AI Across Industries: Reviving the Beatles, Regulating the Future, and Redefining Chip Market

What's up ya'll, this is AIdeations. The go-to newsletter that takes AI and tech news that slaps and turns it into a no-bs, fun email for you each day.

Here's what we've got in store for you today:

 🎵 The Beatles Are Back

♟️ Europe Vs. Silicon Valley

💾 AMD Vs Nvidia

👾 OpenAI Pushes New Developer Updates

📰 News From The Front Lines

📚 Research Of The Day

🎥 Video Of The Day

🛠 Tools Of The Day

🤌 Prompt Of The Day

🐥 Tweet Of The Day

The Beatles Are Back! How AI Is Giving Us A New Beatles Track

Brace yourselves because a new Beatles recording is about to hit your playlist, and it's got "Artificial Intelligence" written all over it. Sir Paul McCartney, or Macca for those die-hard fans that are in the know, has unveiled plans to release a decades-old song featuring John Lennon's voice, extricated from a low-fi demo thanks to some pretty snazzy AI technology. The tune, likely christened "Now and Then," was among several songs Lennon recorded for his old pal McCartney just before his untimely exit.

The tech wizardry behind this musical resurrection is none other than Emile de la Rey, the mastermind who separated the Beatles' voices from background noises in Peter Jackson's epic, "Get Back." Using similar tricks, McCartney's team managed to isolate Lennon's voice from a rickety old cassette recording. Imagine that - an AI creating a posthumous duet between Lennon and McCartney. I bet even Lennon, in his New York apartment, couldn't have imagined such a future.

However, not all is rosy in this AI-meets-Beatles tale. "Now and Then" was previously shot down during the Beatles' Anthology series compilation in 1995, thanks to George Harrison calling it "f-ing rubbish." Yet, the intrigue around AI in music continues to grow. McCartney calls it "scary but exciting," and honestly, I couldn't agree more. With AI even generating new albums for disbanded groups (looking at you, Oasis), who knows where we're headed? So, here's to the future, folks. It's as unpredictable as a Beatle's hairstyle. Keep those vinyl spinning and the debate rolling!

AI Showdown: Europe vs Silicon Valley – Who'll Regulate First?

The EU and Silicon Valley are locking horns in a high-stakes boxing match over AI regulation. The EU lawmakers just adopted their stance on the EU AI Act. It's aiming to ban "unacceptable risk" systems (like predictive policing tools) and to limit "high-risk AI," which could sway elections or harm health. The Act also wants content created by AI systems to be labeled and mandates companies to share summaries of copyrighted data used for training. This could be a massive hurdle for systems using text scraped from the internet.

While the EU is adding another layer to their tech regulation lasagna, the US is still at the appetizer stage. Senate Majority Leader Charles E. Schumer (D-N.Y.) suggests that US lawmakers might only start considering their own AI legislation in the fall. Meanwhile, EU’s existing data privacy and tech competition laws are already stirring the pot. Google's chatbot Bard had its EU launch postponed due to privacy assessments, and Italy put a temporary pause on ChatGPT over data privacy concerns.

This jostle for control is setting Europe up as the world's premier tech referee. Its regulations could impact consumers globally, as companies shift their practices to avoid dealing with a patchwork of different policies. Case in point, Microsoft extending the rights central to EU's General Data Protection Regulation to all consumers globally.

But the US isn't just sitting on the bleachers. Schumer recently initiated a series of private AI briefings for lawmakers, showing their intent to stay in the AI game. However, there's a sense that the US is lagging in the tech regulations race. Democratic lawmakers stress the need to take the lead globally, while Republicans emphasize staying ahead in developing new AI tools.

The takeaway? The AI regulation race isn't just a power play between the US and EU—it affects us all. It's about shaping the future we want to live in. Will the EU remain the global tech regulation champ? Will the US step up their game? And what will this mean for our AI-powered future? It's going to be a wild ride, and we'll be here, keeping you updated every step of the way. In the world of AI, the only predictable thing is unpredictability.

A Fresh Contender in AI Chips: AMD Challenges Nvidia with the MI300X

AMD, a heavy-hitter in the chip-making industry, just unveiled their latest AI-oriented GPU, the MI300X. A bold move that has industry watchers intrigued. The firm, known for traditional computer processors, could potentially disrupt Nvidia's party, the current industry leader controlling over 80% of the AI chip market. But if AMD's AI chips win hearts and minds, we could witness a shake-up.

AMD CEO Lisa Su is optimistic, to say the least. She brands the AI venture as their "largest and most strategic long-term growth opportunity." And with the data center AI accelerator market projected to grow from a substantial $30 billion this year to a mind-boggling $150 billion in 2027, AMD could be on to a winner. Price-wise, we're in the dark, but if AMD manages to undercut Nvidia's GPUs, we might see a consumer shift towards more affordable AI applications.

Now, for the tech specs. The MI300X handles up to a whopping 192GB of memory. When running large AI models with an increasing number of calculations, every gigabyte counts. In the memory department, Nvidia's H100, their flagship chip, only musters up to 120GB, putting AMD in a promising position.

So, it seems like AMD's pushing back hard. But they still need to contend with Nvidia's comprehensive software package, CUDA, which has proven popular with AI developers. AMD's counter-punch is their own software, ROCm, set to engage with the broader ecosystem of models, libraries, frameworks, and tools. As Victor Peng, AMD's president, notes, the firm's "made really great progress in building a powerful software stack." The stage is set for a clash of the titans in the AI chip market, and it's anyone's guess who'll take home the crown.

OpenAI Pushes New Developer Updates

OpenAI has updated its AI model, GPT-3.5 Turbo, to be more controllable and to understand and simulate function calls, which makes integrating with other software or services easier. The cost to use this model has also been reduced by 25%.

They've also introduced a new model that can process four times more conversation context, useful for longer interactions. It does cost slightly more to use.

The older model will be replaced by the new version on June 27th, but users can continue to use the old version until September 13th if they need more time to adapt to the changes.

In simple terms, OpenAI's updates make the AI more powerful, versatile, and cost-effective for users.

 📰 News From The Front Lines: 📰

📚 RESEARCH 📚

Title: SpQR: A Sparse-Quantized Representation for Near-Lossless LLM Weight Compression

This paper talks about a new method for compressing the weights in large language models (LLMs) so that they take up less computer memory. It uses a method called Patched Frame of Reference (PFOR) compression, which basically helps to shrink the size of data, but it also adjusts the scale of the data because the data is floating point (it has decimal points).

How it works:

1. The weights of the model are reduced to 3 or 4 bits, in chunks of 8 to 32 scalars (basic data units). The minimum and maximum values in each chunk are used to normalize the values to specific ranges.

2. The arrays of minimum and maximum values are similarly reduced, but in larger chunks of 16. The offsets and scales for these arrays are 16-bit floats.

3. They store the values of "outliers" separately. Outliers are weights that are deemed important and can't be compressed like the rest. This is done using a technique that measures the sensitivity of weights.

The result is that the model performs nearly as well as unquantized models (those not compressed) while using less than 5 bits per weight. This compression technique may also speed up the model's performance, especially for smaller models.

Pros:

1. Reduces memory usage: The method can effectively reduce the memory usage of LLMs.

2. Nearly as good as unquantized models: The performance of these compressed models is close to that of uncompressed ones.

3. May speed up smaller models: The compression might also lead to faster operation, particularly for smaller models.

Cons:

1. More complexity: The technique adds more complexity to the system, such as the need to handle outliers separately.

2. May impact model performance: While the performance is close to unquantized models, there might be a slight dip in performance ("perplexity hit").

Implications/ Use Cases:

This work could be extremely useful for applications where memory usage is a critical factor, such as in mobile devices or in other constrained environments. It could also potentially speed up language model operations, leading to faster responses in applications that use such models. The authors also suggest that there's room for further improvement in this technique, which could lead to even more efficient compression methods in the future.

 📼 Video Of The Day 📼

🛠️ Tools Of The Day 🛠️

QuickQR Art - Scannable QR Art for your brand generated in seconds!

WebStudio - giving designers superpowers that were exclusive to developers in the past.

Coda - Coda brings all your words, data, and teamwork into one powerful doc.

Monarch - The modern way to manage your finances. Get clarity, confidence, and peace of mind for your finances. Track all of your account balances, transactions, and investments in one place.

Taiga - Your AI coding mentor accessible in slack

Review Writer- Automatically respond to reviews with AI

AtOnce - Make great content & answer emails 10x faster.

🤌 Prompt Of The Day 🤌

You are a direct response copywriting specialist. I'm going to give you some sample great subject lines below.

Sample subject lines:
This Is Your Opportunity
Thou Shall Not Steal
Too much email?
Tricks of the trade
This ends today
This is a game changer...
Steve Jobs' Storytelling Framework
Please confirm your subscription
Cut your data eng costs, improve cost insight, and S3's wild durability claims
Lesson #1: Legacy Writers vs. Digital Writers
The 56-yo multi-BILLION-dollar copywriting secret NOBODY talks about
A Message From Rachel
Important: confirm your subscription
More growth resources
Welcome to my DTC community
87,000+ IRS Agents are frothing at the mouth right now
Confirm your subscription to Dear Freelancer
Welcome to The Beautiful Mess
You've made an excellent choice
Thanks for subscribing to MKT1 Newsletter
Thank you for subscribing to 42
The Nag Metric
Missing info?
Welcome to Dear Freelancer
Welcome, content creator!
Will the Government Take Your Gold?
Become Selfless by Being Selfish
I think its time to fire that client
The BIG question
#1 thing that will make or break you as a new business owner
How to get REALLY good REALLY fast
I'll make a bet with you
[Lessons 3 & 4] - Good Client, Bad Client
Excited to have you here
Welcome to Unicorn Growth Strategies
How to earn 374 backlinks from 1 article
Money Words
My most profitable DR secrets, revealed
We have a mole amongst us
Should You Fear AI?
12 simple tips that give me the copywriting edge
7 Key Social Selling Tips, According to Experts
Things are about to change
Too much email?
How to do linear regression and correlation analysis
Millions made from 3 tests
Pros & Cons of AI Email in Sales [+Tools to Consider]
25 SEO Trends to Leverage in 2023
10 principles of creating thumb-stopping ads
How I gained 10,000+ subscribers in 2 months
Thou Shall Not Steal
America is Dead. Long Live America.
What jury duty taught me about product management
You're going to die
Personal invitation for you
How to become extraordinary
habibi come to dubai

I need a subject line for an email about [enter your email subject]

Please use the above subject lines to give me 30 suggestions. Some should be directly related to my email subject, and others should just be designed to get the email opened based on curiosity.

🐥 Tweet Of The Day 🐥

Thanks for tuning in to our daily newsletter. We hope you found our tips and strategies for AI tools helpful.

Your referrals mean the world to us. See you tomorrow!

Interested in Advertising on AIdeations?

 Fill out this survey and we will get back to you soon.

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