AI Weekly Digest #4: Private AI, Impact on Jobs, and Regulations

Welcome to this new edition of our AI newsletter, where we bring you the latest updates on artificial intelligence and its impact on our world. No BS!

So much to cover! In this issue, we'll explore:

• The Main Headlines: Is AI going too fast? Elon Musk thinks so! Dolly & GPT4All models, and Microsoft announces Security Copilot.

• Hot Topics: Sam Altman’s (OpenAI CEO) fascinating podcast with Lex. GoldmanSachs report on AI’s impact on jobs, AI vs Mechanical Turk.

• Beyond the Hype, the story: The Age of AGI has Begun

Beyond the Hype, The Story

The Age of AGI has Begin

GPT4 announces the start of the artificial general intelligence (AGI) era: 155 fascinating pages evaluating GPT4 reasoning abilities that conclude the analysis by “Given the breadth and depth of GPT-4's capabilities, we believe that it could reasonably be viewed as an early (yet still incomplete) version of an AGI system”. One of the evaluations show that GPT4 could be hired as a software engineer at Amazon! (summary in the video below).

Main Headlines

• Is AI Going Too Fast?? Y. Bengio, Elon Musk, Steve Wozniak,& over 1,000 others, have called for a temporary pause on training systems exceeding GPT-4. "AI systems with human-competitive intelligence can pose profound risks to society and humanity". It is very unlikely to see OpenAI & Micrososft slow down, but it does raise concerns and some countries are starting to halt the use of ChatGPT (e.g., read about Italian regulators Banning ChatGPT).use-giant-ai-experiments/

• 30min to build your own Private ChatGPT? Since the release of Stanford’s Alpaca model (see our previous newsletter), the research community has been exploring cheap and fast ways to build your own chatGPT. With several breakthroughs! In a nutshell:

  • Dolly: Databricks announced Dolly showing that a “small” GPT-J 6B parameters model trained for 30min on Alpaca's 50k dataset, leads to a competitive instruct like model (e.g., chatGPT).

  • GPT4All: a new model released taking the Alpaca model to the next level by training LLaMA on 800k instructions generated by ChatGPT! It is also packaged in a easy way to be deployed locally on your machine!

• Microsoft announces Security Copilot: a new GPT-4 AI assistant for cybersecurity. You’ll be able to feed the AI files, Urls or snippets of code as ask information about their security and vulnerability. In the demo, we even see it reverse engineer a malicious PowerShell script and explained how it worked step by step!

Hot Topics!

AI continues to raise questions about its impact, short-term on jobs, and long-term on humanity in general

• Sam Altman (OpenAI CEO) shared his thoughts about the future of AI (podcast): You can see the full video here, or get the summary in the twitter thread below. “I empathize with people who are a lot afraid [of AI]”.

• Goldman Sachs report on AI’s impact on jobs: In a recent report Goldman Sachs predicts that 300 million jobs could be lost or degraded by artificial intelligence (AI) in the next decade. It also predicts that the global economy will benefit from AI with a 7% GDP increase fueled by AI productivity ; while sharing concerns regarding risk of displacing many workers in sectors like office administration, legal, healthcare and art and design.

• AI vs Mechanical Turk: Goldman Sachs’s report resonates with a new paper showing that ChatGPT outperform humans at labeling tasks (usually relying on mechanical turk kind of services). Many companies rely on human labor to check invoices, extract data, label documents, etc. Today, AI can do enable fast, around the clock & cheap labor.

Bonus!

Here’s a great learning resource to start your journey with NLP and Deeplearning technologies.
Enjoy!

That’s it for today! If you made it this far, I’d appreciate a quick feedback 😋! I know there’s room for improvement! So don’t hesitate to share with me the things you liked and those that you didn’t.

Have a great Sunday and may AI always be on your side!