- Everyday AI
- Posts
- GPT-4o: The real and the fake of ChatGPT's new model 🤥
GPT-4o: The real and the fake of ChatGPT's new model 🤥
7 things to know about OpenAI's new GPT-4O, Replit cuts 20% of its staff, and U.S. and China meet on AI
👉 Subscribe Here | 🗣 Hire Us To Speak | 🤝 Partner with Us | 🤖 Grow with GenAI
Outsmart The Future
Today in Everyday AI
7 minute read
🎙 Daily Podcast Episode: There’s more confusion and bad info out there about OpenAI’s new GPT-4O model than there is good. We do something about it, giving you the 7 things you need to know about GPT-4O. Give it a listen.
🕵️♂️ Fresh Finds: A Mac app that allows you to build agent workflows, voice actors suing AI voice studio, and Sony Music going after 700+ companies for training on their music. Read on for Fresh Finds.
🗞 Byte Sized Daily AI News: U.S. and China meet on China's 'misuse' of AI, Stability AI could be up for sale and Microsoft's emissions are soaring, read on for Byte Sized News.
🚀 AI In 5: You ever spend like 17 minutes just looking for a chat inside of ChatGPT? Same. We found a super simple trick around that. See it here.
🧠 Learn & Leveraging AI: Don’t waste hours wondering what’s what when it comes to the newest version of OpenAI’s LLM, GPT-4O. We break it down and show you some secrets (legit) no one is talking about. Keep reading for that!
↩️ Don’t miss out: Did you miss our last newsletter? We talked with a Microsoft PM leader about her AI secrets, Google’s new AI search filter, the U.S. government’s $32 billion plan for AI an more. Check it here!
Microsoft PM Speaks - How AI is Shaping Product Management 🔓
Straight facts about GPT-4O 👇
↳ Everyone's talking about GPT-4o but no one's actually investigating it.
↳ No, not everything is free.
↳ Yes, OpenAI is still not telling the truth about some stuff.
↳ Yes, we did all the homework for you so you just gotta show up and learn
Today, we sifted through the noise, BS and misconceptions around OpenAI’s new model.
Join the conversation and ask Jordan any question on the new GPT-4O here.
Also on the pod today:
• OpenAI’s secret answers engine 🤓
• The new model that’s coming next 🔮
• OpenAI stil fibbing about memory 🧠
It’ll be worth your 52 minutes:
Listen on our site:
Subscribe and listen on your favorite podcast platform
Listen on:
Upcoming Everyday AI Livestreams
Friday, May 17 at 7:30 am CST ⬇️
Here’s our favorite AI finds from across the web:
New AI Tool Spotlight – UserCall uses AI to conduct customer research at scale, Glato uses AI avatars to make UGC videos at scale for your business, Vidix is an AI-powered platform that allows you to build agents that interact with your Mac computer.
Trending in AI – Voice theft by AI? Apparently. A group of actors are suing AI voice startup LOVO over alleged misuse.
Big Tech - Microsoft is offering to relocate up to 10% of their staffers in China to move to the U.S. or other nations that are allies with America.
AI Coding — AI coding platform Replit, according to The Information, has chopped 20% of its staff.
AI coding startup Replit has cut 20% of staff as it looks to pivot to enterprise sales.
The unicorn startup, which has raised over $220MM in funding, follows other AI startups like Tome that have made similar enterprise pivots.
More here:
— Stephanie Palazzolo (@steph_palazzolo)
7:35 PM • May 16, 2024
AI in Music — Sony Music sends letter to 700 companies saying they’re not authorized to use their music to train their models.
AI and Computing - Chip giant Qualcomm and Ampere Computing team up to release a new chip to power the future of AI.
Future of AI and Chatbots — Are AI voices and low-latency interactions too dangerous? Some think that chatbots are going to easily manipulate us humans in the future.
Read This – A 5-person team in San Francisco created an open source version of AI smart glasses for only $20.
1. U.S. and China Finally Meet on AI, U.S. talks China’s ‘Misuse’
The U.S, and China held talks on AI safety and risks, with US raising concerns about China's misuse of AI while Beijing criticized US restrictions and pressure in the field of AI.
The first U.S.-China talks on AI were a result of a November meeting between presidents Joe Biden and Xi Jinping. China has been heavily investing in AI technology while the US has imposed restrictions on AI trade and investments in China. Read more on the meetings here.
2. For Sale? Former Juggernaut Stability AI Could Be For Sale 💸
British AI startup Stability AI is facing financial troubles and has held discussions with potential buyers for a sale. The company generated low revenue and high losses in the first quarter of 2024 and owes a significant amount to cloud computing providers.
Stability AI is facing financial difficulties and has held discussions with potential buyers for a sale. In the first quarter of 2024, the company generated less than $5 million in revenue and lost over $30 million. The company owes close to $100 million in outstanding bills to cloud computing providers and others, adding to their financial struggles.
3. Microsoft’s Stock Price (and Emissions) Are Way Up 📈
Microsoft's 2024 Sustainability Report shows that their emissions and water usage have increased, with AI technologies being a major contributor. Microsoft's emissions and water usage have increased, primarily due to new AI technologies.
Despite setting ambitious climate goals, Microsoft has been unable to decrease their carbon footprint as they invest more in AI. However, the company is taking steps to address these issues and make their operations more sustainable. The company plans on investing in more sustainable practices, such as using renewable energy and reducing waste.
4. Hugging Face Offering $10 Million of Compute For Devs 💻
Hugging Face, a major player in the AI field, is investing $10 million in shared GPUs to support smaller developers and academics in the AI community.
The initiative, named ZeroGPU, aims to provide accessible computational resources for creating advanced AI technologies without the need for hefty investments. By democratizing access to GPUs, Hugging Face hopes to level the playing field and foster a more collaborative and transparent environment in the AI industry. This move could potentially pave the way for more diverse innovations and opportunities for aspiring AI enthusiasts and startups.
5. AI-powered Robot Gives Real-Life Graduation Speech 🤖
An A.I. robot named Sophia gave a commencement address to the graduating class at D’Youville University in Buffalo, New York. The decision was met with mixed reactions.
Sophia gave generic advice to the graduates, sharing common themes found in traditional commencement speeches. The decision to have a robot speak sparked backlash from students who felt it was impersonal, especially for those who had virtual high school graduations due to the pandemic.
Slay the ChatGPT chaos with one simple Chrome extension
You ever find yourself in a never-ending scroll to find that chat from last month?
Like… yo. Where’d it go?!
It’s kinda crazy that OpenAI still hasn’t given us a way to better organize chats with folders and simple color coordination.
Sure, there’s a handful of Chrome extensions that do this pretty well, but they’re all mostly full of bloat, adding like 50 other features you may or may not want.
We found one of the simplest Chrome extensions to better organize your ChatGPT chats.
Check out today's AI in 5.
Or see this related video:
Does Claude-3 beat GPT-4 or Gemini?
🦾How You Can Leverage:
What’s that?
You don’t have like 15 hours to explore the depths of the new GPT-4O release from OpenAI?
We don’t blame ya.
After all, isn’t Generative AI supposed to help us win back our time?
Don’t worry — we’re dorks and have spent (maybe a little too much) time digging into the depths of the new powerhouse LLM offering from OpenAI.
After OpenAI announced its new model and features at its Spring Event, the AI world has been buzzing ever since about what it can (or can’t) do.
Lettuce warn ya — there’s a lotta bad info out there. And confusion.
So today, we simplified it be creating the 7 things you need to know about GPT-4O, and what OpenAI isn’t really telling the truth about.
And there’s SOOOOO much info in today’s show, that we’re gonna keep the 7 point breakdown short and to the point.
Here we gooooooo. 👇
7 – Not everything is out yet or free ⁉️
Not all capabilities of GPT-4O are live and most won't be free. Only the ‘base’ GPT-Omni model will go to free users, but not all of its capabilities will.
GPT-4O isn’t currently only available to paid ChatGPT Plus users, but should roll out to free users any day/week now.
Free users will likely not have access to the show-stopping features — the “Her” AI assistant that can see, hear and talk in real-time, and the AI agent-esque desktop version of ChatGPT.
Those also haven’t been rolled out to paid users yet, but likely will be soon.
Our take: People are saying, “Oh! GPT-4O is free. We can cancel ChatGPT Plus!” Kinda. But also kinda wrong. The GPT-4O model itself isn’t the big takeaway here. It’s the “Live” Omni assistant and the desktop program that will make ChatGPT Plus well worth the $20/mo.
6 – Some of the best features weren’t announced 📢
Like… whoa.
Did you know the new version of GPT-4O can create new fonts, spin up 3D renderings, create more consistent illustrations inside Dall-E, and create intricately detailed photo illustrations?
Our take: Call us keto, but we’d say some of the unmentioned portions of the GPT-4O model are straight nuts.
If you want to see the whole list, go to this page and check out the examples an try some of the prompts yourself.
5 – We’ve seen GPT-4O before 🕵️♂️
Halt! Who goes there?
Oh, that’s you im-also-a-good-gpt2-chatbot!
Like we talked about a few weeks ago, OpenAI had ‘leaked’ some mystery models on the Chatbot Arena.
And OpenAI confirmed one of those mysterious models was in fact GPT-4O.
GPT-4o is our new state-of-the-art frontier model. We’ve been testing a version on the LMSys arena as im-also-a-good-gpt2-chatbot 🙂. Here’s how it’s been doing.
— William Fedus (@LiamFedus)
5:01 PM • May 13, 2024
Our Take: The ELO scores for im-also-a-good-gpt2-chatbot were almost quite literally off the charts. For those new around here, ELO scores are kinda just a point system for which LLM is best.
Users put in a prompt, and see two responses from unnamed models. Then, they chose the best response. And im-also-a-good-gpt2-chatbot (now known as GPT-4O) was crushing faces and taking names, well ahead of Claude 3 Opus, Gemini 1.5 Pro and Meta’s Llama 3.
4 – There’s more models out there 🔍
What’s that?
More rumors?
Well, im-also-a-good-gpt2-chatbot wasn’t the only mystery model that made its way to the Chatbot Arena.
im-a-good-gpt2-chatbot and GPT2-chatbot also made dorky splashes around the AI corners of the interwebs.
Our Take:
Sooooo…. are there more models coming soon?
Yes. OpenAI literally flashed this on the screen for about 0.2 seconds during their Spring Event presentation.
Mistake or subliminal messaging?
![](https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/e6d84be4-4dc1-48d4-b137-f4052cb43e6e/CleanShot_2024-05-16_at_14.21.31.jpeg?t=1715887945)
Like we talked about on today’s show, we expect to see the paid ChatGPT Plus be upgraded from GPT-4O a few weeks/months after this rollout is complete.
3 – UI/UX updates brought more features 🛠️
Small tweak? Or big change to how we use models?
Aside from some tidying up of the interface, the user experience inside ChatGPT did change.
Below your output, you can now swap the model.
![](https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/e736e797-3e3b-496f-bb1d-71697c5525f4/CleanShot_2024-05-16_at_14.38.31.jpeg?t=1715888388)
Our Take:
Prompting is a never-ending journey. Kinda like distance running. You’ve gotta always get your steps in and pound the pavement, or you’ll lose your skills.
Trying the same prompt between GPT-4 an GPT-4o is a great way to see how prompt handling has changed between models.
The sooner you learn, the more you’ll be able to take advantage.
2 – ChatGPT kinda fibbed about memory. Again. 🧠
Did ChatGPT flat out lie about ChatGPT’s memory?
Not really.
Semantics, maybe.
But OpenAI has touted GPT-4o’s 128K memory, or the amount of information a chat can store in its context window before it starts to forget things. That’s roughly 95,000 words of back and forth conversation with ChatGPT before it starts to just respond with:
Huh? Whuuuuut?
128K isn’t that bad, but it’s well behind other LLM competitors.
Buuuuuuuuuuut…..
OpenAI isn’t super clear about the whole 128K limit. Turns out, that’s still only if you’re using the API. So the vast majority of everyday users getting our LLM inside of ChatGPT are still stuck with 32K memory, or about 28,000 words.
Our Take:
Memory and small context window is one of the last areas where OpenAI is not able to compete.
That might be one of the reasons Sam Altman is out there trying to raise $7 trillion?
1 – ChatGPT DID build an answers engine 🚂
Real talk — this is dang impressive y’all.
OpenAI didn’t even mention what we think may be its biggest updates: Browse with Bing.
For the uninitiated, Browse with Bing is a built-in feature in ChatGPT Plus that allows the model to query the internet (with Bing) to find information past its 2023 knowledge cutoff date.
About nine months ago, the Browse with Bing functionality was so finicky and unreliable, we had to warn people not to use it.
There’s been a few updates since, but now?
Dang. It’s actually a straight up answers engine now.
Our take:
AI search is the future of the internet. And we believe that OpenAI is still hard at work on a search engine type platform, or maybe an answers engine.
What’s that?
Think of Perplexity or Google’s Search Generative Experience.
OpenAI wants a slice of the search pie. And with this under-the-radar update to Browse with Bing, we think they could start that shift sooner rather than later.
Check out this impressive response:
![](https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/e1ad1fc5-c62f-4bb7-9b33-213f855228f4/CleanShot_2024-05-16_at_00.25.15.png?t=1715889102)
⌚
Numbers to watch
$10 Million
Hugging Face committed $10 million of free GPU compute to take on bigger AI companies.
We wanna know below!
What's your favorite part of the newsletter? |
Reply