4 Min Read • Intensity: ☕️☕️

Good morning, Caffeinators☀️

Hope you have a soaring Friday as we dive into United Arab Emirates’ Falcon LLM!

Today’s Starting Sips:

• 🚨 Are Our Jobs Under AI’s Radar?

• 🦅 Falcon Soars High in LLM Skies

• 🍎 Apple’s Review on GPT

• 👐 How to Train Your Falcon

🚨 Are Our Jobs Under AI's Radar?

The "AI takes our jobs" theory isn't sci-fi anymore. Data from Challenger, Gray & Christmas reveals nearly 4,000 US workers lost their jobs last month because of AI.

Get ready, 'cause the numbers get more chilling!

Over 80,000 layoffs were announced in May by US employers. Yes, you read that right - an alarming 20% increase from last month and a 287% jump from a year ago. Out of these, 5% of job cuts were AI's doing.

Here's the million-dollar question: Who's running faster? The job losses or the AI industry growth?

Analysts seem optimistic, forecasting job growth and economic escalation within the emerging AI industry. So, is it an AI-induced job apocalypse or a dawn of new opportunities? Stay tuned!

More on job apocalypse? Click here!

🦅 Falcon Soars High in LLM Skies

Abu Dhabi's Technology Innovation Institute has unleashed the Falcon models that shook the LLM landscape.

The heavyweight champion, Falcon-40B, has dethroned Meta’s LLaMa-65B. On the other side of the weight scale, Falcon-7B reigns supreme in the lightweight division.

These robust models have been trained on a staggering 1 and 1.5 trillion tokens! For perspective, GPT-3 only had 500 billion. A third of what Falcon has.

Where do they get the funds from? UAE government * cough * oil money 💸

This breakthrough demonstrates two critical insights. Firstly, the race to build the most potent LLM is far from over. Though smaller parameters, Falcon stands out because of its high-quality and vast amount of training data. With more parameters without sacrificing training quality, we can unlock even more potential.

Secondly, the UAE's investment in Falcon isn't about generating immediate revenue. It’s a strategic move to attract AI talent worldwide. They've made Falcon open-source, essentially rolling out the red carpet for AI engineers worldwide. Interesting strategy, wouldn’t you agree?

To learn how to download and use Falcon, checkout Learning of the Day ⬇️

🍎 Apple’s Review on GPT

Tim Cook, Apple's CEO, just dropped a bombshell - he's a fan of ChatGPT and he's excited about its possibilities. Now, that's something!

He's not just playing around with the tool, but the company is 'looking at it closely'. Hints at possible integration into Apple products?

Though AI already dwells within Apple's shiny gadgets, we can all agree that Siri needs an upgrade 🤭

But hold on! It's not all rainbows and unicorns. Just like all AI specialists, Cook points out the darker side: AI bias, misinformation, and possibly worse.

This revelation surely has stirred up the tech world, eager to see how Apple embraces or regulates AI. Coming on the heels of similar cautionary statements from tech leaders, Cook's comments highlight AI's rapidly evolving nature. It's a thrilling tale that keeps unfolding!

Learning of the Day: 👐 How to Train Your Falcon

Feeling adventurous? Let's play with a simpler yet jazzy version of Falcon-40B - the Falcon-7B.

But first, we gotta get the essentials. Install these buddies if you haven't already:

!pip install transformers
!pip install einops
!pip install accelerate
!pip install xformers

With the tech gear on, we're ready for the Falcon 7-B Instruct. Here's your magic spell:

from transformers import AutoTokenizer
import transformers
import torch

model = "tiiuae/falcon-7b-instruct"
tokenizer = AutoTokenizer.from_pretrained(model)
pipeline = transformers.pipeline(
    "text-generation",
    model=model,
    tokenizer=tokenizer,
    torch_dtype=torch.bfloat16,
    trust_remote_code=True,
    device_map="auto",
)

prompt = "Give me one fun fact about coffee"
sequences = pipeline(
    prompt,
    max_length=200,
    do_sample=True,
    top_k=10,
    num_return_sequences=1,
    eos_token_id=tokenizer.eos_token_id,
)

for seq in sequences:
    print(f"Result: {seq['generated_text']}")

The code downloads Falcon-7B via HuggingFace transformer pipeline. Then, we give the model an instruction, “Give me a fun fact about coffee”.

If all goes well, Falcon responds something like:

And behold! Royalty FREE advice from UAE highness: drink more coffee ☕️

To-Go Cup 🥤

• 🤖 AI job apocalypse is happening today. 5% of unemployment from last month was because of so

• 🐣 UAE dropped the number one open source LLMs: Falcon 40B and 7B

• 🍏 Tim Cook likes ChatGPT but is worried about our current AI regulations

• 🪢 Learn how to train your own Falcon today

Closing Remarks

Here at CaffeinatedGPT, our mission is to sound the alarm on AI safety. AI threats happen more often than you think.

We're not fretting about some dystopian future where robots rule… at least not yet. Instead, we're focused on the here-and-now threats AI presents:

• Spread of misinformation

• Awareness of job displacement

• Lack of data privacy and transparency

We invite you to join us on this journey, and together, we can make the digital world a safer space. So, next time you sip your morning coffee, remember, every byte of awareness helps!

If you enjoyed this article, I think you will like this one too! It is about calling GPT API.

spoiler alert: It is a lot cheaper and you only need 6 lines of code.

Until next time,

Your friendly barista and ☕️ GPT!