- Everyday AI
- Posts
- No. That's not how ChatGPT works. ā
No. That's not how ChatGPT works. ā
A guide on LLMs, Google Workspace AI update, OpenAI accused of copyright violations on YouTube videos, and more!
š Subscribe Here | š£ Hire Us To Speak | š¤ Partner with Us | š¤ Grow with GenAI
Outsmart The Future
Sup y'all š
Today's #HotTakeTuesday was a fun one!
We've probably spent more time inside of ChatGPT than 99.9% of humanity, so we figured today we'd wanna bust down some common myths about ChatGPT, and give it to you straight about how ChatGPT actually works.
Oh, and keep reading, as we have a pretty fun poll down below and the results might surprise you. š
āļø
Jordan
P.S. we completed our NVIDIA GPU giveaway today live on the show. Weāll announce the winner at the end of this newsletter! š
Today in Everyday AI
7 minute read
š Daily Podcast Episode: Thereās a lot of bad info out there about ChatGPT. Weāre cutting down the fluff and bringing you the truth. Give it a listen.
šµļøāāļø Fresh Finds: AI-powered customer research, DeepMindās co-founder frustrated with Google merger, and IBM partners with German AI startup. Read on for Fresh Finds.
š Byte Sized Daily AI News: Google Workspace gets an AI update, OpenAI accused of copyrighting YouTube videos, and Meta to release smaller version of Llama 3. For that and more, read on for Byte Sized News.
š AI In 5: Donāt know who to trust to learn ChatGPT? Hereās 5 red flags that tell if someone knows ChatGPT or not! See it here
š§ Learn & Leveraging AI: Sorry but thatās not how ChatGPT works. Hereās your guide to actually understanding LLMs. Keep reading for that!
ā©ļø Donāt miss out: Did you miss our last newsletter? We talked about Elon Musk predicting when AI will be smarter than humans, Spotify's new AI feature, and a game-changing ChatGPT feature in DALL-E 3. Check it here!
No that's not how ChatGPT works. A guide on who to trust around LLMs ā
Whether you're browsing social media or searching the web, there's so much bad advice out there about ChatGPT.
That super long one-fits-all prompt you found online? Not gonna cut it.
We canāt help but say it:
No. Thatās not how ChatGPT works.
We're tearing down the rumors about what ChatGPT is and how to use it.
Here's a beginner's guide on knowing who to trust around Large Language Models.
Join the conversation and ask Jordan questions on ChatGPT here.
Also on the pod today:
ā¢ Misconceptions around ChatGPT š¬
ā¢ Usage of Generic Business Prompts š
ā¢ Risks of Sole Reliance on AI Tools š«
Itāll be worth your 42 minutes:
Listen on our site:
Subscribe and listen on your favorite podcast platform
Listen on:
Hereās our favorite AI finds from across the web:
New AI Tool Spotlight ā Odaptos is AI-powered customer research, ScreenSnapAI is an AI screenshot manager for Mac, and Chat2DB is an AI data development and analysis platform.
Trending in AI ā DeepMindās co-founder is apparently frustrated with Googleās AI DeepMind merger.
Big Tech - IBM is partnering with German AI startup Aleph Alpha to push GenAI in Europe.
AI in Society ā South Korea is investing $7 billion in AI to keep an edge in AI chips.
AI Startups ā AI startup AlphaSense, valued at $2.5B, has crossed $200M in annual revenue.
1. Google Unveils New AI-Powered Features in Workspace Suite š¼
Google is stepping up its game with Gemini, offering AI-powered tools to enhance meetings and data security starting at $10 per user per month. With features like 'Take notes for me' and live translation in Google Meet, users can expect a seamless meeting experience. Additionally, Google is ramping up AI security measures.
2. OpenAI Accused of Unauthorized YouTube Video Use š³
OpenAI's alleged unauthorized use of YouTube videos to train GPT-4 has stirred major controversy in the tech world. The New York Times revealed that OpenAI may have breached YouTube's policies by using video transcripts for AI training. Google's response to these claims hints at a potential showdown between tech giants, raising questions about copyright laws in the AI landscape.
3. Google Unveils Custom Arm-based CPU for AI Data Centers š»
Google is shaking things up in the tech world with its latest custom Arm-based CPU, Axion, designed to support AI workloads in data centers. This new chip promises a 30% performance boost compared to general-purpose Arm chips and a whopping 50% improvement over Intel's existing processors.
4. Meta to Release Smaller Versions of Llama 3 š¦
Meta is gearing up to unveil two smaller AI models as a precursor to the much-anticipated Llama 3 release next week. These smaller models are designed to build excitement for the grand unveiling of Llama 3, which promises to be a game-changer in the world of open-source AI. As Meta dives into the realm of text generation and image processing, these upcoming models are just a taste of the technology to come.
5. Microsoft to Invest $2.9B in AI Infrastructure in Japan šø
Microsoft is making a bold move by pouring $2.9 billion into expanding its cloud and AI infrastructure in Japan, its largest investment in the country in 46 years! This investment will not only support AI development but also skill three million individuals in AI, showing a commitment to enhancing tech capabilities.
5 red flags that tell you if someone knows ChatGPT or not!
Everyone seems to be a āChatGPT proā now. Sorry to say, but theyāre giving you bad adviceā¦
So how do you know who to trust when it comes to learning ChatGPT?
Hereās 5 red flags that tell you if someone knows ChatGPT or not!
Check out today's AI in 5.
š¦¾How You Can Leverage:
No, thatās not how ChatGPT works.
Maybe weāve thought this in our head so many millions of times that we decided to say the quiet part out loud for #HotTakeTuesday.
Because hereās the reality.
Even some seemingly smart people who have their š© together in all things AI want to share AI tips and tricks.
Which is really, really bad advice.
Follow randos at your own risk, because most of the advice they give is ā¦. NOT HOW CHATGPT WORKS.
Not saying weāre the hands-down experts, but:
Our team has literally been using the GPT technology almost daily since 2020.
Weāve talked the literally the smartest people on the planet about GenAI for a year.
Our host Jordan literally spent his weekend re-reading these papers below.
Without further chit chat, letās go over 5 tell-tale warning signs that signal whoever youāre listening to about ChatGPT probably has no idea what theyāre talking about.
Oh, and make sure to go watch/listen to todayās show.
Like Anneli said, we all need to debunk the B.S. in the GenAI world to stay ahead.
![](https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/b30c8e6d-186a-4cbf-b5a5-a926edcabebe/debunking_bs.png?t=1712678503)
Letās get after itā¦ hereās 5 red flags that āexpertsā are getting wrong about ChatGPT.
1 ā ChatGPT is not good. Look at this response. š
If someone honestly thinks that a single screenshot of ChatGPT means anythingā¦. Turn around and run.
In todayās show, we gave an example online where someone entered a simple prompt, asking ChatGPT to come up with an advertising slogan for a Nike ad for womenās basketball icon Caitlin Clark.
(For context, Clark recently broke the NCAA all-time scoring record and is so good itās silly.)
After seeing Nikeās great ad copy, this person pondered what kind of result ChatGPT might give.
In other words ā a very simple input and output from ChatGPT shared on social media.
(Thatās not how ChatGPT works)
For fun, we took the simple copy and paste example, the ACTUAL ad text that Nike ran, and then we properly trained ChatGPT for the task and produced our own headline using our Prime, Prompt, Polish method.
Which advertising slogan do you think would best complete this ad?
![](https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/cec9700b-4112-4960-b875-315cbc34dad1/clark_screenshot.jpeg?t=1712678980)
Which advertising slogan do you think would best complete this ad?One of these is from a human copywriter, one is from a zero-shot prompt, and one is from ChatGPT that has been created with our PPP method. |
(Vote live to see the results, and weāll tell you in tomorrowās newsletter which one was written by who.)
What it means:
Well, you can skip straight to the good stuff and see what our live audience thought.
But in short, doing a simple input-output inside of ChatGPT and sharing results means absolutely nothing, and this is not a good way to learn about GenAI.
2 ā ChatGPT is never out of date with Bing! š
Wrong.
If anyone tells you that the āBrowse with Bingā feature inside of ChatGPT will keep your outputs up to date and hallucination-free doesnāt really know what theyāre talking about.
Thatās not how ChatGPT works!
Between the fact that Browse with Bing canāt actually visit specified pages to the finicky nature of the SEO that Browse with Bing relies on, your outputs arenāt necessarily better when you use Browse with Bing.
What it means:
Youāve really gotta understand the pros and cons of Browse with Bing and also how it works.
Luckily, weāve already covered that one in depth.
3 ā Try these 25 prompts to 10X š¤¦
Grossbarfyuck. š¤®
If anyone swears to you that their magical ChatGPT prompts are gonna do absolutely anything for you or your company, run for the hills shorties.
Weāll say it loud so those of yāall in the back can hear.
Prompts.
Do.
Not.
Work.
What it means:
Generative AI systems like large language models areā¦. Generative!
We think that LLMs like ChatGPT are deterministic like search engines. Theyāre not.
(Hint: Deterministic = the same/consistent output that is predetermined. Generative = can get wildly different results.)
And jumbo models with trillions of parameters really fail with zero-shot prompts. Thatās when you go into a new chat, paste in one giant prompt with no examples, hit enter, and expect a usable output.
(Againā¦. Thatās NOT how ChatGPT works.)
4 ā How to make ChatGPT write like you āļø
Anyone that gives you advice on and doesnāt break down covering unstructured data into structure rules for ChatGTP to followā¦. Doesnāt know what theyāre talking about.
#Sorrynotsorry.
Can a random prompt and pasting in examples make ChatGPT slightly less robotic and a little more like you?
Sure.
And if your (or your companyās) writing style is a straightforward, vanilla-ish tone, ChatGPT might do OK at that.
But can a simple prompt and pasting in examples of your writing actually produce results similar to a unique writing style or brand voice?
No. Thatās not how ChatGPT works.
What it means:
Hereās the thing - Large Language Models donāt truly even understand words.
They convert words to tokens based on context and use super smart deep learning superpowers to try and predict the next token in its output based on its translating of your text into tokens.
In other words, for your AI efforts to work, you really need to understand LLM tokenization and go through the process of turning unstructured words into structured rules for ChatGPT to follow.
This show will give you a head start.
5 ā Free version of ChatGPT is good enough š¬
Can we skip all the nonsense here and say, āNo, thatās not how ChatGPT works?ā
Would you rather have a paper, pen and landline to run your business or the worldās fastest computer?
Yeah, thatās kinda the difference between the free version of ChatGPT and the paid version.
What it means:
If someone tells you that the free version of ChatGPT is good enough or if theyāre trying to give you āX free alternatives to ChatGPTāā¦. That means they donāt actually know what theyāre doing.
This show might be a month or 5 old, but the gist of it still holds true.
The free version of ChatGPT shouldnāt be used. Likeā¦. Ever.
ā
Numbers to watch
$15 Million
Texas will save $15 million by using GPT technology to grade written answers for its STAAR tests.
Now This ā¦
Let us know your thoughts!
Vote to see live results
What other LLMs / AI models do you want us to cover on the show? |
ā¦.. and the winner is
Congrats to Justin L for winning our NVIDIA GPU giveaway. Weāll be in touch so check your email.
Reply