectory of generative artificial intelligence (AI), looked at how innovations such as Google’s Gemini and the anticipated OpenAI Q* project are reshaping research priorities and applications in different domains, and also included an analysis of the impact on generative AI research taxonomy. The report evaluated the computational challenges, scalability, and impact on real life of these technologies, while highlighting their potential to drive significant progress in fields such as healthcare, finance, and education. It also addressed new academic challenges posed by the proliferation of both AI-themed papers and AI-generated papers, and investigated their impact on peer review processes and academic communication. This study highlighted the importance of incorporating ethical and human-centered methods into AI development and making them conform to social norms and well-being, and described future AI research strategies focused on balanced and conscientious use of MoE, multimodal, and AGI in generative AI.
arXiv: https://arxiv.org/abs/2312.10868
Browse: https://browse.arxiv.org/pdf/2312.10868.pdf
PDF: https://arxiv.org/pdf/2312.10868.pdf
arXiv-vanity: https://www.arxiv-vanity.com/papers/2312.10868
Paper page: https://huggingface.co/papers/2312.10868
Papers with code: https://paperswithcode.com/paper/from-google-gemini-to-openai-q-q-star-a It’s been a year since mankind has been with ChatGPT, and now there’s an article like this.
“How to use ChatGPT to set innovative goals for 2024
- Simplify and improve your goal-setting process with ChatGPT.”
But it’s quite useful. Someone might use the method presented here for performance evaluation and employee counseling.
Step 1: Ask me
Step 2: Inform ChatGPT about yourself
“I want to set professional goals for 2024. Here’s my introduction.
“I’m a private entrepreneur focused on HR improvement. There are three main ways to do this. I write about HR and employment laws for multiple clients. I present webinars primarily focused on US compliance and AI use in HR. And thirdly, please give me a keynote speech. I have 44,000 followers on LinkedIn, 32,000 followers on Facebook, and a large social media presence with 12,000 newsletter subscribers. Can you tell me five goals that will help me increase my revenue in the coming year?”
Step 3: Ask follow-up questions
Step 4: Keep asking questions
Step 5: Repeat until you have the necessary information.
Link: https://www.inc.com/suzanne-lucas/how-to-use-chatgpt-to-set-transformative-goals-for-2024.html2024 is really about to open the era of On device LLM. With the recent increase in SLLM cases and installation attempts, it’s going to be a really exciting year.
A few days ago, a Canadian company called Spass.ai finally published a paper on the 70B model on arXiv, aiming for “LLM for medical devices.” Interestingly, he also has an AI-based patient shock prediction hardware product named ShockRange, has a partnership with Catholic St. Mary’s Hospital, and is headquartered in Korea.
In the era of Phi-2, Gemini-nano, SLLM, open-source LLM, and On device LLM, regulators’ concerns about how to verify and manage LLM installed in medical devices are likely to deepen.
제목: SM70: A Large Language Model for Medical Devices
Summary:
We present SM70, a large language model with 70 billion parameters specifically designed for SpassMed’s medical devices, under the brand name ‘JEE1’ (pronounced G1, meaning ‘life’). This large language model provides a more accurate and secure response to healthcare questions. To fine-tune SM70, we used approximately 800,000 data items from MedAlpaca, a publicly available dataset. The Llama270B open-source model became the basis for SM70, using QLoRA technology for fine-tuning. Evaluation is performed on three benchmark datasets: MEDQA – USMLE, PUBMEDQA, and USMLE, each representing unique aspects of medical knowledge and reasoning. The performance of SM70 contrasts with other notable LLMs including Llama2 70B, Clinical Camel 70 (CC70), GPT 3.5, GPT 4, and Med-Palm, providing comparative understanding of their capabilities within the healthcare domain. Our results show that SM70 outperforms several existing models on these datasets by demonstrating its ability to handle a wide range of medical queries, from fact-based questions derived from PubMed abstracts to complex clinical decision scenarios. Specifically, SM70’s robust performance on USMLE and PUBMEDQA datasets suggests its potential as an effective tool for clinical decision support and medical information retrieval. This paper highlights the need for further development, especially in tasks that require extensive medical knowledge and complex reasoning, as SM70 admits areas that lag behind GPT4, the most advanced model, despite promising results.
arXiv: https://arxiv.org/abs/2312.06974
Browse: https://browse.arxiv.org/pdf/2312.06974.pdf
PDF: https://arxiv.org/pdf/2312.06974.pdf
arXiv-vanity: https://www.arxiv-vanity.com/papers/2312.06974
Paper page: https://huggingface.co/papers/2312.06974
Papers with code: https://paperswithcode.com/paper/sm70-a-large-language-model-for-medical
Homepage: https://www.spass.ai/