Some people asked why we don’t have speech as the mechanism to communicate with LLMs. Latency. The answer is latency. 3s for a round trip answer and you get BORED waiting.
COMPUTING
-
UntetherAI Joins MLCommons to Accelerate ML Innovation
By
–
@UntetherAI has joined @MLCommons to accelerate #ML innovation! We are thrilled to collaborate and contribute to their renowned MLPerf benchmarks in our commitment to providing high-performance, energy-efficient #AI inference acceleration. Press Release: https://
tinyurl.com/4yswn2n7 -
Device Capacity Limits and Systemic Disruption Challenges
By
–
That's too much to put on this device. It sets up the disruption.
-
Real-time performance challenges in multi-thousand GPU clusters
By
–
Is anyone doing real time work on the multi-thousand GPU clusters that are popping up? It should be possible with the right software, but I generally see step times measured in seconds rather than milliseconds.
-
Technology Accessibility: How Hardware Costs Dropped Since 1989
By
–
If you can't afford the rich person's computer toys, stay alive. In 1989 I was so jealous of Apple cofounder Steve Wozniak who bought a $45,000 color printer to print from Photoshop 1.0. Today a $70 printer is way way better. The Israelis who are making the optics for the
-
Replit Deployments: Fast Node.js App Building Platform
By
–
Replit is one of the fastest places to build and deploy Node.js apps.
— Replit ⠕ (@Replit) 6 juin 2023
Try our Replit Deployments for your next Node.js project.
Deployments give you reliability, security, and scalability so you can focus on what you do best: building. pic.twitter.com/F0Kgr5hd9uReplit is one of the fastest places to build and deploy Node.js apps. Try our Replit Deployments for your next Node.js project. Deployments give you reliability, security, and scalability so you can focus on what you do best: building.
-
Apple Uses Transformers for Autocorrect, Positions Mac Studio for ML
By
–
There were two essentially throwaway lines from WWDC yesterday. First, it's using a transformers-based approach for autocorrect, of all things. And the second was it wanted to semi-position its new Mac Studio with a comically large level of memory for ML training workflows.
-
Meta’s MTIA AI Inference Accelerator: Key Problems Solved
By
–
3️⃣ Important problems that MTIA, our first-generation AI inference accelerator solves. A closer look from Mahesh Maddury, Roman Levenstein & Linda Cheng; part of the team behind this work.
— AI at Meta (@AIatMeta) 6 juin 2023
Details on Meta Training and Inference Accelerators ➡️ https://t.co/9A4I1iwNGp pic.twitter.com/aQmJ4wqwN4Important problems that MTIA, our first-generation AI inference accelerator solves. A closer look from Mahesh Maddury, Roman Levenstein & Linda Cheng; part of the team behind this work. Details on Meta Training and Inference Accelerators https://
bit.ly/42sAnv9 -
Quantum Technology Implementation Requires Significant Time Investment
By
–
Achieving successful implementation of quantum technology demands a significant investment of time from companies to foster the development of both skills and cutting-edge technological capabilities. Source @Capgemini Link https://
bit.ly/429ar7r via @antgrasso #QuantumComputing -
Cutting-edge AI solutions presented at Holst Centre Innovation Day
By
–
Save the date! We will be attending the
@HolstCentre Innovation Day! On June 21, Eindhoven, Netherlands. Our team will present our cutting-edge AI solutions that offer industry-leading performance and usability at a fraction of the cost of existing solutions. #ID23 #edgeAI