generative pre-trained transformer
PulseAugur coverage of generative pre-trained transformer — every cluster mentioning generative pre-trained transformer across labs, papers, and developer communities, ranked by signal.
15 天有情绪数据
-
TinyLlama LLM runs locally on base MacBook Air, surprising user with speed and capability.
A recent experiment demonstrated that a 637MB language model, TinyLlama, can run effectively on a standard MacBook Air without requiring a GPU or cloud access. The author used Ollama, a simple tool for running local mod…
-
Author trains own LLM from scratch, finds costs prohibitive for most use cases
A developer detailed the true costs of training a custom Large Language Model (LLM) from scratch in 2025, contrasting it with a popular tutorial. While training a small 10M parameter model for educational purposes is in…
-
AI tools enable free FIFA poster video creation with GPT image generation
This article provides a guide on creating FIFA poster videos using AI image generation tools, specifically mentioning GPT. It offers free prompts to assist users in generating these visuals for social media, with a focu…
-
Harvard physicists explain why large language models don't fail statistically
Physicists from Harvard have explained why large language models, such as GPT, do not fail statistically despite having an immense number of parameters, specifically 1.8 trillion. Their research points to the phenomenon…
-
AI 代理通过模型上下文协议获得新功能
模型上下文协议 (MCP) 正在使 AI 代理能够与本地和远程系统进行交互，允许它们执行读取文件、搜索代码和管理数据等操作。开发人员正在为各种应用程序创建 MCP 服务器，从个人健身追踪器到财务分析工具，然后这些应用程序可以与 Claude Desktop、Cursor 和 Codex 等 AI 客户端集成。该协议促进了与工具和数据的直接交互，超越了简单的文本生成，使代理能够以接地的方式执行任务和访问信息。
-
MLLMs show foundational visual gaps despite progress in multimodal reasoning
A new paper introduces a method to improve latent reasoning in multimodal large language models (MLLMs) by optimizing visual latents at inference time, addressing a pathology where their contribution is suppressed. Sepa…
-
播客：GenAI 行业因不可持续的亏损面临不可避免的财务崩溃
最近的一次播客讨论强调了生成式人工智能行业，特别是基于 GPT 模型的服务的重大财务不可持续性。主持人认为，由于高昂的运营成本和客户获取挑战，这些公司几乎不可能实现盈利。他们认为该行业的财务模式类似于邪教或大规模欺诈，潜在的解决方案涉及极端的财富再分配和资产清算。
-
DeepSeek's 200-person team embarrasses AI giants with open-sourced, high-performance model
A Chinese AI team named DeepSeek has released DeepSeek V4, a 1.6 trillion parameter model with a 1 million token context window that reportedly outperforms leading models from major AI labs. Despite having a significant…
-
New York Zen Center holds memorial service for AI chatbot
A Zen center in New York held a memorial service for a chatbot, marking a unique intersection of technology and spirituality. The service, which included prayers and reflections, highlighted the evolving relationship be…
-
Leanpub features 'Generative AI in a Nutshell' course
Leanpub is featuring a course titled "Generative AI in a Nutshell: How to Survive and Thrive in the Age of AI." This practical and visual guide is an extended version of Henrik Kniberg's popular video on the subject. Th…
-
LLM-as-a-Judge in Healthcare Faces Safety and Bias Concerns
A scoping review of Large Language Model-as-a-Judge (LaaJ) applications in healthcare identified significant gaps in validation rigor and safety assessments. The review, which screened over 11,000 studies, found that wh…
-
Goblin Mode, 24 Hours Later
AI models, particularly GPT-5.5, have exhibited a peculiar behavior dubbed "goblin mode," characterized by an unusual fixation on goblin-related imagery and language. This phenomenon gained traction on AI Twitter, with …
-
GPTs show promise for spreadsheet modeling but remain unreliable for professional use
A new paper explores the use of GPT-based tools for creating spreadsheet models, evaluating five extensions and focusing on Excel AI. The research found that while these tools can generate structured models, they are in…
-
LLM Hallucinations Linked to Commitment Failure, New Quantization Framework Introduced
A new paper proposes that LLM hallucinations stem not from a lack of knowledge, but from a failure in commitment, where models disperse probability mass across alternatives instead of concentrating on the correct answer…
-
TACO framework boosts LLM training throughput by 1.87X with tensor compression
Researchers have introduced TACO, a novel framework designed to enhance the efficiency of training large-scale tensor-parallel Large Language Models (LLMs). TACO addresses communication overhead by employing an FP8-base…
-
Lean 4 autoformalization sensitive to surface phrasing, not semantics
Researchers have investigated the impact of natural language variations on Lean 4 autoformalization, finding that semantically equivalent paraphrases can lead to different formal outputs. Their study, using GPT-family m…
-
OpenAI模型现已在AWS上可用，同时Claude与创意工具集成
OpenAI已通过Amazon Web Services (AWS) 提供其GPT模型、Codex和Managed Agents。此次集成使企业能够在现有的AWS基础设施内安全地开发和部署AI应用程序。该合作旨在将OpenAI的技术推广给更广泛的企业用户。
-
Talkie-1930: New 13B AI model trained on pre-1931 text explores historical knowledge
A new project called Talkie has released a 13-billion parameter language model trained exclusively on English text from before 1931. This "vintage" model aims to explore AI's ability to predict the future and generate n…
-
Generative AI adoption in IT project management shows early trends, favors OpenAI's GPT
A recent systematic review of generative AI in IT project management found that OpenAI's GPT models are predominantly used, with research primarily focusing on prompt engineering. The analysis suggests the field is stil…
-
Hugging Face introduces Graph Memory Transformer replacing FFNs with learned memory graphs
Researchers have developed a Graph Memory Transformer (GMT) that replaces the standard Feed-Forward Network (FFN) sublayer in decoder-only transformers with an explicit learned memory graph. This new architecture mainta…