- Dev Notes
- Posts
- New Open Source Granite Models, Watsonx Capabilities, and IBM Concert
New Open Source Granite Models, Watsonx Capabilities, and IBM Concert
Good morning! IBM unveiled major AI advancements at their THINK conference, including open-sourcing powerful coding models and introducing new enterprise AI tools. Researchers demonstrated a relay attack that can steal Teslas despite their keyless entry system. Kotlin 2.0 launched with a unified compiler that boosts performance across platforms.
New Open Source Granite Models, Watsonx Capabilities, and IBM Concert
IBM went big at their annual THINK conference, giving their watsonx AI platform a serious upgrade.
The craziest release was open-sourcing their Granite family of language models specifically designed for coding tasks like:
Modernizing legacy applications
Generating new code
Documenting projects
These code-focused Granite models are no joke, packing 3 to 34 billion parameters and training on a whopping 116 programming languages. Even better, they outperformed much larger open-source models on industry coding benchmarks.
IBM also unveiled a cool new project with Red Hat called InstructLab. It's a novel approach that uses synthetic data to customize and continuously evolve foundation models through an open ecosystem.
With InstructLab, developers can finetune models for their specific needs, contribute improvements over time, and keep the AI evolving through community collaboration.
IBM also announced:
New code assistants for enterprise Java apps and aging Z mainframes
Upgraded data governance and automation capabilities
Access to hot third-party models like Meta's Llama 3
Tighter integration with cloud giants AWS, Microsoft, and Salesforce
Last but not least, IBM teased a new product called Concert - an AI-powered central hub. Concert will provide insights and recommendations across your entire enterprise landscape - applications, infrastructure, data repositories, you name it.
So it's clear IBM is going all-in on open AI, powerful language models, and making the technology more accessible to businesses. Their THINK announcements showed an impressive commitment to that vision.
Read More Here
Researchers Discovered That Teslas Are Easy To Steal Despite Adopting New Keyless Tech
So Tesla owners were pretty hyped when the company added ultra-wideband (UWB) radio tech for keyless entry on their newer models like the Model 3. The big selling point was that UWB can precisely measure distance, which should stop those annoying relay attacks where hackers can basically extend the signal range to unlock and steal your ride.
But here's the catch:
Researchers from this cybersecurity firm in Beijing called GoGoByte just showed they can still pull off a relay attack on the latest Model 3, even with UWB on board.
By relaying the Bluetooth signal from the owner's phone, they could unlock and fire up the Tesla from up to 15 feet away.
Now Tesla did respond saying: "This behavior is expected, as we are currently working on improving the reliability of UWB. UWB ranging will be enforced when reliability improvements are complete."
So in other words, they haven't actually implemented the UWB proximity checks for unlocking yet. The feature is there, but not being used for security. Until they push an update to legitimately use UWB ranging, their advice is to enable that PIN-to-drive option to add an extra security layer.
Read More Here
Kotlin 2.0 Now Available With Stable Version of K2 Compiler
Kotlin 2.0.0 has arrived, and it's a major level-up for the language. The highlight is definitely the new stable K2 compiler unifying all the platforms Kotlin supports while providing a huge performance boost. It also lays the foundation for faster language evolution going forward.
Some other awesome changes in 2.0.0:
Smart casting improvements to track types more accurately across scopes like try/catch blocks
Better separation of common and platform code in multiplatform projects to prevent weird cross-platform bugs
Support for compiler plugins like the experimental Power-assert for more useful test failure messages
On the JVM side, lambda generation now uses invokedynamic by default to reduce binary sizes. And for native developers, you can now monitor GC pauses in Xcode Instruments using signposts.
The JavaScript crew gets some love too:
Modern JS compilation targeting ES2015
Suspending functions compiled to generators
Per-file output for smaller bundle sizes
The Gradle plugin has a neat new DSL for configuring compiler options across targets. And the standard library stabilizes handy APIs like enumEntries() and AutoCloseable.
Read More Here
🔥 More Notes
Youtube Spotlight
COMPUTER SCIENCE explained in 17 Minutes
Was this forwarded to you? Sign Up Here