Odaily Planet Daily News Lightning Labs has launched new artificial intelligence Bitcoin tools designed to enable developers to seamlessly integrate Bitcoin and the Lightning Network into their artificial intelligence applications. The suite includes LLM Agent BitcoinTools, It allows developers to create AI agents that can hold Bitcoin balances, send/receive Bitcoin on the Lightning Network, and interact with Lightning Network daemon nodes. Additionally, this release includes the Aperture reverse proxy server, which supports Lightning Node Connect and provides dynamic API endpoint pricing. These tools will enhance the accessibility and functionality of AI infrastructure. (The Block)

Babbitt News, June 9th, according to the Chosun Ilbo report, it has been confirmed that Samsung is fully developing its own AI large language model (LLM) for internal use. This product is similar to ChatGPT. It has been developed since June and is led by Samsung Research Institute. . Sources said that almost all of Samsung's GPU computing resources have been invested in large-scale model training. The company plans to complete the development of the first version of LLM within two months. At present, Samsung plans to use this large model for document summary, software development and language translation, and has not yet decided whether to provide the product to consumers.

PANews news on June 2, according to PR Newswire, blockchain analysis company Elliptic announced the integration of large language model (LLM) ChatGPT into its off-chain intelligence and research collection work, helping researchers and investigators to more quickly , Synthesize and organize intelligence on new risk factors on a larger scale. Integrating ChatGPT will allow Elliptic to understand exactly a client's risk exposure to make more informed and secure decisions, the company said.

According to 36 Krypton’s report on June 5, the AI data company “Integer Intelligence” recently completed tens of millions of Pre A rounds of financing. Yishang Capital is the exclusive financing advisor. This round of financing is mainly used for the iterative upgrade of the intelligent data engineering platform (ABAVA Platform). The new ABAVA platform will integrate AI large models + small models to achieve efficient automatic data annotation. In addition, integer intelligence will also integrate RLHF (human feedback reinforcement learning) data service capabilities to provide solutions for the development and iteration of large language models (LLM).

According to 36 Krypton’s report on June 5, the AI data company “Integer Intelligence” recently completed tens of millions of Pre A rounds of financing. Yishang Capital is the exclusive financing advisor. This round of financing is mainly used for the iterative upgrade of the intelligent data engineering platform (ABAVA Platform). The new ABAVA platform will integrate AI large models + small models to achieve efficient automatic data annotation. In addition, integer intelligence will also integrate RLHF (human feedback reinforcement learning) data service capabilities to provide solutions for the development and iteration of large language models (LLM).

Lenovo Group launched its first next-generation Copilot+ PC, the Lenovo Yoga Slim 7x and Lenovo ThinkPad T14s Gen 6, which are equipped with the latest Qualcomm Snapdragon XElite processor. According to reports, these two AIPC devices, powered by the new Snapdragon XElite processor, feature a 12-core Oryon CPU, Adreno GPU, and dedicated Hexagon NPU. With the enhanced features of Microsoft and Copilot+, users can now use large language models (LLM) even when offline.

According to Interface News on June 27, the database company MongoDB announced that it will cooperate with Google Cloud to accelerate developers' use of generative AI and the development of new applications. Developers can take advantage of MongoDB Atlas and integration with Google Cloud Vertex AI Large Language Model (LLM) to accelerate software development.

IT House issued an article today stating that the blind heap volume of the AI model is actually not necessarily better. It depends more on the quality of the training data. Microsoft recently released a language model phi-1 with 1.3 billion parameters, using "textbook Level" high-quality data set training, it is said that "the actual effect is better than GPT 3.5 with 100 billion parameters". The model is based on the Transformer architecture, and the Microsoft team used "textbook-grade" data from the web and "logically rigorous content" processed with GPT-3.5, as well as eight Nvidia A100 GPUs, to complete the training in just 4 days . The Microsoft team said that rather than increasing the number of parameters of the model, improving the quality of the model's training data set may enhance the accuracy and efficiency of the model. Therefore, they used high-quality data to train the phi-1 model. In the test, the score of phi-1 reached 50.6%, which is better than GPT-3.5 (47%) with 175 billion parameters. Microsoft also stated that phi-1 will be open sourced in HuggingFace next, and this is not the first time Microsoft has developed a small LLM. Before that, they created a 13 billion parameter Orca, which was trained using GPT-4 synthetic data. The performance is also better than ChatGPT.

According to the Securities Times, on June 20, Jinshan Office held a 2022 annual performance briefing. Regarding the development of AI-related products, Zou Tao, chairman of the company, said that AI is one of the important product strategies of Kingsoft Office. At present, the company chooses to "walk on two legs", and will first choose to cooperate with technical service providers related to AIGC and LLM, and carry out product research and development through their technical empowerment; at the same time, they are also continuing to contact some excellent start-up companies. In addition, "WPSAI" related functions will be gradually launched, which is expected to fully empower the company's entire line of products.

According to Businesskorea, according to South Korean sources, the Device Solutions (DS) division in charge of Samsung Electronics’ semiconductor business is developing its own large language model (LLM) of GPT-3.5 or higher. Customized artificial intelligence to improve productivity and work efficiency. It is reported that the service is expected to start its basic service as early as December this year, and provide professional search services including company knowledge and data in February next year.

On June 12, the Golden Ten Data reported that, according to a survey by the global market research firm TrendForce, the first quarter saw a traditional off-season for consumer terminals. Although there were occasional urgent orders in the supply chain, most of them were individual customer inventory replenishment behaviors, and the momentum was slightly weak. At the same time, the demand for automotive and industrial control applications was affected by factors such as inflation, geopolitical conflicts, and energy. Only AI servers, under the trend of global CSP giants investing heavily and enterprises deploying large language models (LLMs), emerged as the only bright spot supporting the supply chain in the first quarter. Based on these factors, the quarterly output value of the top ten global wafer foundries in the first quarter decreased by 4.3% to $29.2 billion.

According to a report by the Financial Associated Press on June 14, Bank of America Securities issued a report stating that AMD recently announced its latest AI GPU chip "MI300 X", and predicts that the AI chip market will increase from US$30 billion this year to US$150 billion in 2027. . However, the bank remains "neutral" on AMD as the company did not mention any commercial customers participating in its new GPU products. AMD is still far behind Nvidia, as AMD will not start sampling AI chips until next quarter. In terms of timing, when AMD starts to stabilize with 5nm-based GPU accelerators, Nvidia may launch its next-generation 3nm-based, LLM-optimized Blackwell architecture GPU, further extending its already sizable lead in the AI market.

According to PANews news on July 7, according to The Block, Lightning Labs has launched a new artificial intelligence bitcoin tool. The tools are designed to help developers seamlessly integrate Bitcoin and the Lightning Network into their AI applications. The suite includes LLM Agent BitcoinTools, which allows developers to create AI agents that can hold Bitcoin balances, send/receive Bitcoin on the Lightning Network, and interact with Lightning Network Daemon nodes. Additionally, this release includes Aperture Reverse Proxy Server, which supports Lightning Node Connect and offers dynamic API endpoint pricing. These tools will enhance the accessibility and functionality of AI infrastructure.

Odaily Planet Daily News Bitcoin mining company Hive Blockchain said in an earnings call with analysts on Friday that the company will allow customers to train large-scale language artificial intelligence models in its data center, compared with competitors such as OpenAI's ChatGPT, The company claims better privacy. Aydin Kilic, CEO and President of the company said: “Companies are now noticing that they don’t want to upload sensitive customer data to companies like OpenAI that have a public LLM (Large Language Model). We want to provide privacy in Hive through Hive Cloud, Companies can have a service agreement, own their data and their privacy, and still run AI compute workloads on our GPUs." Shares of Hive rose nearly 2 percent on Nasdaq on Friday. (CoinDesk)

According to a report by IT House on June 25, developer Iván Martínez Toro recently launched the PrivateGPT open source model, which allows users to ask questions based on their own documents without an Internet connection. It is reported that the PrivateGPT open source model can be run locally on home devices, and the open source large language model (LLM) called gpt4 all needs to be downloaded before running. Then, instruct users to put all their related files into a directory so that the model pulls in all the data. After training the LLM, the user can ask the model any question and it will answer using the provided document as context. PrivateGPT can ingest more than 58,000 words and currently requires a lot of local computing resources (high-end CPU is recommended) to set up. Toro said that PrivateGPT is currently in the proof-of-concept (PoC) stage, which at least proves that a large model similar to ChatGPT can be created completely locally, and the potential of this PoC conversion into actual products can be foreseen, allowing companies to access personalization, security and privacy ChatGPT to increase productivity.