生活

2025年08月随想

作者:liurw 日期:2025-08-08

字体大小: 小中大

2025-08-12

这两天关注的长江电力今天上涨，感觉这个股票前面由于260亿元投资，投资者担心的分红现金流问题应该消化了吧。

中午情况及K线如下

2025-08-10

中国科学家破解60年谜团，合成稀有钻石 --->笔记中国科学家合成稀有钻石

一直觉得钻石行业就是一个被垄断的、价格虚高的市场。发表在《自然》上的这篇文章有意思。

2025-08-09

GPT5界面

看看介绍

GPT-5 brings a few upgrades over earlier versions, mainly in three big areas:

Better reasoning and planning
- It’s more accurate when following multi-step logic.
- Handles trick questions, subtle wording, and hidden assumptions more reliably.
- Can plan and keep track of longer, more complex tasks without losing context.
Expanded multimodal abilities
- Can work with images more flexibly — analyzing, comparing, and integrating visual and text information in the same conversation.
- More robust at turning visual data into structured insights (e.g., charts, diagrams, or itemized reports).
Longer memory and context management
- Can hold much longer conversations and recall earlier details more precisely.
- Keeps relevant information in mind for complex projects, making it feel more “continuous” in multi-day sessions.

Plus, it’s generally faster, more concise when you want it, and more nuanced in tone — better at adapting to your preferred style, whether that’s professional, casual, or playful.

Here’s a side-by-side comparison of GPT-4 and GPT-5 so you can see the main upgrades clearly:

Feature / Capability	GPT-4	GPT-5
Reasoning Accuracy	Good at multi-step reasoning but prone to small logic slips	Stronger logical consistency, better at detecting hidden assumptions and trick questions
Context Length	Tens of thousands of tokens	Larger context window → can handle significantly longer conversations, documents, or code without losing track
Multi-Modal Abilities	Text + some image understanding (in certain models)	Text + image fully integrated; better at combining visuals and text in the same reasoning chain
Memory / Continuity	Can recall details within a single session	Improved memory handling → maintains more detail over longer sessions for complex projects
Planning & Sequencing	Handles short to medium plans well	Handles multi-day or complex project plans with fewer errors and better adjustment mid-process
Tone Adaptation	Can adjust style with prompts	More nuanced tone shifts — matches your preferred style naturally without overdoing it
Speed & Responsiveness	Good, but slows slightly with longer prompts	Generally faster, even with long or complex inputs
Error Detection	Often misses subtle mistakes in text, code, or data	More vigilant in catching subtle factual, logical, or calculation errors
Creativity	Strong for stories, brainstorming, and ideas	More consistent creativity with deeper contextual integration across long narratives
Precision in Calculations	Sometimes “shortcuts” in math, leading to mistakes	Enforces step-by-step calculations for higher numeric accuracy