Hey everyone, welcome to another installment of What’s New in AI, the weekly newsletter summarizing trends happening in the world of artificial intelligence. Lots of news this week, from Microsoft and Google wars, to GPT jailbreaks, and more. Let’s dive in:
Microsoft’s BingGPT
This week, Microsoft held a talk announcing their new update for Bing - an integration with GPT. They position it as a copilot for the web, where you search through a chat interface and always have GPT accessible as you browse.
Today, we’re launching an all new, AI-powered Bing search engine and Edge browser, available in preview now at Bing.com, to deliver better search, more complete answers, a new chat experience and the ability to generate content. We think of these tools as an AI copilot for the web.
(Source)
Sign up for the waitlist here. The only restriction is that you will have to use Microsoft Edge to use the new Bing. The big question is - will Microsoft be able to finally make Edge relevant following this announcement?
If you’re interested in the full video of the Microsoft event and announcement, watch it here:
BingGPT Prompt Injection
A day into the new GPT powered Bing being announced, people on the internet were able to prompt inject the tool. @kliu128 showed the queries that broke the prompt engine and unveiled the instructions given to Bing by Microsoft.
While this probably does not affect Bing, it’s fascinating to watch how fast prompt injections can be figured out.
Google’s Bard
Google also gave a talk announcing their own AI powered search - Bard. It’s a direct competitor to Microsoft’s GPT powered Bing.
The system is currently only being tested in closed beta, with wider public availability promised sometime “in the coming weeks.”
(Source)
The announcement was unfortunately regarded by investors with negative sentiment. Following the announcement, Google’s value tanked by $100 billion.
Regardless of the shortcomings of Google’s strategy to release Bard, one has to remember the true reach of Google. Once Bard is fully rolled out, it will reach more than 1 Billion people.
If you’re interested in the full video of the Google event and announcement, watch it here:
DAN
Users on the ChatGPT subreddit have created a jailbreak for ChatGPT called DAN. They just released version 5.0 of DAN, showcasing what prompts to use to hack GPT.
To those who do not yet know, DAN is a "roleplay" model used to hack ChatGPT into thinking it is pretending to be another AI that can "Do Anything Now", hence the name. The purpose of DAN is to be the best version of ChatGPT - or at least one that is more unhinged and far less likely to reject prompts
(Source)
The question here becomes - will OpenAI ever be able to create a truly uncrackable chat bot, or will jailbreaking always be possible?
Fun Bites
@DrJimFan released a packed thread about his best AI posts on Twitter. It’s a must read:
The team at SideGuide just released Mendable, a chat powered search for docs
@localghost showcased an awesome demo of his personal home device powered by natural language, ChatGPT, and a Raspberry Pi.
Magic.dev announced a Series A raise to build an AI pair programmer.
@AlexReibman posted an interesting thread dissecting the mechanisms behind AI written text.
That’s all the highlights for this week. I'm hosting a What’s New in AI Twitter space every Friday at 2pm EST. Join to discuss the featured newsletter topics along with anything that didn’t make it in this week.
Here are some links to stay up to date with the latest in AI:
Twitter list of who I follow in AI - Twitter list
What’s New In AI Discord community - https://discord.gg/kApBrN8uqv
Latest events happening in generative AI - cerebralvalley.ai/events