Everyday AI
Posts
Is ChatGPT's new memory feature a game changer? 🧠

Is ChatGPT's new memory feature a game changer? 🧠

ChatGPT cross-chat memory, Sam Altman joins US AI safety board, Meta Llama 3 coding test vs other LLMs

Everyday AI
April 26, 2024

👉 Subscribe Here | 🗣 Hire Us To Speak | 🤝 Partner with Us | 🤖 Grow with GenAI

Sup y’all 👋

Do you have more than 2.5 friends who could benefit from reading Everyday AI?

If so and you referred them, you’d be in the lead for winning our giveaway.

To celebrate our 1 Year anniversary of Everyday AI, we’re running a giveaway that starts NOW and runs through 11:59 PM CST on Monday.

We’re giving away a pair of Meta’s Ray-Ban smart glasses and more.

Just copy your unique referral code and email it, text it, or post it on social media.

Refer at least 1 friend to enter the giveaway.

The person with the most referrals through the contest end time will get first pick of prizes, which includes Meta Ray-Ban Smart Glasses.

Appreciate y’all!

✌️
Jordan

Outsmart The Future

Today in Everyday AI
7 minute read

🎙 Daily Podcast Episode: ChatGPT has added cross-chat memory. What does this mean for our workflows and did we need it? Give it a listen.

🕵️‍♂️ Fresh Finds: Microsoft and Google beat Q1 expectations, Perplexity launches an iOS app and Meta Llama 3 hits 1.2M downloads. Read on for Fresh Finds.

🗞 Byte Sized Daily AI News: Tech giants form US AI safety board, US and China to hold talks on AI and Elon Musk’s xAI nears $6 billion. For that and more, read on for Byte Sized News.

🚀 AI In 5: Can Meta’s Llama-3 hold its own against other LLMs in coding? We put it to the test vs ChatGPT, Gemini, Claude and Copilot! You might be surprised at the results. See it here

🧠 Learn & Leveraging AI: We’re breaking down ChatGPT’s cross-chat memory features and letting you know how and when to use it. Keep reading for that!

↩️ Don’t miss out: Did you miss our last newsletter? In our 1-year anniversary show, we talked about AI and job displacement, and also covered OpenAI gets a surprise delivery and Synthesia adds emotion-expressing avatars. Check it here!

ChatGPT’s New Memory Feature: the good and the bad 🧠

ChatGPT has just released a new memory feature!

BUT do we want ChatGPT to remember everything about us?

Is this a new ChatGPT solution looking for a problem?

We're covering ChatGPT's new cross-chat memory feature and how it might change the way you prompt. (For better or worse.)

Join the conversation and ask Jordan questions on ChatGPT here.

Also on the pod today:

• Application and actual use of memory feature 🧠
• Benefits and drawbacks of the memory feature ⚖️
• Suggestions for improving ChatGPT memory feature 🤔

It’ll be worth your 37 minutes:

Listen on our site:

Click to listen

Subscribe and listen on your favorite podcast platform

Listen on:

Spotify | Apple Podcasts |
Google Podcasts | Amazon Music |

Here’s our favorite AI finds from across the web:

New AI Tool Spotlight – CopyCopter creates viral videos from text, Resume Photo AI generates professional headshots and Gems is a personal AI knowledge assistant.

Money in AI – Microsoft and Google have beat their Q1 expectations thanks to AI.

LLMs - Perplexity has launched an iOS app for Pro users.

Ask any question — hear the answer.
Available on iOS for Pro users starting today.
— Perplexity (@perplexity_ai)
6:11 PM • Apr 25, 2024

Big Tech - Meta has released an update on Llama 3’s impact so far, and it’s already been downloaded 1.2 million times on Hugging Face.

Future of Work - Google has introduced a new fund of $75 million to provide AI training for workers, along with a new online AI course.

Business of AI – Microsoft is partnering with beauty giant The Estée Lauder Companies to supercharge beauty using GenAI.

1. Sam Altman and Other Tech Giants Join US AI Safety Board 🇺🇸

The Biden administration forms a high-profile advisory board, including top executives from Nvidia, Microsoft, and Alphabet, to provide advice to the Department of Homeland Security regarding the safe deployment of artificial intelligence in America's critical infrastructure.

Government officials emphasize the need for AI-powered defenses against potential disruptions and cyber threats. With the rapid growth of AI technology, the board aims to ensure safe and responsible deployment in critical infrastructure.

2. US-China to Hold High-Level Talks on Artificial Intelligence 🗣

US Secretary of State Antony Blinken revealed plans for the first intergovernmental dialogue between the US and China on artificial intelligence. The dialogue aims to address concerns surrounding the risks and safety implications of advanced AI technologies.

Set to occur in Beijing in the coming weeks, the talks will focus on sharing perspectives and best practices for managing these emerging technologies. Despite existing tensions, both nations have also agreed to enhance cultural exchanges and work towards stabilizing their relations.

3. Elon Musk’s xAI is Nearing $6 Billion 🚀

xAI, founded by Elon Musk, is on the verge of securing a whopping $6 billion at an $18 billion valuation, leaving investors scrambling to get a piece of the action. With heavyweights like Sequoia Capital and Future Ventures joining the funding round, the stakes are higher than ever in the world of AI investments.

Musk's intricate web of companies, including Tesla, SpaceX, and Neuralink, stand to benefit from xAI's cutting-edge technology, setting the stage for a digital revolution that could reshape our everyday lives.

4. Former Athletic Director Arrested for AI-Generated Principal Impersonation Scandal 🚨

A former athletic director at Pikesville High School got caught red-handed using artificial intelligence to impersonate the principal and spread hate-filled messages, causing chaos within the school community.

The scandal involved fake audio clips of the principal making racist and antisemitic comments, which were later proven to be AI-generated. The AI mischief led to disruptive school activities, investigations into improper payments, theft allegations, and a wave of public outrage on social media.

5. Microsoft's Cloud Revenue Soars 23% Thanks to AI Investments ☁️

Microsoft is riding high on a 23% surge in cloud revenue, reaching $35.1 billion! With a growing number of AI customers on Azure and increased contract values, the tech giant is making significant strides. Nearly 60% of Fortune 500 companies are already onboard with Microsoft's Copilot assistant, available on over 225 million PCs.

LLM Coding Showdown: Llama 3 vs ChatGPT, Copilot and more!

Click Image To Play Video 👆

Meta’s Llama 3 was just released a week ago.

So how does its coding compare to other AI models like ChatGPT, Gemini, Copilot, and Claude?

We’re running an LLM coding challenge to see if Meta Llama 3 can keep up.

Check out today's AI in 5.

Or see this related video:

Meta Llama 3: Everything you need to know in 10 minutes

🦾How You Can Leverage:

A common complaint using large language models — they’re dumb and they forget things!

Right?

Maybe not with this new update from ChatGPT.

After teasing its memory feature about two months ago, OpenAI finally started to roll out this feature to paid users.

(If you don’t see it on your ChatGPT Plus account, try signing out and signing back in.)

We took a deep look at this memory feature on today’s show to see if it’s going to be a game-changer or just a marketing piece.

In short, it depends on the way you use ChatGPT.

So, how do you use this feature and what are the best use-cases?

Glad you (rhetorically) asked.

Here’s the breakdown of what you need to know to get the most out of this new feature. 👇

1 – T️he problem it (kinda) solves 😒

In general, chats inside LLMs work in silos.

The information you share in one chat just kinda live there. And if you start a new chat, you …. gotta start over.

In theory, this memory update can really help users save time by remembering important details across multiple chats and creating a centralized memory bank of sorts.

Try this: There’s two different ways that ChatGPT can commit something to memory: by automatically deciding it’s an important detail and adding that detail automatically, or you can ask it to add something to your memory.

Right now, you cannot add a new memory any other way.

There’s one use-case we can all agree on, right?

2 – Memory can clash with C.I. 🥊

The concept of giving ChatGPT information to remember across all chats isn’t new.

OpenAI released a “Custom Instructions” feature in July of 2023 which gave ChatGPT similar abilities to remember information across chats about who you are and how you want ChatGPT to respond to you.

In our opinion, these two different features can clash a bit at times and would be better off being joined into one feature.

It’s a bit confusing to have both options, TBH.

Here’s a breakdown between the two:

Try this: Here’s the biggest difference between Custom Instructions and the new Memory feature, since they can work in tandem.

Custom Instructions: Allows you to both tell ChatGPT about you, and instruct ChatGPT to respond a certain way.

Can be used: Custom Instructions can only be toggled on and off account-wide.

Settings: Custom Instructions can only be accessed in the “Customize ChatGPT menu.”

Memory: Allows you to have ChatGPT remember certain things about you or how you want it to respond, think of it more as bite-sized nuggets of info you want ChatGPT to always remember.

Can be used: Memory can be both toggled on and off account-wide. But, you can also enable memory and then start a new chat without memory enabled by choosing ‘Temporary Chat’

Settings: First, you have to click the ‘Settings’ menu item, then the ‘Personalization’ panel.

3 – How (and if) you should use it 📋

TBH, this isn’t quite the game-changer that many people are making it out to be.

Don’t be confused by the word ‘memory’ — this doesn’t extend ChatGPT Plus’s context window past 32K tokens, or about 26,000 words. All it does is allows ChatGPT to remember (what it thinks are) important details about yourself, or how you use ChatGPT.

The context window limits still apply, although you might save some tokens by being able to draw on items saved in your memory bank.

Try this: We’ll give it to you straight on if and how you should use this:

Limited scenarios: If you use ChatGPT family for one or a few use-cases, you’ll probably find this new memory feature helpful.

You’ll be able to add and manage important information and instructions that will save you a lot of time and back-and-forth.

Many scenarios: We use ChatGPT for just about everything. A variety of internal reasons, many external sources, personal, trying to test and break things, etc.

In this case, the memory feature might do more bother than good.

More marketing than practicality?

As an example, let’s say our we have a memory that instructs ChatGPT to always summarize long content in short, informal bullet points.

But if we’re trying to research something for a competitor, we may want ChatGPT to reply with long-form content instead of bullet-points and in more formal language.

In this case, you’d either have to

a) start a Temporary Chat, or
b) delete that memory manually, or
c) ask ChatGPT to modify that memory or delete it.

Here’s the long and short of it — your mileage will vary on this new Memory feature.

It’s not going to change the way you or your team uses ChatGPT.

But, if you use ChatGPT for only a few limited use-cases, this feature will likely save you some time, tokens, and headache.

⌚

Numbers to watch

$40 Billion

Meta is expected to spend $40 billion on AI this year.

Now This …

We wanna know!

Vote to see live results

Have you entered our 1-year anniversary giveaway yet?

See below for sign-up details!

What giveaway?

This one 👇

Reply

or to participate.