šŸ§  What You Need To Know About GPT-4

ChatGPT-4 release, Headlines, Links and more

Weā€™re not asking for your sympathy.

Itā€™s just that when you put together an informative, entertaining newsletter like we did the other day for a noonish send, and Sam Altman decides that 1:00 p.m. that day is the time for the Chat GPT-4 release, it stings.

The sun rose again today, though, and as you can imagine, this missive is going to be ChatGPT heavy. Stay tuned to the end of the email for a GPT-4 deep dive.

In the email today:

  • What you need to know about GPT-4 šŸ§ 

  • Headlines šŸ“°

  • Links šŸ‘€

  • SPECIAL: GPT-4 deep dive šŸ³

Here we go.

Here Is What You Need To Know About GPT-4 šŸ§  

SmokeBot had one of those ā€œwhere were you when X happenedā€ moments on Tuesday when this Sam Altman tweet cryptically-but-clearly told us that GPT-4ā€™s launch was super-imminentā€

The tweet included Sam sort of blankly looking into the camera sitting next to a backpack which we can only assume contained $50 million in hundred dollar bills. Thatā€™s what weā€™d do if we were him, anyway.

Hereā€™s what you need to know:

  • GPT-4 is officially what was released.

  • It is the new large-language model (AI algorithm) that will power an updated version of ChatGPT.

  • You can join the waitlist or use it now if you're a ChatGPT Plus subscriber. ļæ¼

  • It will also power new AI tools built by third-party developers.

Those are the basics.

But what does it do?

  • Unlike the previous version, GPT-4 can read, analyze and make sense and use of images. That feature is not available yet, but itā€™s coming.

  • ChatGPT-4 is far more powerful than the prior version, with a broader knowledge base and enhanced problem-solving capacity.

  • The new release has a cutesy creative side too ā€” it gave a forced but accurate summation of the plot of ā€œCinderellaā€ in 26 words beginning A-to-Z, in order.

  • Where the prior iteration was essentially a C+ student and a coin flip to pass a medical licensing exam, ChatGPT-4 is as smart as the most successful professional you know.

  • ChatGPT-4ā€™s capacity to produce content is significantly bigger ā€” it can now take in 25,000 words and generate long-form content. This will be a boon to formula authors like John Grisham and James Patterson, though we doubt ChatGPT-4 is going to produce the next ā€œThe Great Gatsby.ā€ At least not yet.

  • Remarkably, GPT-4 can identify and comprehend images and what you might want to do with them. Think a picture of an apple pie and the chatbot spitting out a recipe, or GPT-4 taking a meme you donā€™t understand and telling you what it refers to. Pretty wild.

So there is plenty to be excited about here. But you know SmokeBot. SmokeBot spills tea when tea needs spilling.

ChatGPT-4 still ā€œhallucinates,ā€ i.e., like a 6-year-old, if it doesnā€™t know the answer, it just makes one up. Itā€™s also still vulnerable to manipulation by prompts nakedly intended to trick it into saying something dumb.

Perhaps worse yet, ChatGPT-4 is not only still wrong too often, itā€™s confidently wrong, like your drunk uncle at Thanksgiving.

Candidly, SmokeBot thinks that in the haste to press OpenAI/Microsoftā€™s advantage over Google and Meta and everybody else, ChatGPT-4ā€™s release was just a tiny bit rushed.

Headlines šŸ“° 

PricewaterhouseCoopers LLP implements AI for its lawyers: Approximately 4,000 attorneys in 100 countries worldwide will be using a firm chatbot to ā€œspeed up work from due diligence or regulatory compliance to broader legal advisory and legal consulting services.ā€ āš–ļø 

Google plans to broaden AIā€™s application in medicine: ā€œHealth is a company-wide effort,ā€ said Karen DeSalvo, the tech monolithā€™s chief health officer. AI will be integrated into ā€œmedical exams and AI-assisted research.ā€ šŸ„¼ 

April 18 is coming, and Chat GPT-4 is ready to help: Itā€™s not just the lawyers who might be in trouble, as Chat GPT-4 could separately become a go-to tool for taxpayers and put accountants out of work. šŸ§¾

New York City seeks ā€œAI Czarā€: The city that never sleeps wants to hire a ā€œdirector of Artificial Intelligence and Machine Learningā€ to educate and inform the various bureaucratic entities on AIā€™s ā€œupsides and risks.ā€

Links šŸ‘€

SPECIAL: GPT-4 Deep Dive

With so many discussions and debates going on all over the internet regarding GPT-4 today, it can be easy to get overwhelmed. To help you out, here are the five most important aspects/changes to know about the new ChatGPT-4:

  1. New Multimodal Functionality

  2. Better Performance

  3. More Built-In Safety and Accuracy

  4. "Steerability"

  5. AI is Becoming Big Business

 

The Launch of GPT-4

While there have been several major AI advancements hitting the headlines recently, none have matched the impact of Open AIā€™s ChatGPT, which introduced the softwareā€™s last update, GPT-3.5, in late 2022.

The AI-powered language modeling technology has been one of the hottest topics in the business world since the release of GPT-3.5, and itā€™s created some heated arguments and controversy along the way.

With the announcement of GPT-4, OpenAI introduced the fourth major update to its ChatGPT system first published in 2018.

Due to the extreme complexity of this technology, it will take months to assess all of the different elements involved with ChatGPTā€™s update to GPT-4. But after diving into all the information, Smoking Robot has identified five of the most important things to know about this potentially world-changing release from OpenAI Labs.

1) ā€œMultimodalityā€

Last week, Microsoft Germanyā€™s CTO Andreas Braun started a surge in speculation when he leaked the upcoming release of GPT-4 and hinted at the multimodal capacity of the new system.

That was confirmed with the release of GPT-4, which now includes the ability to read images as prompts and explain what the images show as a text response. It can also provide further analysis on an image. For example, you could ask what is most likely to happen next with an image showing a baseball as it hits the bat and get a kind of basic "prediction" from the AI.

In another example on the OpenAI website, users can upload a photo showing several ingredients, ask "What can I make with these ingredients?", and GPT-4 will provide a list of options based on the photo.

Although this is obviously one of the biggest changes/additions to the technology, many expected to see more than image recognition capabilities, and Braun did mention the possibility of videos as a new medium while speaking to the press last week.

2) Upgraded Performance

Along with the added functionality in image analysis, ChatGPT is also getting a significant upgrade to its overall performance according to the tests done by OpenAI Labs.

In a post on its website, OpenAI claims that GPT-4 is "more reliable, creative, and able to handle much more nuanced instructions than GPT-3.5." You can dive even deeper into the testing and results by checking out the company's GPT-4 Technical Report.

GPT-4's upgraded creative capabilities help it generate better results for use cases like song composition, screenplay creation and custom writing styles. It is also now capable of analyzing over 25,000 words of text.

3) Improved Safety and Accuracy

OpenAI also emphasizes GPT-4's ability to "solve difficult problems with greater accuracy" adding that the new system is "82% less likely to respond to requests for disallowed content and 40% more likely to produce factual responses than GPT-3.5."

The organization says that it utilized more human feedback along with many lessons learned from previous versions of the system when training GPT-4. They added that they also worked with a group of experts from areas relating to AI security and safety during development.

4) ā€œSteerabilityā€

GPT-4 also includes a new feature called "System Messages" that introduces more "steerability" for developers and eventually, users.

While previous versions of ChatGPT used a fixed tone, verbosity and style for responses, GPT-4 will allow API users to significantly customize their usersā€™ experience by prescribing the AI's style and task using "system messages."

OpenAI also acknowledged that this new feature is also now "the easiest way to 'jailbreak' the current model."

5) Making AI Profitable

Microsoft has made headlines over the past few months for its partnership with OpenAI Labs for exclusive access to ChatGPT and other OpenAI models, and the AI research firm made sure to thank Microsoft for their help throughout the development process.

Microsoft Azure's AI-empowered infrastructure played a major role in the testing and creation process, and the tech giant certainly took this opportunity to leave its mark on the platform.

OpenAI also listed a group of other significant organizations as official GPT-4 partners including Stripe, Morgan Stanley, Duolingo, the Khan Academy and the Government of Iceland.

Many in the AI world are critical of OpenAI's move towards corporate partners and leadership, including former team member and co-founder Elon Musk, who posted a tweet about his concerns last month.

 

Impressively Imperfect

It's obvious to see that OpenAI has come through with some impressive upgrades to its ChatGPT system with the release of GPT-4, but does the product match the unprecedented level of anticipation and excitement that grew over the last few months?

That's a question we'll leave up to you to answer. Just don't try GPT-4 expecting perfection. This isn't the final solution to all the world's problems, but it's still a big step in the advancement of AI technology.

Oh, and if you're not already a ChatGPT Plus member, don't forget to have your credit card handy. GPT-4 requires ChatGPT Plus members and a $20/month subscription fee.