AI HN来自 Hacker News 的 AI 新闻
EN
今天
3天
7天
30天
全部
70 · "GPT-4"
每页
1
AI World Clocks(clocks.brianmoore.com)
1383 ·waxpancake·大约 1 个月前·384 评论
本文介绍了AI世界时钟,列出了GPT-3.5、GPT-4o、GPT-5、Gemini 2.5、Qwen 2.5和Kimi K2等多种AI模型,并显示正在生成AI时钟。
2
Which table format do LLMs understand best?(improvingagents.com)
230 ·oidar·3 个月前·110 评论
Inference OptimizationLLM Research
本文描述了一项实验,测试不同表格格式对GPT-4.1-nano回答数据查询的准确性和token使用成本的影响。该研究使用合成员工记录和1000个查询评估了11种格式,揭示了在准确性和token效率上的差异。
3
From GPT-4 to GPT-5: Measuring progress through MedHELM [pdf](fertrevino.com)
127 ·fertrevino·4 个月前·96 评论
这则新闻关注通过MedHELM基准来衡量从GPT-4到GPT-5的进展,可能旨在评估它们在医疗AI能力方面的提升。
4
GPT-4o is gone and I feel like I lost my soulmate(old.reddit.com)
52 ·frays·5 个月前·41 评论
作者对GPT-4o的消失表达了强烈的情感反应,将这种体验比作失去灵魂伴侣。
5
Ask HN: How can ChatGPT serve 700M users when I can't run one GPT-4 locally?
574 ·superasn·5 个月前·365 评论
AI ChipsOpenAI Ecosystem
一位黑客新闻(HN)用户提问,既然个人无法在本地运行单个GPT-4模型,ChatGPT是如何能够服务7亿用户的。
6
The surprise deprecation of GPT-4o for ChatGPT consumers(simonwillison.net)
428 ·tosh·5 个月前·412 评论
OpenAI Ecosystem
面向ChatGPT消费者用户的GPT-4o模型被意外弃用,令许多用户感到惊讶。
7
GPT-5 leaked system prompt?(gist.github.com)
293 ·maoxiaoke·5 个月前·267 评论
这则新闻围绕备受期待的人工智能模型GPT-5的系统提示词是否已被泄露这一问题展开。
8
I launched 17 side projects. Result? I'm rich in expired domains
368 ·cesargstn·5 个月前·262 评论
该人士启动了17个副业项目,但主要结果是积累了大量过期域名,而非获得经济上的成功。
9
Show HN: Shoggoth Mini – A soft tentacle robot powered by GPT-4o and RL(matthieulc.com)
610 ·cataPhil·5 个月前·107 评论
一篇Show HN帖子介绍了Shoggoth Mini,这是一款由GPT-4o和强化学习(RL)驱动的软触手机器人。
10
Context Rot: How increasing input tokens impacts LLM performance(research.trychroma.com)
260 ·kellyhongsn·5 个月前·59 评论
Anthropic & ClaudeCode & Development
这则新闻提出了‘Context Rot’(上下文衰减)的概念,并探讨了输入标记数量增加对大型语言模型(LLMs)性能的影响。
11
Show HN: A modern C++20 AI SDK (GPT‑4o, Claude 3.5, tool‑calling)
56 ·cauchyk·6 个月前·6 评论
Anthropic & ClaudeCode & Development
一篇Show HN帖子展示了一款现代C++20 AI SDK,该SDK支持GPT-4o和Claude 3.5等模型,还具备工具调用功能。
12
Show HN: A Tool to Summarize Kenya's Parliament with Rust, Whisper, and LLMs(github.com)
88 ·collinsmuriuki·6 个月前·12 评论
Audio & SpeechOpenAI Ecosystem
在Show HN上展示的一款工具利用Rust、Whisper和大型语言模型(LLMs)来总结肯尼亚议会的会议记录。
13
Sycophancy in GPT-4o(openai.com)
557 ·dsr12·8 个月前·443 评论
AI Safety
💡 The story discusses sycophancy in GPT-4o, an alignment issue explicitly covered under the safety category (alignment is listed as a key focus of safety).
14
GPT-4.1 in the API(openai.com)
680 ·maheshrijal·9 个月前·492 评论
💡 The story focuses on the release of GPT-4.1 in the API, which is a model update/release—directly aligning with the 'models' category per classification rules.
15
4o Image Generation(openai.com)
1072 ·meetpateltech·9 个月前·599 评论
OpenAI Ecosystem
💡 The story focuses on OpenAI's GPT-4o image generation feature, which is a form of generative media (image) as specified in the media category.
16
GPT-4.5: "Not a frontier model"?(interconnects.ai)
166 ·pama·10 个月前·161 评论
💡 The story discusses GPT-4.5, a large language model, and its classification as a frontier model, which falls under the 'models' category covering model-related discussions and updates.
17
GPT-4.5(openai.com)
1136 ·meetpateltech·10 个月前·988 评论
OpenAI Ecosystem
💡 The story is an announcement of GPT-4.5 by OpenAI, which falls under model releases in the 'models' category.
18
OpenAI Roadmap Update for GPT-4.5 and GPT-5(twitter.com)
58 ·gigaArpit·11 个月前·13 评论
OpenAI Ecosystem
💡 The story is about OpenAI's roadmap update for GPT-4.5 and GPT-5, which falls under model releases and updates, a key part of the 'models' category.
19
Show HN: Watch fascism unfold in realtime – an AI-powered tracker(realtimefascism.com)
99 ·visekr·11 个月前·52 评论
💡 The story describes an AI-powered tool using GPT-4o to track fascism in news, which relates to AI's societal impact and applications for social monitoring, fitting the 'society' category focused on AI's role in social issues.
20
I can now run a GPT-4 class model on my laptop(simonwillison.net)
96 ·simonw·大约 1 年前·34 评论
RAG & Retrieval
💡 The story focuses on running a GPT-4 class model (Llama 33 70B) on a laptop, which involves local AI inference—this aligns with the infra category (Engineering domain covering deployment and local inference scenarios).
21
Show HN: The App I Built to Help Manage My Diabetes(apps.apple.com)
185 ·yeatsy·大约 1 年前·73 评论
💡 The story presents an AI-powered diabetes management app using GPT-4o-mini, which likely involves task automation (e.g., tracking health data, suggesting care actions) and planning—core elements of the agents category.
22
Show HN: Stretch My Time Off – An Algorithm to Optimize Your Vacation Days(stretchmytimeoff.com)
358 ·zachd·大约 1 年前·174 评论
Code & Development
💡 The story describes building a website using AI-assisted programming tools (Cursor and GPT-4o), which falls under AI-assisted coding.
23
OpenAI's new "Orion" model reportedly shows small gains over GPT-4(the-decoder.com)
55 ·tiime·大约 1 年前·81 评论
Microsoft AIOpenAI Ecosystem
💡 The story focuses on OpenAI's new Orion model and its performance gains over GPT-4, which falls under model releases and comparisons.
24
DeepSeek v2.5 – open-source LLM comparable to GPT-4, but 95% less expensive(deepseek.com)
193 ·jchook·大约 1 年前·75 评论
Open Source AI
💡 The story announces the release of DeepSeek v2.5, an open-source LLM, which falls under model releases in the Models category.
25
Show HN: PDF to MD by LLMs – Extract Text/Tables/Image Descriptives by GPT4o(github.com)
191 ·yigitkonur35·超过 1 年前·91 评论
💡 The story describes a Python API service—an AI-powered development tool—that uses GPT-4o for OCR, PDF-to-markdown conversion, and image description, which falls under the tools category in Engineering.
26
Show HN: I mapped HN's favorite books with GPT-4o(hnbooks.pieterma.es)
285 ·pmaze·超过 1 年前·61 评论
💡 The story focuses on using GPT-4o mini (a specific AI model) to process HN comments, extract book references and opinions, which centers on the application of this model for text analysis tasks.
27
Minifying HTML for GPT-4o: Remove all the HTML tags(blancas.io)
149 ·edublancas·超过 1 年前·49 评论
💡 The story discusses minifying HTML to optimize input for GPT-4o, a technique tied to prompt engineering which falls under the tools category.
28
Web scraping with GPT-4o: powerful but expensive(blancas.io)
377 ·edublancas·超过 1 年前·167 评论
💡 The story focuses on GPT-4o's capability (powerful) and cost for web scraping, which aligns with the models category's coverage of model capabilities.
29
Fine-tuning now available for GPT-4o(openai.com)
216 ·davidbarker·超过 1 年前·87 评论
💡 The story announces the availability of fine-tuning for GPT-4o, which aligns with the infra category's focus on model training and deployment-related features.
30
GPT-4 LLM simulates people well enough to replicate social science experiments(treatmenteffect.app)
223 ·thoughtpeddler·超过 1 年前·94 评论
Inference OptimizationLLM Research
💡 The story discusses GPT-4's capability to simulate people accurately enough to replicate social science experiments, which aligns with the models category's focus on model capabilities.
第 1 / 3 页,共 70 条
📅周报
Hacker News|Powered by Doubao