Good Morning! AI dominated Google's I/O event with new language models, developer tools, and AI chips unveiled. CPUs employ a hierarchy of small, fast caches and larger, slower ones to bridge the CPU-memory performance gap efficiently. The Ebury malware crew has infected nearly 400,000 Linux servers globally by late 2023, upgrading their malware to steal credentials and crypto wallets while staying hidden.

— Forrest Knight & Meghanadh Vasireddy

“AI” - Sundar Pichai’s Only Known Word

We’ve been releasing bits and parts about the event over the last week on the newsletter but here’s the full overview. Google's I/O event was all about AI. Let's start with the Gemini upgrades. Gemini 1.5 Pro can now handle up to 2 million tokens of context. That's an insane amount of information to process.

They also unveiled:

• Veo - For generating slick, high-quality videos

• Imagen 3 - Their best image generator yet for photorealistic creations

As for the developer toys, Google went all out:

• Gemini integration in Android Studio - Build that AI power directly into your Android apps

• Firebase Genkit - “Because making AI apps shouldn't be harder than using the tech itself”

• Project IDX - A browser-based coding environment turbocharged with AI smarts

Some other model updates to look at:

• Gemma 2 with 27 billion parameters

• LearnLM - Designed specifically for education and online learning

But the true flex was their 6th gen Tensor Processing Units, called "Trillium." These AI chips are promising a 4.7x performance boost over the previous version.

Now, Google knows all this AI power needs to be handled responsibly. So they've got upgrades to their SynthID watermarking to detect AI-generated content. They're also trying out new "AI-Assisted Red Teaming" for more intense security practices.

Read More Here

Why do CPUs have Multiple Cache Levels?

You know how CPUs are crazy fast but main memory is sluggish? Well, modern CPUs have multiple levels of caches to help bridge that performance gap.

Each CPU core has its own private L1 cache which is small but blazing fast. Then there's a slightly larger but slower private L2 cache per core. On top of that, there's a shared L3 cache that's even bigger but higher latency, acting as a holding area for stuff evicted from the L2s.

So why have multiple cache levels instead of just one big cache? It all comes down to different design goals and constraints:

• L1s prioritize minimal delays to avoid stalling the CPU

• Larger caches emphasize higher density and bandwidth

• Private caches let cores access them freely without dealing with other cores

• Shared caches require coordinating between cores

The caching algorithms also differ between the levels:

• L1s handle fine-grained accesses down to individual bytes

• Larger caches operate on full cache lines at a time

It's a balanced design meant to satisfy the conflicting demands of low latency, high bandwidth, large capacity, and reasonable area/power overhead. No single cache level can optimize for all of those, hence the hierarchical approach.

Read More Here

Ebury is alive but unseen: 400k Linux servers compromised for cryptocurrency theft and financial gain

The Ebury malware crew ESET exposed way back in 2014 is still infecting Linux servers. These people have managed to hack into almost 400,000 systems globally by late 2023.

But, it gets worse. They leveled up Ebury with some seriously malicious upgrades:

• New modules for redirecting web traffic

• Capabilities to steal credentials and crypto wallets

• You name it, they've added it

Their latest shady tactic is compromising hosting providers to rapidly deploy Ebury across thousands of servers in one fell swoop. Efficient but completely underhanded.

The new Ebury 1.8 version is a real piece of work too. It's using:

• Fancy obfuscation techniques

• Randomized domain generation algorithm

• A kernel-mode rootkit to stay hidden on infected boxes

But ESET’s report lays out all the details on Ebury's new attack methods. It also has IOCs and a nifty detection script to check if your systems are owned.

Read More Here

🔥 More Notes

• NetBSD bans use of Copilot-generated code

• Upstreaming Linux kernel support for the Snapdragon X Elite

• Google AR at I/O 2024: New Geospatial AR features and more

Youtube Spotlight

The Imitation Game

I promise I won’t do anymore movie recommendations but another free movie dropper on YouTube called the Imitation Game - a movie on the life of Alan Turing.

Was this forwarded to you? Sign Up Here