Friday Funtime

Hi this is Brainyacts, it's like IKEA instructions for Generative AI, but with less 'where does this screw go?' and more 'wow, this actually makes sense!

Let’s keep smilin’.

A special welcome 👋 to my NEW SUBSCRIBERS! 

To read previous posts, go here.

In this edition we will

1. get savvy on Google’s updated LLM

2. learn how Claude just exploded its capacity!

3. learn how to use multiple LLMs

4. learn how over $100k was raised via ChatGPT biz coaching

5. meet TrumpGPT (yup, that’s right)

6. AI goes Absolut(ely) local

Why You Should Definitely Read This‼️

Understanding Google's latest AI language model, PaLM 2, and Anthopic’s update to Claude is crucial for professionals across the global. This is not a technical discussion. It is about explaining how fast this space is moving which means you will be able to do so much more than you are right now. Both of these announcement are significant.

These developments will 10x to 100x the utility of how we use and interact with generative AI - so please dig in. I tried to make it easy reading.

2 LLMs Get Major Updates  

Google’s Latest LLM

Hey there! Let's chat about this new AI language model Google's whipped up called Pathways Language Model 2, or PaLM 2 for short. Think of PaLM 2 as the brain behind Google's AI chatbot Bard, just like OpenAI's GPT-3.5 is the mastermind behind ChatGPT.

Now, PaLM 2 is the new and improved version of Google's first language model, PaLM. While PaLM was a bit of a one-trick pony, only understanding English, PaLM 2 is a globe-trotter! It's been trained on text from over a hundred languages, so it's got a pretty good grasp on idioms, riddles, and even poetry in all these languages. Plus, it's a bit of a code whiz, able to write in Python, JavaScript, Prolog, and Fortran, and it can even solve mathematical equations and make sense of images.

What's fun is that PaLM 2 comes in all shapes and sizes, depending on what you need it for. These different versions are named after different animals - Gecko, Otter, Bison, and Unicorn. Google tells us that the smallest of these, Gecko, is a little speedster, able to work quickly even on mobile devices and offline.

So, how does PaLM 2 stack up against OpenAI's GPT-4? Well, both are pretty impressive, boasting multilingual capabilities and the ability to understand images. They've also both been trained on a ton of data, though neither Google nor OpenAI are giving us the exact numbers. But Google's PaLM 2 has a bit of a leg up when it comes to logic and reasoning, and it even has special versions for healthcare (Med-PaLM 2) and cybersecurity tasks (Sec-PaLM 2).

Another key area the company plans to upgrade Bard with PaLM 2 is the "Workspace AI collaborator." PaLM 2 will feature in the company's full product range—in Cloud, Colab, Meet, Docs, Maps, Sheets, and others.

As of now, Google's keeping Bard free to use, while OpenAI's got a paid version of ChatGPT.

Of course, PaLM 2 isn't perfect.

The smaller versions, like Gecko, while portable, don't perform as well as their bigger siblings. And sometimes, these models can spit out information that's made up - a bit like daydreaming or "hallucinating". Plus, Google hasn’t really solved the whole transparency on the training data issue as whether and to what extent PaLM 2 includes copyrighted or pirated text.

But all in all, PaLM 2 is a pretty cool leap forward in the world of AI language models!

Now onto Anthropic’s Claude.

I’m truly excited to bring you an update about Claude, an AI model that's making strides in the realm of natural language processing. Their context window has been expanded from 9,000 to a whopping 100,000 tokens, roughly equivalent to 75,000 words. This means that you could potentially have conversations with Claude that span hours or even days.

What is a "context window?"

For large language models (LLMs) like GPT-4 the context window refers to the maximum number of previous words or tokens (units of text) the model can consider when generating a response or prediction.

To put it in simpler terms, imagine having a conversation where you only remember the last 10 sentences said. That's your "context window" – the amount of past conversations you're keeping in mind to understand and respond to the current sentence.

The context window is important for a few reasons:

1. Understanding Context: If the model's context window is too small, it might not fully understand the context of the discussion. For instance, if you're discussing a complex topic and ask a question related to it, the model might not remember the details if they were mentioned too long ago. This could lead to answers that are out of context or not as helpful.

2. Coherent and Relevant Responses: By remembering recent conversations, the model can generate more relevant and coherent responses. This is crucial for maintaining a sensible and meaningful conversation.

3. Resource Constraints: There's a balance to be struck, however. Larger context windows require more computational resources, making the model slower and more expensive to run. So it's a trade-off between performance and resources.

So Claude’s massive enhancement means it can now analyze large volumes of text, like multiple business documents or even entire books, in mere seconds. To give you a sense of scale, an average person would need approximately 5 hours to read 100,000 tokens, and even more time to fully digest and analyze the information. Claude, on the other hand, can complete this task in less than a minute.

A notable demonstration of this was when the entire text of 'The Great Gatsby' was fed into Claude, with a single line altered. Claude was able to identify the difference correctly in just 22 seconds.

Claude's capacity to retrieve and synthesize information from extensive documents, following the user's instructions like a human assistant, is a major step forward.

I love this demo on their site as I have used this previously too and struggled with a limited context window in OpenAI’s products.

Here it reads an entire 10k. These are typically 100-pages or more of financial and strategy information. The US SEC requires all publicly listed companies to file these. They contain a lot of awesome info but they are challenging to consume.

You have to request access to Claude. And you should. But in the meantime you can always use Poe.

USE CASE: Accessing Multiple LLMs

If you're familiar with using ChatGPT, then you'll be intrigued by Quora's AI tool, Poe. Just like ChatGPT, Poe allows you to interact with AI, but it comes with a twist – a variety of chatbots with their own specialties.

I’ve covered Poe before when it launched but it has only gotten better and for the same $20/mo you might pay to access OpenAI’s GPT-4, you could get access to that and a handful of other LLMs, including the newly enhanced Claude. Not a bad deal.

Subscribers are guaranteed at least 600 GPT-4 (about 20 a day) and 1000 Claude+ messages per month at normal speeds. After message limits are reached, at times when Poe is over capacity, messaging speed, quality, or bot availability may be reduced.

NOTE: If you are an OpenAI paid user, you get 25 GPT-4 messages every 3 hours. So if you are a power user of GPT-4, OpenAI is the way to go. Unless you geek out on me - then you have Poe, OpenAI, and others. 💸 😎

Here's everything you need to know about Poe:

Understanding Quora Poe:

1. Quora Poe is an AI chatbot service offering a collection of chatbots developed by different AI companies.

2. Just like you've been using ChatGPT, you'll also find it here, along with its advanced version, the GPT-4 language model.

3. Additionally, Poe introduces you to new chatbots like Claude+, developed by Anthropic, an American AI startup.

Differentiating Quora Sage and Other Bots:

1. Each bot in Poe's lineup has its own unique strengths, similar to how ChatGPT specializes in certain tasks.

2. For instance, Sage is Quora's own general knowledge bot. It's great at answering general questions and assisting with simple coding tasks. However, like ChatGPT, its knowledge is limited to information up until 2021.

How to Use Quora Poe:

1. To start using Quora Poe, visit the Poe website and sign up or log in with your existing account.

2. 🚨 If you're an iOS user, you can also download the Poe app from the App Store. (Android users, hang tight! Your app is on its way later this year.)

3. Once you're in, select the chatbot you want to interact with and type out your query. You'll receive a response fairly quickly.

Getting More from Quora Poe:

1. There is a free version and you can interact with all the chatbots, but the service has some limitations. If you want to unlock more messages or get more access to the advanced chatbots, consider getting a Poe subscription.

2. The subscription costs $20 per month or $200 for a year, and you can subscribe via the iOS app or the official website.

In a nutshell, Quora's Poe is a step forward from ChatGPT, providing you with a wider range of AI personalities and capabilities.

News you can Use: 

ChatGPT As a Business Coach/CEO

Entrepreneur João Ferrão dos Santos used ChatGPT to create a successful online business, AIsthetic Apparel, which raised over $100K in investment and generated over $10K in sales revenue within a few days of launch.

ChatGPT suggested a t-shirt business as it required minimal upfront investment, had a broad potential customer base, and could be easily automated. Using an initial capital of $1,000, ChatGPT created a business plan for an online t-shirt printing-on-demand operation. The AI-generated designs were then printed on high-quality shirts and shipped directly to customers. João's LinkedIn post about the project went viral, contributing significantly to initial sales.

Key Takeaways:

• AIsthetic Apparel generated over $10K in sales revenue within a few days of launch and raised over $100K in investment.

• João plans to continue growing the business, leveraging AI technologies for unique and appealing designs.

• João's success has attracted global attention, and he is considering expanding the company's reach in the United States.

• He will be sharing insights on how AI can revolutionize the fashion industry at a startup and investor event in New York, and will also be featured in a CBS News documentary about Artificial Intelligence.

News you can Lose: 

TrumpGPT

Ever wondered what a chatbot Trump might sound like? No, me either, but someone did. You've heard of Deepfakes, now get ready for... Textfakes!

In a humorous (and slightly eerie) experiment, researchers have used our trusty AI friend, Chat GPT, to replicate the unmistakable rhetoric of Donald Trump. The results? Spot-on! The AI version of Trump not only echoes the ex-president's famous linguistic flourishes but also proves to be more coherent than the man himself.

Now, here's a quick rundown of this AI experiment:

• Researchers fed ChatGPT with questions asked to Trump during a CNN town hall.

• The AI responded in a way that mirrored Trump's style, hitting similar points, making similar pivots, and reflecting similar political calculations.

• The AI version of Trump was able to articulate his typical responses to controversial topics, such as his role on January 6, allegations of the 2020 election being rigged, and his stance on issues like the Second Amendment and abortion.

• In a side-by-side comparison of real Trump answers and those generated by ChatGPT, the AI convincingly reproduced Trump's manner of speech and thought patterns.

• However, it was highlighted that ChatGPT, despite mimicking Trump, doesn't inherently possess any political beliefs. It's merely reflecting the style and substance of the data it was trained on.

So, kick back, sip your drink, and enjoy this intriguing dive into the world of AI, where even the most distinctive personalities can be convincingly replicated in text form.

Weekend Inspired: 

You don’t need to drink alcohol to appreciate this. Interesting idea and ad to attach your brand to the AI hype. And listen to that music, it definitely has a “Stranger Things” vibe to it. Sorta fits.

DISCLAIMER: None of this is legal advice. This newsletter is strictly educational and is not legal advice or a solicitation to buy or sell any assets or to make any legal decisions. Please /be careful and do your own research.