- Inclined.ai
- Posts
- 🤖 GPT-4: ChatGPT's Big Upgrade
What's up? You're reading Inclined AI. TED LASSO season 3 drops today, and yet somehow, that’s not the biggest news out there.
Here's the big deal:
OpenAI released GPT-4 yesterday
Google announced new generative AI integrations
Anthropic is opening early access to Claude
THREE HUGE COMPANIES, ONE DAY FULL OF ANNOUNCEMENTS
OpenAI, Google, and Anthropic fought for eyeballs yesterday by releasing back-to-back-to-back big drops.
It’s a special day. So to celebrate, I decided to break this down as one mega post instead of two (or three) different headlines.
I’ll focus on the big takeaways for each one.
GPT-4 is Here
OpenAI’s graphic for the release
They did it. Everyone expected OpenAI to launch GPT-4 on Thursday of this week.
It turns out Pi Day was the perfect moment.
GPT-4 is, in fact, a text and vision model and is available to ChatGPT Plus users right now. The API currently has a waitlist (you can sign up here.)
They failed to mention when the model would open to public access.
Here are the highlights:
GPT-4 nails reasoning tests. That includes VQA testing that focuses on visual questions. You can now upload an image, and ChatGPT will grasp the idea. If you want it to explain a meme, it will explain a meme.
They ran GPT-4 through a wide array of exams. The results knocked my socks off. It tested in the top 10% of law students in the BAR Exam and received a 4 (out of 5) in the AP Calculus B Exam.
Sam Altman and company what to shake up AI alignment. Jailbreaking prompts will now work less, but hallucinations will still happen. Soon, their goal is to allow users to customize ChatGPT and steer it toward flexible behaviors.
GPT-4 scores higher on factual outputs and is more accurate. However, it will still miss the mark. It’s not perfect.
That last thing. “It’s not perfect” was a common statement from OpenAI’s talking heads.
You’ll marvel at the initial uptick in performance, but this update will feel like second nature in a matter of moments.
I like OpenAI President Greg Brockman’s quote from their demo, “it’s not perfect, and neither are you.” But, together, you make a hell of a team.
Google Steps Up, Sort Of
Pictured above: a trusted Google tester
Alphabet seemed to know the news was coming and tried to jump ahead of OpenAI’s monumental moment.
Google Cloud is adding generative AI support to Vertex AI and their own Copilot-like tooling. It’s a big step up from where they were regarding accessibility.
Speaking of access, the PaLM API is landing in developers' laps. I don’t know what people will do with a 540 billion parameter Large Langauge Model, but you can bet I’m sticking around to find out.
The API will allow builders to create AI tools, but the pricing details are sparse. So don’t switch off your current setup just yet.
Finally, the biggest news from the big-boy tech company does affect everyone: generative AI is coming to Google workspace.
Wait, technically; it’s already here. Autocomplete in Gmail is a form of generative AI, but these new tools are a level-up.
Picture a version of Google Slides where you can auto-generate images, audio, AND video into your slides. That’s a future I will race you towards.
Users can also prompt within Google Docs and Gmail to get a draft/template ready at a moment's notice.
Are you ready for the bad news?
All of this is for “trusted testers” right now. Why does Google keep doing this?
Every time they announce something unique like LLaMDA or the PaLM API, they mention that only trusted testers will get to use it.
Google, wake up. You have so many users that want these AI tools. Figure it out, plz.
They wanted to stand out this week, but that caveat shrouded any excitement I had for their news.
Anthropic’s News Gets Buried
The header from their Claude Blog…same that five times fast
They announced an early access release for Claude.
Claude is their Large Language Model that directly competes with GPT-4. Their new web app is launching with a Slack bot and an API.
I’ll link the early access request form here.
I don’t want it to sound like this news does not matter or that Claude is not powerful.
In fact, Claude powers Notion and Quora’s AI interfaces. It’s not a measly ChatGPT clone. Instead, it’s a clear competitor that hallucinates much less than the betting favorite.
The most puzzling part of this is not how promising Claude is. You should check it out.
What left pundits scratching their heads was Google jumping over this announcement. Google backs Anthropic. The tidal wave of news stories hurts Anthropic, and stacking news did not help.
On a typical day, the Claude news would lead this newsletter. But, sadly for them, this was not a typical day.
One Last Thing
Every one of these companies emphasized AI responsibility.
There were a lot of assurances that they could handle the data influx and control the AI they were putting out into the world.
It’s the biggest reason OpenAI waited to release GPT-4; they wanted to ensure it was ready for public use.
As the crowd begins to worry more about AI and the future, these companies building the next wave will need to develop high levels of trust.
Responsibility is key. Remember that.
A SPONSORED INTERRUPTION
Do you need to extract data from websites, PDFs, databases, and more?
You need Kadoa.
Kadoa is an AI web scraper that can collect any data for you in seconds. Kadoa uses generative AI to create web scrapers for any source you want. You define the data, sources, and schedule—Kadoa does the rest.
Kadoa is easy to set up, robust, and reliable. You can receive the data in any format you prefer. Plus, you can also integrate it with an API or other tools.
Don’t miss this opportunity to get early access to Kadoa and start extracting data effortlessly!
Quick Nuggets
📦 Unethical: Microsoft lays off their ethical AI team to double down on OpenAI
🗑️ Separating trash is made easy through machine-learning tools
📈 AI Acquisition: Spilo acquires Tinyclues to cement its lead in Europe’s AI-powered CRM market
⛑️ MedTech is not adopting AI right now—its adopting machine learning
🇺🇸 Gerrymandering is annoying, but will AI play a factor in the next round?
❗️ AI awareness is not that high, according to the Pew Research Center, and it still freaks a lot of people out
🩺 Wait, what: Nabla shows that MedTech is, in fact, using AI to talk to patients, lol
🎙️ 70 talks on generative AI are scheduled for Nvidia’s GTC online event
💊 New medicines are being created with the help of AI (that first MedTech article is looking silly)
🤝 Meet & Greet: discover more about Together, the team making an open-source version of ChatGPT
🤔 a16z asks who’s behind the generative AI platforms springing up today.
👀 Take a look at some medical students’ attempts to use ChatGPT at their university
🔥 Fresh Products
GooseAI - wants you to pay less for AI infrastructure (link)
Khan Academy - is launching a GPT-4-powered AI tutor (link)
Duolingo Max - is a new learning experience powered by GPT4 (link)
Shortify - save time by abbrv. anything you come across (link)
GPTweet - instantly reply to tweets for free (link)
BlogSmith - the first CMS fully loaded with AI tooling (link)
Relum Ipsum - create web copy without leaving Figma (link)
Jason AI - a conversational AI focused on B2B interactions (link)
Codenull - no-code AI that you can try for six months for free (link)
Momentum - takes every sales call and turns it into CRM data instantly (link)
Genius Ai - a smart keyboard assistant (link)
Study Crumb - a free paraphrasing tool for student essays (link)
Good Content, Diligent Dad
If you ever wondered if Darth Vader did chores around the Death Star, now you know.
That’s it for today. I hope you enjoyed the latest edition of inclined.ai - Davis.
Want more?
Upgrade today for less than the price of a latte a month and get benefits like:
A weekly Sunday deep dive into AI trends
Access to Q&A privileges to get your biggest AI questions answered
Exclusive discounts & invites to beta AI tools
