ChatGPT 即将大更新!免费用户自动升级 GPT-4 、创建 GPTs!
最新消息,国外大佬AIPRM 的首席工程师 Tibor Blaho 在 X 曝光了 ChatGPT 即将迎来史诗级更新。
可惜的是并不是 ChatGPT5 ,本次更新更偏向应用层的交互以及增强。
压力再次给到了二级开发者😄~
废话不多说,快来看看都更新了什么吧!
更新内容:
原文地址:https://twitter.com/btibor91/status/1782181937861316994
人们都在努力寻找 GPT-5 的任何线索或提及,但遗憾的是,我还没有看到任何线索或提及。
从现实情况来看,根据我目前所看到的,OpenAI 接下来/即将推出的 ChatGPT,已经在工作中,已经部署,并且可以公开访问--以下是一个总结:
#1 新的 ChatGPT 界面
- 新的 ChatGPT 用户界面原型(隐藏和更改用户界面的某些部分)以及对话页面和消息的显示方式
- 不再有侧边栏--取而代之的是显示对话历史的下拉菜单(至少部分是这样)
- 账户和设置菜单移至右上方(头像图标)
- 模型选择器移至信息输入处
#2 语音模式
使用顶部导航中的语音按钮启用语音模式布局
交互式开始和停止对话(在聊天对话中转录)
#3 自定义 GPT(小工具)
允许无报酬用户创建和/或与自定义 GPT 互动
带有指示块的 Gizmo 意图(根据意图加载更多指示)和保存状态(在所有对话中按用户存储自定义 GPT 数据)
Gizmo 联系人和联系人工具--允许自定义 GPT “联系 ”和使用其他自定义 GPT(类似于新的人工智能代理生成器)
内存作为自定义 GPT 的工具
设置/个性化 - 现在有一个将 GPT 添加到 Gizmo 名册的选项
#4 模型
带浏览和视觉功能的 GPT-3.5 (Sahara-V)
GPT-4 Lite(Scallion;POR)
动态模型切换器在 GPT-3.5 和 GPT-4 之间自动路由请求
为每条信息提供模型调谐器和动态再生模型选择器
为免费用户提供从 GPT-3.5 到 GPT-4 的自动升级响应(gpt-4-upsell)
所有型号和用户的内存(免费和付费)
#5 书写(Jidori)
Jidori 提示控件侧重于在指导下进行写作和改写
基于 ProseMirror(开源项目,所见即所得的编辑器)的新 Jidori 编辑器/输入法
以写作为重点的特殊自定义 GPT(“专家作家编辑写作草稿”)(与 DALL-E 控件类似,仅适用于 DALL-E 自定义 GPT)
#6 上下文连接器
谷歌驱动器、谷歌文档、微软 365 和 Notion
上传文档、文件或文件夹以进行分析、搜索或用作知识(上下文答案)
引用卡和文档分析器
使用连接应用程序的上下文连接器选择器
同步/更新引用文档
#7 浏览
自定义 GPT 的引用/网络属性
用于浏览的简单工具指示器
并行浏览(gpt-4-pbrowse)和研究
#8 数据分析 V2(石墨)
新版 gpt-4-ada-v2
带编辑器的数据网格
图表工作室和交互式可视化
#9 聊天声波 - SNC(SearchGPT)
搜索引擎 - Bing (POR)、Sydney 和 Fortis
GPT-4 Lite 型
内部搜索(拉布拉多)
图像搜索
保守或激进(POR)机器人模式
小工具 - 天气、计算器、体育(赛程、排名、NBA)、财务、时差
后续提示
#10 其他
分享聊天的简化模式,可选择使用网络搜索+通过社交网络分享
域名迁移到 chatgpt[.]com
使用自定义指令定制 GPT-4 功能(启用或禁用浏览、DALL-E 和代码解释器)
这里为一些不方便的同学准备了英文原版:
People are trying to find any clues or mentions of GPT-5, but I haven't seen a single one yet, unfortunately.
Realistically, what OpenAI could launch next/soon for ChatGPT, based on what I've seen so far and is already in the works, deployed and publicly accessible - here's a summary:
Fruit Juice and Fruit Juice C
New ChatGPT UI prototype (hiding and changing certain parts of the UI) and how conversation page and messages are presented
No more sidebar - replaced with dropdown featuring conversation history (at least partially)
Account and settings menu moved to the top right (avatar icon)
Model selector moved to the message input
Audio Mode
Voice mode layout enabled using voice button in top navigation
Start and stop conversation interactively (transcribed in chat conversation)
Custom GPTs (Gizmos)
Allow unpaid users to create and/or interact with custom GPTs
Gizmo intents with instructions block (load more instructions based on intent) and saved state (per-user storage for custom GPT data in all conversations)
Gizmo contacts & contacts tool - allows custom GPT to "contact" and use other custom GPTs (similar to the new Vertex AI Agent Builder)
Memory as a tool for custom GPTs
Settings / Personalization - now has an option to add GPTs to your Gizmo Rolodex
Models
GPT-3.5 with browsing and vision (Sahara-V)
GPT-4 Lite (Scallion; POR)
Dynamic model switcher auto-routing requests between GPT-3.5 and GPT-4
Model tuner & dynamic regenerate model selector for every message
Auto-upgraded responses from GPT-3.5 to GPT-4 for free users (gpt-4-upsell)
Memory for all models and users (free and paid)
Writing (Jidori)
Jidori prompt controls focused on writing and rewriting with guidance
New jidori editor/input based on ProseMirror (open-source project, WYSIWYM editor)
Special custom GPT ("expert writer editing a draft of a piece of writing") focused on writing (similar to DALL-E controls only for DALL-E custom GPT)
Context Connectors
Google Drive, Google Docs, Microsoft 365, and Notion
Upload documents, files, or folders to analyze, search, or use as knowledge (contextual answers)
Citations cards and document parser
Context connector picker for using connected apps
Sync/update referenced documents
Browsing
Citations/web attributions also for custom GPTs
Simple tool indicator for browsing
Parallel browsing (gpt-4-pbrowse) and research
Data Analysis V2 (graphite)
New gpt-4-ada-v2
Data grid with editor
Chart studio and interactive visualizations
Chat Sonic - SNC (SearchGPT)
Search engines - Bing (POR), Sydney, and Fortis
Model GPT-4 Lite
Internal search (Labrador)
Image search
Conservative or aggressive (POR) robots mode
Widgets - weather, calculator, sports (schedule, standings, NBA), finance, timezone difference
Follow-up prompts
Other
Simplified modal for sharing of chats, optionally discoverable using web search + share via social
Domain migration to chatgpt[.]com
Customize GPT-4 capabilities using custom instructions (enable or disable Browsing, DALL-E and Code Interpreter)
It's interesting that so many work-in-progress changes are pushed to production and published for testing/feedback? Some of these might be already abandoned experiments, but who knows...
As always, I'll try to make sure AIPRM for ChatGPT is ready for all of these changes