[上半期节目](https://www.xiaoyuzhoufm.com/episode/67ad6e322b6a6452657f1f63)在让子弹飞
等我这期剪辑完成,已然成为了AI历史回顾类节目
25年年初AI产业界风起云涌,后起之秀喧嚣尘上,agent应用层的结构也逐渐清晰。再来听24年底OpenAI的连续12天发布会已如昨日黄花
DAY 1: release o1 & chatGPT Pro
DAY 2: Reinforcement Fine-Tuning
DAY3: Sora
DAY4: Canvas ChatGPT版的
DAY5: ChatGPT in apple intelligence
DAY6: Advanced voice with video 更自然流畅的语音模型
DAY7: Projects in ChatGPT
DAY8: ChatGPT search
DAY9: for Developer 开放 o1 模型,时效性优化,Go 和 java 的 sdk
DAY10: 1-800-CHATGPT 怎么说呢 电信诈骗完美方案演练?
DAY11: ChatGPT to integrate and function seamlessly across multiple applications
DAY12: o3 and o3-mini
模型:o1, o3, advanced voice with video, RLFT, Sora
ChatGPT: Pro, Canvas, iphone, projects, search, phone call, work with apps
Developer: o1, RT-API, FT, GO & Java SDK
23 May 2024
Vol.18 所谓人生的价值 聊聊Lifetime Value feats.詹同学&妙妙
Lifetime Value - By Suno.ai
Counting dollars
Adding cents
Talking numbers and percentages
Lifetime value
The golden thread
Calculating all we've ever said
加入Booking以后一直在做的领域LTV,用户生命周期可以带来的商业价值,在data driven的财务决策,PPC,甚至AB testing都有应用。
没想到这么一个主题居然聊了一个多小时哈哈哈哈�
Topics
* 简单介绍 LTV
* LTV 的定义
* LTV 在股票估值市场的应用:资本收益率 vs CAC(customer acquisition cost)/ LTV
* 如何评价一套 LTV 体系
* LTV 项目背景
* 启动 LTV 项目的条件
* 数据条件
* 收入是否可以划分到用户
* user 的 LT 和 Value 在不同场景存在不同定义
* LTV预估模型:LTV 预估方法演进
* 算法数据,有效特征,模型评价等
* LTV 365 时间比较长,真实值要一年后才产生,如有市场变化怎么办,比如疫情
* LTV 的实际业务应用和效果,基于 LTV 指标体系有什么用户洞察
* 未来可能的方向
* 其他
* 神经网络还在用 tf estimator 那套东西吗?有什么变化?
* 如何看待适用机器学习模型与神经网络模型的应用场景
* 分别适用什么样的场合?数据量是个很大的瓶颈吗?
* 针对一个陌生任务(没有标准 DNN 网络结构的情况下),如何从零开始搭建网络?
11 Nov 2023
Vol.17 AI春晚!浪潮之巅!
和詹同学一起回顾了OpenAI首次Dev Day,简直是一次里程碑式的发布了!
OpenAI完整的呈现了它的野心和路径:chatbot -> AI Agent -> AGI,并且坚定的产品化的不断实践
在竞争对手还在做chatbot的show cases时,OpenAI已经完整展开了两个生态:一个是ChatGPT+GPTs,以chat为核心交互的全新体验;一个是以Assistants API为核心的能融入现有任何产品形态做AI+的模式
目前GPTs及GPT Builder都已对订阅用户开放,来体验下我做的小东西吧:https://chat.openai.com/g/g-O8VXL4opJ-emoji-sparkles
OpenAI开发者大会 DevDay 2023 直播视频(中英文精校)】 https://www.bilibili.com/video/BV15e411X75q/?share_source=copy_web&vd_source=1dc1940899d23c495584b2398c77593a
00:49 OpenAI 2023回顾
2022.11 GPT-3
2023.03 ChatGPT; Whisper; plugins
2023.04 GPT-4
2023.05 ChatGPT app for iOS
2023.06 Function calling
2023.07 Custom instructions for ChatGPT
2023.08 GPT-3.5 Turbo fine-tuning
2023.10 多模态:语音,图片,DALL-E3
ChatGPT Enterprise
2M Developers, 92% HVC,1亿WAU
How people use the product, how people use AI
26:27 GPT-4 Turbo
context length to 128k: 16X comparing to GPT-4. 300 pages
more control: JSON Mode; call multi-functions at once; reproducible output using seed
better world knowledge: up to 2023.04
new modalities API: DALL-E3, GPT-4 Turbo with vision, TTS, Whisper V3
Customization, Custom Models: fine-tuning for long context GPT-3.5 and GPT-4
Higher rate limits: 2x tokens per minute
Copyright Shield
Price: 3x less for Input tokens(1 cent for 1000 input tokens) and 2x less for output tokens(3 cent for 1000 output tokens); lower cost for GPT-3.5 and fine-tuning as well
Partnership with Microsoft
Azure: Infrastructure building
Mission
Safety
50:19 ChatGPT
Now using GPT-4 Turbo
All in one 可以使用现有tools ChatGPT会自己选择使用什么tools → AI Agent
Tell AI what you need and it’ll do all of these tasks for you.
GPTs: 针对特定用途定制的ChatGPT
三个最重要的自定义组件: instructions; expanded knowledge; actions
在执行操作和数据交互时的用户确认
三个主要场景: private; public; business
GPT Builder: 使用NLP构建GPTs
GPT Store: for public GPTs
Revenue sharing 头部GPTs
62:13 Assistants API (API version AI Agent)
Threading: 对话历史记录
Retrieval: expended knowledge
Actions
Code interpreter: If there’s no proper tools
Function calling
Playground
Name
Instruction
Model
Tools
Threads and Messages: keep adding messages into a thread
67:27 谈谈感受与未来
两个生态:ChatGPT+GPTs & Assistants API
benchmark
AI颠覆娱乐业
Contact DeepFM 深度调频
Guest appearances
Books guests
Based on episode analysis; this does not confirm that the show is currently accepting guests.
Host of DeepFM 深度调频?
Claim your podcast to manage its listing and keep your show details accurate.
Pod Engine is an independent podcast discovery and analytics service and is not affiliated with or endorsed by this podcast. Artwork and show content belong to their owners. Full legal notice.
Explore this show Podcast research with Pod Engine