- Everyday AI
- Posts
- Stop making this one ChatGPT mistake š«
Stop making this one ChatGPT mistake š«
š How to stop ChatGPT from lying, Meta and IBM's AI Alliance, Runway's Motion Brush feature, and more!
Everyday AI
December 05, 2023
Outsmart The Future
Today in Everyday AI
7 minute read
š Daily Podcast Episode: Thereās a ChatGPT mistake that people keep making and itās causing hallucinations. Hereās how to fix it. Give it a listen.
šµļøāāļø Fresh Finds: AI that upscales your images, Yahooās new AI advertising, and bedtime stories from an AI Jimmy Stewart. Read on for Fresh Finds.
š Byte Sized Daily AI News: Meta and IBM's AI Alliance, concerns for Teslaās AI, and schools using AI for safety. For that and more, read on for Byte Sized News.
š AI In 5: Runway has a great new(ish) feature called Motion Brush! Weāre showing you how it works. See it here.
š§ Learn & Leveraging AI: Wondering why ChatGPT keeps getting things wrong? Weāll show you the mistakes a lot of people are making after the recent ChatGPT update. Keep reading for that!
ā©ļø Donāt miss out: Did you miss our last newsletter? We talked about the future of AI video generators, Google's Gemini AI postponed to 2024, and telling the difference between AI and humans. Check it here!
The ChatGPT Mistake You Donāt Know Youāre Making š«
Recently people have been reporting āmore mistakesā from ChatGPT.
But ChatGPT isnāt the problem.
You keep making the same mistake that's causing hallucinations and incorrect information.
And you probably don't know you're making it.
We'll tell you what it is, and how to avoid it so you can get better results.
Join the conversation and ask Jordan any questions about ChatGPT.
Also on the pod today:
ā¢ Limitations of custom GPTs š
ā¢ ChatGPT knowledge retention š§
ā¢ Quick ChatGPT basics ā
Itāll be worth your 36 minutes:
Listen on our site:
Click to listen
Subscribe and listen on your favorite podcast platform
Listen on:
Upcoming Everyday AI Livestreams
Wednesday, December 6th at 7:30 am CST ā¬ļø
Thursday, December 7th - Grow your brand with AI and a podcast
Friday, December 8th - Product Strategy in the Age of AI: A User-Centric Approach
Hereās our favorite AI finds from across the web:
New AI Tool Spotlight ā Spellar AI is AI-powered speaking copilot, Magnific AI helps to upscale your images, and Hitab lets you customize your tabs with ChatGPT.
Big Tech ā OpenAIās COO thinks AI for business is overhyped?
Trending in AI - Yahoo released Blueprint, a new AI feature for its advertising platform DSP.
LLMs ā This new trick allows AI models like GPT-4 to be jailbroken usingā¦AI.
Pop Culture ā The Calm App now offers bedtime stories from an AI Jimmy Stewart.
1. Meta and IBM's AI Alliance š¤
Meta Platforms and IBM have launched a coalition with over 50 AI companies and research institutions, advocating for an open model of AI. This AI Alliance, including Intel, Oracle, and academic institutions, aims to foster open innovation and science in AI, countering the proprietary approach of firms like OpenAI. The alliance focuses on diversifying AI development, emphasizing the need for multiple providers to reduce reliance on single vendors.
2. Tesla's AI: Whistleblower Concerns š
Elon Musk, CEO of Tesla, touts the company's AI as the "best real-world AI," powering its self-driving technology. However, whistleblower Lukasz Krupski, a former Tesla technician, raised concerns about the readiness of Tesla's Autopilot feature for public use. Krupski leaked confidential Tesla files, including customer complaints about the Autopilot system, to German newspaper The Handelsblatt. These documents revealed issues like sudden acceleration and emergency braking in Tesla vehicles.
3. Trump Accuses AI of Tarnishing His Image šŗ
Former US President Donald Trump has accused the Lincoln Project, a political advocacy group, of using Artificial Intelligence (AI) to negatively portray him in ads. The ads, aired on Fox News, featured Trump's gaffes, which he claims were doctored using AI. The Lincoln Project, known for its opposition to Trump, denied these allegations, asserting they simply showcased reality.
4. AI for School Safety š
School districts across the United States are turning to artificial intelligence for enhanced safety measures. In Louisiana's Iberville Parish, the district has implemented ZeroEyes gun detection software into school cameras. This AI system, developed by former military and law enforcement personnel, can identify potential guns and alert authorities. The software is designed to work with existing camera systems, providing an additional layer of security without compromising privacy.
5. Getty vs. Stability AI: Copyright Lawsuit Heads to Trial šļø
Getty Images' lawsuit against Stability AI, the company behind the AI image generator Stable Diffusion, is set to go to trial in the UK. Stability AI attempted to have the case dismissed, but the court found inconsistencies in its claims. This legal battle centers around copyright infringement issues, highlighting the complexities of intellectual property rights in the age of AI-generated content.
Runway Motion Brush Simple Tutorial! š
Runway has a great new(ish) feature called Motion Brush!
This tool allows you to animate certain parts of your image to create a custom video.
We're showing you how to use it!
Check out today's AI in 5.
š¤·āāļø Whatās Going On and Why It Matters:
Hereās one of the biggest ChatGPT mistakes you probably donāt know youāre making: knowledge retention.
If youāve moved past the beginners stage, youāve probably gone all-in on some new features of ChatGPT.
Like Jordan talked about on todayās show, the Default mode is actually useful now, with the ability to upload files and connect to the internet.
And GPTs? Everyoneās running to create a custom GPT with their companyās knowledge. And it only takes minutes.
But hereās the problem, yāall.
Knowledge retention.
See, ChatGPT can be a little skittish when it comes to properly understanding and recalling the documents that you upload.
And with the GPT store reportedly now opening in January 2024, this knowledge retention issue could be even more problematic.
As always, when a monumental update comes out, we donāt blindly follow the crowds.
At Everyday AI, we carefully and meticulously explore GenAI systems, pushing them to their limit and finding the holes no one knows about.
And then we share them with you.
So letās break down the knowledge retention issue and what it means to you and your biz.š
š¦¾How You Can Leverage:
New bells and whistles often means new frustration and facepalms.
But donāt worry, shorties, we did all the hard work for you.
As we found out on todayās show, knowledge retention inside ChatGPTās new default mode and custom GPTs can be a real headache if you donāt know what youāre doing.
Uploading docs as a knowledge base is supposed to improve your outputs. Likeā¦ a legit blessing!
Right?
But, if you dive in too fast, it could actually be a pseudo blessing wrapped in a false sense of security with a pretty red bow of hallucination.
Each mode handles knowledge retention and knowledge calling a bit differently. So, letās go over the essential 1-2-3 here so you can confidently unwrap the gift of GPT.
1 ā Test before you deploy ļøš
Yoooooo. You gotta be careful.
As we showed on the show, always test the connection/recollection abilities anytime you share a doc with ChatGPT. Whether youāre using the default mode or the custom GPTs, we recommend āhidingā random facts in the latter half of the doc.
And every so often, ask ChatGPT a very non-directed question about those facts. If it fails that test, you need to improve your prompting inside the default mode, or improve your configuration instructions in custom GPTs. (More on that below)
Try this:
TBH, custom GPTs usually do a better job at retaining knowledge.
Inside default mode, youāll want to keep an eye on the 32K token limit. And it never hurts to both keep track of tokens and also re-upload the doc every so often to make sure the default mode can continue to access the doc.
2 ā Call the knowledge š£
Hello, document, are you there?
Even after you upload your docs for ChatGPT to use as knowledge, it doesnāt mean it will remember. Or use that knowledge for each reply.
You may think that ChatGPT is āusingā the documents you upload in your responses. But in reality ā it might not be. Thatās why youāve gotta call your knowledge base, like we showed in todayās show.
Try this:
When in doubt, add a simple call to the knowledge base. As in, āYoooo ChatGPT, check the doc named āyour doc nameā for this information. Can you tell me about it?ā
Whether youāre using a custom GPT where youāve uploaded company docs or a chat in the default mode, it doesnāt hurt to call ChatGPT to use any docs youāve uploaded.
Something as simple as:
āBefore responding, please make sure to carefully analyze the documents in your knowledge baseā in a custom GPT.
or
āBefore responding, please make sure to carefully analyze the (document name) that I uploaded previously.ā In the default mode.
3 ā Always be skeptical
Sorry not sorry.
Hereās whatās gonna happen in early 2024 when the GPT App Store opens. Everyoneās gonna rush into the store and immediately use every GPT under the sun.
Cool.
(Prolly us too, TBH.)
Yes, the OpenAI GPT store will likely be bonkers when it comes to building your biz or career. (The good kind of bonkers and the bad kind, prolly.)
But, make sure to be skeptical when that time comes.
Check who the developer is. See their privacy policy. Are they using outside APIs? Is their website even secure!? Lolz. (But seriously. If a dev has an unsecured website, you really shouldnāt touch anything they develop.)
Try this:
When the GPT Store drops, due your due diligence. Especially when using a third-party GPT for your day-to-day business operations.
Not only do you need to make sure the GPT works correctly, but you should also go through a privacy and security checklist on the GPT if youāre going to rely on it for your biz.
Now This ā¦
What are your thoughts?
Vote to see live results
Are you excited about the OpenAI GPT store? |
Reply