TSMC thinks it will take the rest of the industry 6 more years to achieve ¼ of the transistors that we already have today. Would you like to enjoy 4 trillion transistors today on CS-3 or 1-trillion transistors on a GPU in 2030? Read TSMC’s full report (
https://
spectrum.ieee.org/trillion-trans
istor-gpu
…)
AI HARDWARE
-
TSMC’s Transistor Advantage: Today’s 4 Trillion vs 2030 Projections
By
–
-
ARM and Untether AI Shape Computing Future for Everyone
By
–
The future of #computing. For everyone.https://t.co/zQDppuaZA0 https://t.co/0VIzDHtQl9
— Untether AI (@UntetherAI) 15 avril 2024The future of #computing. For everyone. https://
arm.com/partners/catal
og/untether-ai
… -
Untether AI joins Arm Partner Catalog for future development
By
–
We can now be found in the @Arm Partner Catalog! We’re excited to be working together to build the future #onArm. Learn more at https://
arm.com/partners/catal
og/untether-ai
…. -
Limitless Pendant LED Recording Consent Mode Privacy Features
By
–
This bit is interesting: > An LED lights up whenever it’s recording, and the Limitless Pendant also has a “Consent Mode” that detects new voices and doesn’t record them until the software hears them agree to being recorded. (It’s worth noting this mode is off by default.)
-
Google Glass Backlash: Surveillance Concerns and Tech Industry Reactions
By
–
Google Glass wearers were being attacked in San Francisco ten years ago – though that could partly have been a reaction against the tech industry generally, hard to know if it was directly due to surveillance concerns
-
Hacking AI Pin to Run Gemma 2B Locally Without Internet
By
–
Who is going to be the first to find a way to hack the ai pin and run like Gemma 2B locally on it and see how it works instead of needing an internet connection (or only periodically phoning a cv/tts model)?
-
Quantization Dramatically Compresses LLMs for Consumer Hardware
By
–
LLMs can take gigabytes of memory to store, which limits what can be run on consumer hardware. But quantization can dramatically compress models, making a wider selection of models available to developers. You can often reduce model size by 4x or more while maintaining reasonable… pic.twitter.com/ASQ28fzgkB
— Andrew Ng (@AndrewYNg) 15 avril 2024LLMs can take gigabytes of memory to store, which limits what can be run on consumer hardware. But quantization can dramatically compress models, making a wider selection of models available to developers. You can often reduce model size by 4x or more while maintaining reasonable
-
Adaptive Compute: Innovative Solution for Technical Challenges
By
–
Adaptive compute could be an interesting/possibly necessary way to solve this.
-
Hardware Success Requires Excellent Software Quality
By
–
The issue here is that we all know hardware is tough, but weirdly, that seems to be the best part of the pin. In this day and age, you have no excuse for bad software. Being torn down because of software is just bad, and totally preventable.
-
Europe launches semiconductor testing facilities for digital innovation
By
–
Europe commits to semiconductor innovation with new testing facilities | Shaping Europe’s digital future https://
digital-strategy.ec.europa.eu/en/news/europe
-commits-semiconductor-innovation-new-testing-facilities
… via @EU_Commission #digitaleu #semiconductor #Innovation #technology #europe #digital @DigitalEU @ArturHabant @elaniaz @BetaMoroney @PawlowskiMario