Wrapping up a productive time at @SEMIconex Taiwan 2024!
We want to thank our customers, partners, and everyone who visited us for connecting and engaging in insightful discussions about our innovative technology and cutting-edge solutions.
COMPUTING
-
SEMIconex Taiwan 2024: Showcasing Innovative AI Technology Solutions
By
–
-
New 70B LLM Outperforms Claude-3.5-Sonnet and GPT-4o with Innovative Fine-Tuning
By
–
> A new 70B open-source LLM beats Claude-3.5-Sonnet and GPT-4o! Matt Schumer, CEO from Hyperwrite AI, had an idea he wanted to try out: why not fine-tune LLMs to always output their thoughts in specific parts, delineated by tags? Even better: inside of that, you
-
AI Training Progress Requires Years More Data and Compute
By
–
Sadly you will need about 2-4 more years of training data, additional compute, and a bunch of post training before you can expect major improvements.
-
Metaverse Digital Twins Revolutionize Global Energy Market Growth
By
–
#Metaverse https://
openpr.com/news/3647001/m
etaverse-digital-twin-in-energy-market-is-booming-worldwide
… -
Databricks Simplifies Serverless Compute Cost Optimization Solutions
By
–
We’re simplifying serverless compute cost optimization for Notebooks, Jobs, & Pipelines! Improvements include: 25%+ automatic cost reductions for most customers Enhanced cost observability Easy controls for optimization 50% intro discount https://
dbricks.co/3Zevs30 -
SambaNova Reduces AI Energy Consumption with Llama Meta
By
–
Just listened to this insightful episode of @BBCRadio4 The Artificial Human with @aleksk and @Kevin_Fong
, where they discuss the problem of high energy consumption in AI. https://
bbc.co.uk/sounds/play/m0
0208g4
… With our record breaking performance, SambaNova is proud to run Meta’s Llama -
Attention Scarcity Over Time: The Digital Age Challenge
By
–
The true scarcity is attention, not time.
-
AI clusters’ power consumption to rival entire countries by 2028
By
–
> By 2028, AI Clusters will reach the power consumption of entire countries Reminder : “Scaling laws” are empirical laws saying that if you keep multiplying your compute by x10, your models will mechanically keep getting better and better. To give you an idea, GPT-3 can barely
-
Google’s problem: too many steps to use Gemini
By
–
Here's Google's problem. To use Gemini, you need to create a GCP account, create a service account, go to Google AI Studio, activate the product, then generate your key, which you use using AI Studio but the product is actually called Generative API.
-
MongoDB’s Aggregation Queries: Developer Convenience After 12 Years
By
–
Another developer talked abt the difficulty of queries in some databases. Of MongoDB he said, "I genuinely *love* [@mongdodb's] aggregation queries." When I first joined MongoDB in 2012, convenience/ease was its killer app. 12 years later it's still true (but now with ACID, etc)