Open Source LLM

阿里巴巴開源ZeroSearch,讓AI模擬搜尋以降低訓練成本

阿里巴巴上週開源了ZeroSearch ,它是一個強化學習框架,鼓勵大型語言模型在訓練期間透過模擬搜尋進行學習,取代與真實搜尋引擎的互動,有望大幅減少9成的訓練成本。

阿里巴巴通義實驗室(Tongyi Lab) 亦發表相關研究論文 ,表明有效


Open Source LLM

GSoC 2025 Projects: AI-Powered Log Analyzer For Fedora, Better AMD ROCm On Debian

This week Google announced all of the accepted projects for this year's Google Summer of Code (GSoC). There are 1,272 accepted students/projects this year for student developers working on various interesting open-source efforts over the summer.

In going through all of the accepted projects for GSoC 2025, below is some of the work I found most interesting that Google will be funding via Google Summer of Code.

Fedora has projects lined up to create a service to get a new project
Continue reading...

Open Source LLM

LegoGPT creates Lego designs using AI and text inputs — tool now available for free to the public

(Image credit: Shutterstock)

A research team from Carnegie Mellon University built an AI model called LegoGPT that outputs valid LEGO designs from text inputs. According to the team’s research paper that's posted on GitHub , they trained “an autoregressive large language model to predict the next brick to add via next-token prediction,” but the key takeway is that the AI LLM creates LEGO designs from scratch.

The AI was trained on a dataset with more than 47,000 LEGO structures that build

Continue reading...

Open Source LLM

DeepSeek founder Liang Wenfeng ‘takes no short cuts’, Li Auto CEO says

DeepSeek founder Liang Wenfeng, in the background, follows an open-source approach in AI development. Photo: Shutterstock

Li, who is also the founder and chairman of the Beijing -based electric vehicle manufacturer, acknowledged how a single meeting last September with Liang enlightened him about DeepSeek’s “best practices” and the importance of “fighting against some of the laziness of human nature, like taking short cuts”, according to a video posted on Wednesday by the firm on various social media

Open Source LLM

(PR) NVIDIA & ServiceNow CEOs Jointly Present "Super Genius" Open-source Apriel Nemotron 15B LLM

ServiceNow is accelerating enterprise AI with a new reasoning model built in partnership with NVIDIA—enabling AI agents that respond in real time, handle complex workflows and scale functions like IT, HR and customer service teams worldwide. Unveiled today at ServiceNow's Knowledge 2025—where NVIDIA CEO and founder Jensen Huang joined ServiceNow chairman and CEO Bill McDermott during his keynote address— Apriel Nemotron 15Bis compact, cost-efficient and tuned for action. It's designed to drive the next step forward in enterprise
Continue reading...

Open Source LLM

Alibaba’s Qwen3 topples DeepSeek’s R1 as world’s highest-ranked open-source AI model

The broader LiveBench rankings showed that Qwen3 still trailed the top closed-source AI models, including OpenAI’s o3, Google’s Gemini Pro 2.5 and Anthropic’s Claude 3.7. Photo: Shutterstock

Data from LiveBench, an independent platform that benchmarks large language models (LLMs) – the technology underpinning generative AI services like ChatGPT – showed that Qwen3 surpassed R1 in tests that gauge open-source AI models’ capabilities including coding, maths, data analysis and language instruction. Alibaba owns the South China Morning Post.
Hangzhou -based Alibaba’s cloud computing unit last week released the Qwen3 family , which consists of eight enhanced models that
Continue reading...

Open Source LLM

JetBrains開源Mellum語言模型,專精程式碼理解與補全任務

JetBrains開源其第一個針對程式碼完成任務設計的語言模型 Mellum ,並已發布於 Hugging Face平台 ,供研究與開發人員取用。JetBrains稱Mellum為專精模型(Focal Model),其並非試圖涵蓋所有自然語言任務的通用模型,而是聚焦程式碼理解與補全的單一應用場景,


Open Source LLM|Intel

(PR) Intel Achieves First, Only Full NPU Support in MLPerf Client v0.6 Benchmark

Intel today announced that it is the only company to achieve full neural processing unit (NPU) support in the newly released MLPerf Client v0.6 benchmark. The result marks the industry's first standardized evaluation of large language model (LLM) performance on client NPUs. Intel's measurements of MLPerf Client v0.6 show Intel Core Ultra Series 2 processors can produce output on both the graphics processing unit (GPU) and the NPU much faster than a typical human can read.

"We are proud to

Open Source LLM

NotebookLM升級Gemini 2.5 Flash,行動版App即將推出

Google 受歡迎的NotebookLM原本只能從網頁存取,本週預告即將在5月20日釋出Android及iOS版App,將升級使用最新的Gemini 2.5 Flash模型。

現在 Google Play Store 蘋果App Store 分別顯示NotebookLM應用程式。Play Store説明App即將推出,但是App Store説明預計5月20日釋出。用户


Open Source LLM

Google NotebookLM 將登陸手機平台 提供 iOS/Android 版,介面設計與功能詳解

Google 將旗下 AI 文件分析工具 NotebookLM 擴展至流動平台,預告 iOS 和 Android 原生應用程式即將推出。新版應用將支援語音總覽、離線播放及跨平台共享,預期於 5 月 20 日 I/O 2025 開幕當日正式登場。

NotebookLM 是一款由 Google 開發、結合人工智能技術