The Bonnie to my Claude

This is what we got today:

• 🤖 The Bonnie to my Claude

• 🐦 What is Sparrow?

• 🩺 Dr. ChatGPT

The Bonnie to my Claude

You're the yin to my yang.

The pea to my pod.

The Bonnie to my Claude?

Claude is a highly anticipated 'ChatGPT-like' AI assistant.

The AI was created by Anthropic, an AI startup co-founded by ex-staffers of OpenAI.

They have some street cred.

You're probably saying to yourself, "blah, another AI chatbot, why should I care?"

ChatGPT utilizes a process called Reinforcement Learning from Human Feedback (RLHF) to develop preference models.

Claude uses a different approach to develop preference models called Constitutional AI.

This type of AI uses a model (instead of human feedback) to generate outputs. The outputs are based on underlying principles, aka its 'constitution'.

You can check out an in-depth comparison between the two platforms here.

What is Sparrow?

The entire tech industry has been flipped upside down since the rapid rise of ChatGPT.

The chatbot has since been deemed the 'Google Killer'.

I personally liked 'Google Gunslinger' better, but that's beside the point.

So, what was Google's response to the situation?

They declared a 'Code Red'.

I am sorry don't care if you're getting tired of the Jimmy Fallon gifs.

Despite all the noise, Google's AI lab 'DeepMind' has been keeping quiet, however, reports are pointing to the release of 'Sparrow' very soon.

Sparrow is DeepMind's very own AI chatbot. It's already being labeled as the more 'grownup' version of ChatGPT.

What horse are you betting on?

If you're interested in learning more about Sparrow, you can find the whitepaper here.

Dr. ChatGPT

Yesterday, we talked about ChatGPT passing an MBA Exam at Wharton.

One day later and ChatGPT is back, this time passing the US Medical Licensing Exam.

What a nerd.

Dr. Victor Tseng and colleagues at Ansible Health put the AI tool to the test.

They found that ChatGPT could perform at >50% accuracy across all exams, and 60% in most of their analysis.

The passing threshold for the test is typically around 60%.

In another experiment performed by Vivek Natarajan and colleagues, the AI tool known as Flan-PaLM was also given the exam and passed with an accuracy of 67.6%.

I wouldn't be too comfortable with an AI performing open-heart surgery on me, however, if it means getting rid of the 3-hour wait at my doctor's office maybe I'd consider it.

Dr. Tseng and colleagues stated, "These results suggest that large language models may have the potential to assist with medical education, and potentially, clinical decision-making."

That's it for today. We'll see ya back here tomorrow!