8 min read

Uh oh...Is ChatGPT Getting Worse?

6 Skills You Need to Become an Expert Prompt Engineer

Top Insights

Is ChatGPT Getting Worse?

This week, a Stanford University study showing ChatGPT's performance is dwindling lit the AI marketing world on fire.

The team evaluated GPT-4 using a series of 500 math problems in March, of which the chatbot answered 488 correctly. In June, they ran the same test, but this time it only got 12 answers right. From a 97.6% success rate down to 2.4% in just three months.

It's rumored OpenAI is using several smaller and specialized GPT-4 models that act similarly to the large model but aren't nearly as expensive to run. You can find the full study here.

However, there's a solid counter-argument. The folks at AI Snake Oil say the seeming decline may be a result of how the researchers tested the AI, not the AI itself.

They explain AI models like GPT-4 have two aspects: capability and behavior.

Capability is what the model can do, and it doesn't change much. Behavior, on the other hand, can change a lot based on fine-tuning, which is done regularly. In short, the new paper doesn’t show that GPT-4 capabilities have degraded. But it is a valuable reminder that the kind of fine tuning that LLMs undergo can have unintended effects, including drastic behavior changes on some tasks.

Full Rebuttal

How Anyone Can Build Their Own AI Agents (Fast).

You don't need to be a programmer or tech whiz to build your own customized AI agents. AI Marketers ambassador Alex Stewart reveals how easy it easy for you to create:

• Copy agents that follow your tone of voice...
• Content agents that repurpose your videos into short and long form posts/blogs/emails...
•YouTube script analyzer agents that can break down videos into repeatable frameworks...
•Accountability agents that can check-in on your clients/customers/team members...

In his post, Alex breaks down the process of building one of these powerful agents by combining three simple components - APIs, a universal programming language, and low-code tools. On the surface this might sound intimidating. But the truth is, it's so straightforward anyone can do it. No coding or tech background required.

The only limit is your imagination. Whether you're a business owner, marketer, or a curious mind, you're going to want to check this out. Even to just familiarize yourself with these remarkable technologies essential to the future of work.

Full Post

Prompt of the Week

An AI prompt for marketers is a text input designed to guide an AI like ChatGPT in generating creative marketing ideas and copy. It helps you leverage AI to enhance your campaigns and decision-making processes.

One of the most effective market research tactics is reading reviews on websites like Amazon. They're direct, honest feedback from customers. They tell you what works and what doesn't. Reviews give you real-world insights into how your market speaks and their biggest problems and desired solutions.

This weeks prompt helps you analyze as many reviews as you can fit into ChatGpt, and shows the most common responses from various categories. In other words, it allows you to almost instantly uncover your markets most important thoughts, experiences, and emotions.

Here it is:

💡
Make a list of most common things that customers like or don't like about the product from the reviews pasted below. You should only include things that repeat over and over again in the reviews, write the ones that repeat the most first.

Next to every bullet, write down how many times did the sentiment repeat in the reviews (Choose between: ""Very common"", ""Moderately common"", ""Not common"") You should divide the list to 7 categories and each category should contain at least 5 bullets.

Use Markdown to format the content nicely. Important: For every heading, choose an appropriate emoji and paste it next to the heading.

The categories should be:

1. Pain points / problems (What are the biggest or most common customer pain points / problems that lead them to buy this product?)

2. Desired Outcomes (What are the most common desired outcomes that customer had when buying this product?)

3. Purchase prompts (What event triggered them to start considering purchasing this product?)

4. Main Unique Value propositions (What do you think are the main things that set this product apart from others (benefit/advantage your would put into headlines))

5. Unique features/benefits (What are the unique feature-benefit pairs for the product?)

6. Uncertainty & Perceived risk (Why were/are customers doubting? What did they think about the product that scared them before buying?)

7. Objections (""Why did customers think this product wouldn't work for them before buying?) Important: For every heading, choose an appropriate emoji and paste it next to the heading. The reviews you should analyse are here: <INSERT REVIEWS>

*Protip: Use one of the numerous review scrapping tools and combine it with this prompt to take your market research to the next level.

*Prompt courtesy of FlowGPT user, Wilco De Kreij.

Unraveling AI

Subscriber Submitted Question: "I've heard of no-code tools before, but what are low-code tools?"

Our Answer: You know how building a Lego set is easier than carving a toy out of wood? That's because Lego gives you ready-made pieces and you just have to put them together based on a design. Low-code tools are like legos. Instead of writing all the instructions line by line, you get pre-made "blocks of code" that you drag, drop, and arrange to create your program - much like building a toy with Lego bricks.

Low-code tools make it simpler for people who aren't experienced coders to create their own applications. It's like having a Lego kit with an instruction manual; you don't have to be an expert builder or know every single Lego piece, you just need to follow the instructions and put the blocks together.

If you want to delve in further, here is fantastic beginners guide to low-code tools.

Tool to Try

Chord- Chord uses AI to curate relevant, high-quality content and discussions from around the web, then synthesizes what it learns into custom articles, written in 1-2 minutes. A valuable function, but that's not the reason we're featuring Chord.

When testing, we discovered its also a fantastic market research tool. One that can slash the time you spend wading through forums, googling, and reading blog posts.

Say you're a business owner or marketer selling a joint pain relief supplement for dogs. We typed "my dogs joint pain won't go away" into Chord and in a few minutes it spit out a pretty vanilla article. There were reasons why your dog might be having joint issues, common remedies, and popular brands of supplements.

But below the article, Chord lists all the most relevant sources the AI found.

This includes popular articles, forums and a complete treasure trove of other resources WITH links for you to dive deeper into. Here are just a few examples:

In the News

A Big Week For ChatGPT Competitors

There was a lot of exciting progress made in the world of chatbots this week. To start, Anthropic introduced Claude 2. One of the standout upgrade is being able to input up to 100K tokens in each prompt. For comparson, GPT-4 has a 32k limit.

This means that Claude can work over hundreds of pages of technical documentation or even a book. Claude can now also write longer documents - from memos to letters to stories up to a few thousand tokens - all in one go.

What You Need to Know about Claude 2

Source: Andre Havro, Resulta

Meta released a new set of large language models (LLMs), collectively called Llama 2. And it's completely open sourced. This means companies can build their own custom models on top of them for free. Meta says it’s partnered with Microsoft to make the Llama 2 models easily accessible from the Microsoft Azure cloud.

“Our models outperform open-source chat models on most benchmarks we tested,” the Llama 2 research paper abstract reads.

What You Need to Know about Llama 2

Apple is developing artificial intelligence tools according to a new report from Bloomberg. Engineers are internally referring to it as “Apple GPT." The tech giant has yet to determine a strategy for releasing the technology to consumers, but is aiming to make a significant AI-related announcement next year.

The news of the project alone added $71 billion to Apple's market value this week.

Finally, Elon Musk announced his new company xAI. He plans to compete with OpenAI by using Twitter data to train a superintelligence. But unlike the current top dog chatbot, xAI's goal is to create a"maximally curious" AI rather than explicitly programming morality into it.

AI Resurrection of Brazilian Singer for Volkswagen Ad Sparks Joy and Ethical Worries


The late Brazilian singer Elis Regina was resurrected through artificial intelligence for a Volkswagen commercial, performing a duet with her daughter, Maria Rita.

While fans of Regina rejoiced, the AI-enabled performance sparked a heated debate about the ad's morals.

In fact, Brazil's advertising watchdog, Conar, launched an investigation into a possible breach of ethics. Despite the controversy, Volkswagen defended its campaign, stating it had the blessing of Regina's family. Regina's son, João Marcello Bôscoli, praised the discussion triggered by his mother's reappearance and the emotional side of technology revealed by the ad. Since its airing, the commercial has been viewed over 16 million times on YouTube.

Full Post

Mind Fodder

What would a machine dream about after seeing the collection of The Museum of Modern Art? For Unsupervised, artist Refik Anadol uses artificial intelligence to interpret and transform more than 200 years of art at The Museum of Modern Art.

Known for his groundbreaking media works and public installations, Anadol has created digital artworks that unfold in real time, continuously generating new and otherworldly forms that envelop viewers in a large-scale installation

How Can We Make AI Marketers Your 'MUST READ' Newsletter?

When you signed up for this newsletter, we promised to deliver cutting-edge information and actionable insights into the new world of AI powered marketing. What did you think of today's newsletter? Please let us know what you liked and what you'd like to see more of.

Thanks for reading.

Until next time!

The AI Marketers

p.s. Need A-List copy delivered at unprecedented speed? Three of the world's top copywriters have joined forces and are using the power of GPT-4 to write high-converting ads, emails, and VSL's.

Discover how you and your business can leverage them at https://www.calabs.ai/ or email Stefan directly at stefan@calabs.ai.

p.p.s If you're not already a member of our free AI Marketers Facebook group, you can join our rapidly growing community here.