ChatGPT's new prompt-engineering shortcut

PLUS: Google writes the news

TOGETHER WITH

Good morning, human brains. Welcome back to your daily munch of AI news.

Here’s what’s on the menu today:

  • ChatGPT for Android 📲 🧠 

    OpenAI’s app for Android is available for “pre-register.”

  • New ChatGPT’s feature 🦾

    ChatGPT launched “Custom Instructions.” What it is and how it works.

  • Google’s News AI ✨ 

    It’s code-named Genesis, and Google’s been pitching it to a news outlet you probably consume.

APPETIZER

ChatGPT for Android 📲

ChatGPT’s Android App is finally available — to “pre-order,” that is.

Goodness gracious, OpenAI.

Here’s what comes in the package:

1/ Mirrored functionality. It's like the web-based version but on your Android phone.

2/ Sync-capability. Your preferences and history synced between your mobile and desktop experiences.

Rolling out. Eventually.

3/ The US gets dibs. The rest of the world can apparently wait.

Our take: I mean, seriously, what’s taking them so long? The iOS version has Siri and shortcuts functionality, so hopefully Android users will see similar. Time will tell. Developing what is essentially two different apps isn’t easy, but it feels like there’s some foot-dragging on the Android side of things…

BUZZWORD OF THE DAY

Prompt engineering

Prompt engineering refers to the art of finessing LLMs into generating a desirable output.

By making specific requests, adding detail, and specifying how you’d like the LLM to answer, you can engineer the chat AI to produce a much more complete or interesting answer.

FROM OUR PARTNERS

When’s the last time your newsletter wrote itself?

Probably never, until today.

It’s the newsletter writing machine that lets you go from topic, to outline, to full draft — in mere minutes.

With an AI trained specifically for long-form copy, it’s able to write your newsletter in the exact tone you choose. And Hoppy even incorporates images, that dovetail, right into that copy.

It does all the lifting you want, without tying your hands. Tweak the outline, edit the copy on a sentence level, change the images — your piece is completely steerable, every step of the way.

Join 20,000 creators who don’t know what to do with all their new free time.

MAIN COURSE

ChatGPT’s Custom Instructions 🦾

ChatGPT’s launched a new feature called “Custom Instructions. It gives ChatGPT memory recall capabilities, customization options, and more.

1/ Users set preferences that ChatGPT uses in all future outputs. This makes detailed prompting way, way more efficient — just include all the specifics in Custom Instructions instead of pasting them in from scratch every time.

2/ ChatGPT Plus subscribers have Beta access, but it’ll be available for all users soon.

3/ There’s a 1,500 character response limit.

4/ OpenAI claims the data will enhance the model’s performance, but users can opt out of settings.

Our take: OpenAI’s releasing this feature after murmurings of a decline in GPT-4’s performance. Could be planned, could be coincidence, but the timing’s certainly in their favor. We anticipate a growing trend of customizability and settings that let you tailor your AI’s output — just like modules or packages in programming. Stock intelligence, customized personality modules.

A LITTLE SOMETHING EXTRA

Google’s journalism AI ✨ 

So it begins.

Well, that’s a little sinister…

Google pitched Genesis to The New York Times, The Washington Post, and other notable publications for testing.

Some executives said it was “unsettling,” and voiced concerns that the AI wouldn’t be able to produce a faithful story.

Google has stated Genesis is a “responsible technology“ that will help journalists, not replace them.

Ah ok, now I’m convinced.

Our take: Earlier this year, CNET’s use of AI in articles led to factual errors and plagiarism. But, of course, the news industry is using AI and will continue to do so. It’s just a matter of time before news is openly produced by machines — and fact-checked by humans. A new profession?

MEMES FOR DESSERT

YOUR DAILY MUNCH

Think Pieces

LLama 2 is expensive. GPT 3.5? In comparison, it’s pretty cheap.

China officially mandates that AI must have “core values of Socialism.” What the new guidelines mean for the development of Chinese AI.

Will AI end the world? Dr. Strangelove revisits us — and here’s how to use AI creatively.

Startup News

Google releases info on its “Red Team” — a group of AI hackers working to make AI safer.

Gushwork.AI raises $2 million in funds. The goal: outsource AI-specific business tasks.

Assembly AI’s Conformer 2 is here. The automatic speech recognition model was trained on 1.1M hours of English audio.

Research

Where did LLMs come from? An in-depth look at the evolution of 15,821 Large Language Models.

TokenFlow. A method for enhancing video synthesis in image models.

SciBench. A method to get LLMs to develop “college-level“ problem-solving capabilities in Science.

Tools

Hoppy Copy [Sponsored] — write your newsletter effortlessly with the power of AI.

Superhuman AI — the highly anticipated AI-powered email productivity tool.

Articula — an accurate AI-powered translation IOS app for 24 different languages.

Magic Hour — an AI-powered video creation platform.

TWEET OF THE DAY

Elon Musk really likes the letter “X.” So much so, that he’s even renaming “Twitter” to “X“.

Tag us on Twitter @BotEatBrain for a chance to be featured here tomorrow.

RECOMMENDED READING

If you like Bot Eat Brain, we bet you’ll like this newsletter too:

Return on SecuritySave hours of research with a weekly review of the economics of the cybersecurity market in 5 minutes

AI ART-SHOW

Until next time 🤖😋🧠

What'd you think of today's newsletter?

Login or Subscribe to participate in polls.