Digital Platform Policy Highlights - Digest 20
Q3-Q4 2023: This post delineates the impact of digital policies on the enhancement of AI systems that have generative capabilities.
TL;DR→ Policy changes to utilize increased computational capabilities as well as advances in AI include
ChatGPT introducing custom instructions to personalize the chatbot
OpenAI rolling out a powerful chatbot API for all paying customers
With Ernie Bot, Baidu enters the generative AI race
Generative AI not just helps you to write better email but also can help with complex tasks such as simulate prediction for finance and scientific research. Baidu, a Chinese tech giant, has officially launched its AI chatbot, Ernie Bot, after getting approval from the Chinese government. Ernie Bot works similarly to OpenAI’s ChatGPT and is available to Chinese users only. The chatbot is a showcase of Baidu’s generative AI capabilities and its adaptation to China’s strict regulatory environment. Given the tendency of LLMs to hallucinate, it will be interesting to see how well Ernie Bot complies with China’s censorship restrictions. (link)
Google adds generative AI to Workspace with Duet AI
Big Tech companies adding productivity improving features using GenAI was always on cards. Google is rolling out Duet AI, a collection of features that use generative AI to help users with various tasks in Gmail, Drive, Slides, Docs, and more. Duet AI can create charts, write emails, generate images, summarize documents, and more. Duet AI is only available to corporate accounts and will cost $30 per user every month. The price exactly mirrors what Microsoft charges for Copilot, highlighting that some companies believe the productivity-gain per person will far exceed $30 per month :) (link)
ChatGPT introduces custom instructions to personalize your chatbot
Given the LLMs have been "thinking and talking" very similar to a human, why can't it emulate the "thinking and talking" style of the user interacting with it? ChatGPT, a chatbot powered by OpenAI, has launched a new feature called “custom instructions.” It allows users to preface their conversations with information about themselves and their preferences, which the chatbot will use to tailor its responses. The feature is only available to ChatGPT Plus subscribers and seems to be a simple way to make the subscription more attractive to heavy ChatGPT users. Should we expect a tiered subscription from OpenAI where the customization and similarity are tied to the money one pays? (link)
OpenAI’s most powerful chatbot API rolls out for all paying customers
OpenAI made its most powerful chatbot API, GPT-4, available to all paying developers to integrate to their solutions. The API is capable of generating human-like responses to a wide range of prompts but as with previous models, raises concerns about the potential misuse of AI-generated content and the need for ethical guidelines and regulations. GPT-4 faces increasingly stiff competition from Meta’s LLaMa and Mistral’s Mixtral: both have been made available for free for developers and researchers to use. Given how good LLaMa and Mixtral models are (I use both models extensively) it will be interesting to understand OpenAI’s strategy to compete with what is essentially free. (link)
TikTok makes it easier to label AI-generated content
TikTok introduced a new policy that requires creators to disclose if their posts contain AI-generated content, such as deepfakes or synthetic images. There are also policies to protect nonpublic citizens, such as disallowing deepfakes of nonpublic citizens. The feature is meant to allow for creativity with disclosure while allowing TikTok to address regulators’ calls to address fake news. This is in line with the trends where social media companies, including TikTok and Instagram, are rushing to regulate AI-generated content on their platforms. The new policy introduces an interesting challenge: defining “public” citizens using a simple definition. (link)
This is the eighth part of a multi-part post. You can find part one, part two, part three, part four, part five, part six and part seven here. If you know someone who enjoys these kinds of posts, please share it with them :)
Research help from Anantesh Mohapatra and John Mai (Thanks a ton, folks!)