24th issue: You can fine-tune your own ChatGPT now 🤯

🌯 Wrap for the week

A close look at all the main headlines in AI

Top news for the week

Fine-tuning has arrived for GPT-3.5 💸 

This feature has been on the waiting list for many AI enthusiasts and finally, it has arrived this week!

Here are the breakdowns of the recent news:

1. Customize Your AI Experience: Users can now bring their own data to tweak the model, enhancing its reliability and specificity. Whether you're looking for a model that always replies in French or one that cracks a joke in a brand-specific tone, customization is now your playground. And guess what? For those who wished their AI sounded a tad more unique, OpenAI would grant your wish now.

2. Economics of Fine-Tuning: Here's the deal – fine-tuning allows users to shrink their text prompts, which means faster API calls and, most importantly, saving some bucks. OpenAI even dished out some numbers: training at $0.008/1K tokens, input usage at $0.012/1K, and output at $0.016/1K. A 100,000-token meal (or training file), roughly 75,000 words, would set you back just about $2.40.

3. What’s Next & Old News: OpenAI didn't just stop at GPT-3.5. They've also released two upgraded GPT-3 base models (babbage-002 and davinci-002) with more bells and whistles. But, mark your calendars! The original GPT-3 models are taking a bow and exiting stage left on January 4, 2024. And for those already eyeing GPT-4's capabilities (yes, the one that ogles at pictures too), stay tuned. Fine-tuning for it is set to drop this fall.

If you are keen to explore further about fine-tuning and understand how it works, it is worth checking this blog from OpenAI: https://openai.com/blog/gpt-3-5-turbo-fine-tuning-and-api-updates

On top of that, if you are an enterprise user, you can explore OpenAI’s official partner for fine-tuning Scale, check the detail from OpenAI’s blog: https://openai.com/blog/openai-partners-with-scale-to-provide-support-for-enterprises-fine-tuning-models

Main news credit:
OpenAI brings fine-tuning to GPT-3.5 Turbo from TechCrunch

Other top AI headlines for the week

Meta introduces SeamlessM4T, an AI model that translates and transcribes nearly 100 languages from The Tech Portal

If you have also watched The Hitchhikers Guide to the Galaxy before, I believe the Babelfish is becoming a reality soon. Also, if you want to understand further about this new model and try the demo, check this blog from Meta: https://about.fb.com/news/2023/08/seamlessm4t-ai-translation-model/

AI2 drops biggest open dataset yet for training language models from TechCrunch

Good news for the people thinking about training new LLMs!

How Artificial Intelligence Gave a Paralyzed Woman Her Voice Back from the University of California San Francisco

This is the type of advancement I think AI can be so great for. More news like this, please!

Google, Amazon, Nvidia and other tech giants invest in AI startup Hugging Face, sending its valuation to $4.5 billion from CNBC

Seeing the amazing growth of Hugging Face, it is no wonder how these tech giants would be interested in joining this journey.

🎬️ Guess that movie

Adding a bit of fun to your weekend reading, let’s guess the famous movie from the 10-emoji description created by ChatGPT:

🚗💨🇯🇵🌃👦🏻🏁🔄💥🚔❤️

If it is a bit hard to figure out, you can find the answer with some explanation from ChatGPT at the end of this newsletter.

🧰 Toolbox

Check out this week’s tools and resources that give you more superpowers with prompt crafting!

1. 🔧 DemoGPT
DemoGPT allows you to create an app with just some prompts which makes prototyping different apps much faster. It leverages the LangChain framework and you can check here from LangChain regarding this tool: https://blog.langchain.dev/langchain-demogpt-new-era-for-gen-ai-applications/

2. 📺️ Finetuning Large Language Models
Since fine-tuning is all the rage nowadays, it would be beneficial to pick up this course from DeepLearning to learn more about the concept thoroughly. Happy studying!

3. 📚️ How ChatGPT turned generative AI into an “anything tool”
This is a great article from Ars Technica that covers the basic of AI that led AI to develop from a tool for experts in special fields to one for the average Joe and Jane.

👋 Your next XTwitter friend

Connect with great people in the industry to supercharge your growth

🖼️ Image of the week

I have seen a lot of the following images lately where the main picture is shaped using the other items on the image. I will explore in a future issue regarding how this could be done via Midjourney.

🎬️ Guess that movie - answer

The answer is "The Fast and the Furious: Tokyo Drift": 🥳 

Did you guess that right? Here is ChatGPT’s explanation:

🚗💨🇯🇵🌃👦🏻🏁🔄💥🚔❤️

This summary captures the main elements of the story: cars and racing (🚗💨), the Japanese setting (🇯🇵), Tokyo nightlife (🌃), the main protagonist (👦🏻), race challenges (🏁), drifting technique (🔄), crashes and conflicts (💥), police or trouble with the law (🚔), and underlying love interests or bonds (❤️).