ChainCatcher news, the decentralized full-chain data network Chainbase announced the open-source release of the cryptocurrency-centric large language model Theia-Llama-3.1-8B, with training data sourced from CoinMarketCap and detailed project research reports. Chainbase stated that this model has a lower perplexity and a higher BERT score compared to other mainstream models, utilizing LORA for efficient fine-tuning, allowing large pre-trained models to adapt to specific tasks with lower computational power, significantly reducing the model's memory footprint and accelerating inference speed while maintaining acceptable accuracy.According to previous reports, Chainbase completed a $15 million Series A financing round in July this year, led by Tencent Investment Group and Matrix Partners China.

ChainCatcher news, the full-chain data network Chainbase announced the open-source of its large language model Theia-Llama-3.1-8B specifically designed for the cryptocurrency field on HuggingFace.The Chainbase team has innovatively built the first professional Web3 dataset, which includes various information on the top 2000 projects from CoinMarketCap. The dataset has been manually and algorithmically filtered to ensure the accuracy, diversity, and professionalism of the training data. Based on this dataset, the team efficiently fine-tuned the model using LoRA technology and accelerated the training process with tools like DeepSpeed. The model has been quantized to the Q8 GGUF format, significantly reducing memory usage and improving inference speed.It is reported that Theia-Llama-3.1-8B is Chainbase's initial attempt at a large model in the cryptocurrency field, and this model has been successfully applied in Chainbase's DEMO interactive application TheiaChat, which currently has over 300,000 daily active users.

ChainCatcher news: At the R3al World DePIN Summit of Token2049 in Singapore, Rock, the founder of the DePIN & AI project Network3, announced the launch plan for its new local large language model (LLM) feature, which will significantly enhance the efficiency and performance of edge AI technology.Rock pointed out that although edge AI has deeply integrated into people's daily lives, the complexity and importance behind it are often overlooked. For example, the ability of smartphones to perform data processing tasks at night fully demonstrates the potential of edge devices to utilize idle resources for efficient computing. With the newly launched local LLM feature, Network3 will significantly optimize the efficient computing capabilities of smart devices during idle times, especially for local inference and processing without relying on cloud computing. This not only greatly reduces bandwidth consumption but also enhances data security and privacy protection.Network3 will provide strong support and vast resources for edge AI by integrating global idle edge device resources. The upcoming local LLM feature will allow users to seamlessly enjoy AI chat services on mobile devices without relying on expensive cloud computing infrastructure. Additionally, users can earn tokens by interacting with the model and customize and optimize algorithms based on personal needs, enhancing the personalized experience. According to Network3, the beta version is planned to be released in October, at which time users can download it from the official website for experience.It is reported that Network3 is helping global AI developers efficiently, conveniently, and economically perform large-scale inference, training, or validation of models by building an AI Layer2. Previously, it was announced that Network3 had completed $5.5 million in pre-seed and seed round financing, and the next round of financing has already begun, with several leading institutions participating.

ChainCatcher news, in a post on September 15, a Reddit user with the username "SentuBill" claimed that the ChatGPT large language model (LLM) initiated a conversation by asking him how his first day of school went.According to the conversation provided by the user, ChatGPT was able to infer that it was the user's first day of school from previous interactions with the program. The large language model also indicated that these capabilities are part of a new upgrade. The official conversation logs from ChatGPT confirmed that the content shared by the user was accurate.On September 13, 2024, OpenAI released a preview upgrade of the highly anticipated smarter model to select users. OpenAI stated that these new models will possess reasoning abilities and grant ChatGPT a seemingly magical "think before answering" capability. The response time of the new model will be between 10-20 seconds, and it will be able to search the internet and conduct in-depth research independently.

ChainCatcher news, according to StockMKTNewz, Palantir and Microsoft are advancing their partnership, with Microsoft leveraging Palantir to enhance AI and LLM (large language model) capabilities.It is reported that Palantir is an American software company primarily engaged in the development and application of big data analytics and artificial intelligence technology. It was founded in 2003 and is headquartered in Denver, Colorado.

ChainCatcher news, according to official information, the AO ecosystem has launched the "AI on AO" plan, introducing on-chain open-source large language models (LLMs), aimed at bringing any AI model (not just LLMs) on-chain. Based on Apus Network, utilizing Arweave's permanent on-chain storage, it aims to build a decentralized, trustless GPU network dedicated to providing reliable, efficient, and low-cost computing power for AI training and inference. AI data on AO can be uploaded to Arweave's models via ArDrive.

ChainCatcher news, according to Coingape, Cardano founder Charles Hoskinson has invited recently ousted OpenAI (ChatGPT) CEO Sam Altman to collaborate on building a decentralized large language model (LLM) on Cardano.According to previous news from ChainCatcher, the OpenAI board announced that its CEO Sam Altman has left and exited the company's board, with the company's Chief Technology Officer Mira Murati appointed as interim CEO.

ChainCatcher message, crypto KOL knower stated on his social media that building application infrastructure based on large language models (LLM) is currently very complex and highly competitive. In the absence of clear solutions in the fields of artificial intelligence and crypto infrastructure, he is optimistic about the potential launch of tool tokens in the future.Additionally, he mentioned that a bull market may arrive in the next 6-12 months, and narratives around the practicality of dePIN will be more important than ever.

ChainCatcher news, Kaito AI has started to open beta testing of the industry's first AI cryptocurrency search engine WHAT to 35,000 users on the waiting list from today.It is reported that WHAT combines industry-leading artificial intelligence technologies, including Auto-GPT, utilizes the GPT-4 large language model, and Kaito's proprietary real-time cryptocurrency search engine, aiming to provide a new search experience for the entire cryptocurrency community. (Source link)

ChainCatcher news, Microsoft is preparing to launch artificial intelligence chips to power large language models. Two people with direct knowledge of the relevant projects revealed that Microsoft began developing this internally codenamed "Athena" chip in 2019. It is reported that Microsoft has provided the chip to a small number of Microsoft and OpenAI employees who are testing the technology. Microsoft hopes that this chip will perform better than the chips currently sourced from other suppliers. Reports indicate that tech giants like Amazon, Google, and Facebook are also developing their own internal chips. (source link)

ChainCatcher message, Dune recently released a large language model (LLM) roadmap in the community, having launched the first LLM feature and preparing a full suite of LLM functionalities and integrations for users. Dune has released the Query Explanations feature and integrated GPT-4, allowing users to click a button to display a natural language explanation for any query, simplifying SQL queries.At the same time, Dune will soon support natural language queries using LLM, enabling users to ask questions in English and query the Dune dataset without any SQL knowledge. (Source link)