Thinking about how my brain thinks
Joined May 2013
- Tweets97
- Following333
- Followers84
- Likes2.1K
ChinniKrishna retweeted
Did you see @velaura_ai during your Bay Area commute down US-101? If not, now’s your chance.
Most billboards on 101 promote bigger AI promises, but ours is doing something different. While other billboards scream for more, we’re advocating for less: less power, that is.
Velaura AI is advancing ultra low power technology to improve performance per watt across AI data centers and power constrained physical AI applications.
If you're driving through the Bay Area on US-101, look up (safely) and find the @velaura_ai billboard for a limited time!
#NYSE #VelauraAI #Funding
LLMs are proficient in deciphering patterns while detached from the physical realm.
Delve into the intricacies of this concept in my essay "The Ungrounded Flavor".
Read here: highentropythoughts.substack…
Feedback welcome!
#AI #LLMs #CognitiveScience #PhilosophyOfAI
So Physics is figuring out "Ultimate transformation" which takes current state of universe at t and outputs state at t+1?
#Physics #ComplexSystems #Time #ScienceHumor #PhilosophyOfScience #Emergence #Computation
Ever wonder why we use transposes in backprop?
It’s a reflection of dual spaces.
The transpose maps output gradients back to input measurements.
Read all about it here highentropythoughts.substack…
#DeepLearning #LinearAlgebra #Backpropagation #MathTwitter #AI #ML
Self-attention is nothing more than a kernelised, normalised, content read-out operator. #LLMs
Excited to share PaperMiner—a Python script I developed that extracts paper titles from a conferences and uses openAI API to generate summaries.
Check it out on GitHub: github.com/chinnikrishna/Pap…
🚀📚#AI #Python #ResearchTools #MachineLearning #LLM #ML
ChinniKrishna retweeted
Working in AI means I frequently see people I have never heard of presented as "leaders in the field" or having made "fundamental contributions". Probably because the field is so vast, and no-one knows what really qualifies as AI. But also because of: narcissists+sycophants+$$$
While exploring what my TV antenna receives I took a quick detour to make a simple AM radio using python and numpy.
Quick shoutout to @DigilentInc for not only making affordable USB scopes but also giving access to their API with very detailed documentation. #sdr
Thinking about a problem from first principles is both a gift and curse. Gift because you have a new insight, curse as very few might get it or appreciate it.
#MachineLearning #ArtificialIntelligence #causalmodel
ChinniKrishna retweeted
Microsoft researchers & collaborators created an open-source library to explore “what-if” scenarios for machine learning models. Learn how their method generates multiple diverse counterfactuals at once and gives insight into ML algorithm decision making: aka.ms/AA75k8p
ChinniKrishna retweeted
Announced today–The Intel Nervana NNP-T, the Intel NNP-I, and the future @IntelMovidius VPU enable larger, more complex AI algorithms and elevate the reasoning & context missing from today’s AI in the cloud, on the edge, or an IoT device. intel.ly/34Udtzd #IntelAI
ChinniKrishna retweeted
The code is open-sourced as part of our NLP Architect library: github.com/NervanaSystems/nl…. Work by @MosheWasserblat, @peter_izsak and others. #IntelAI
BERT coming to small compute devices?
“We show how quantization-aware training during the fine-tuning phase of BERT can compress it down by 4×. The quantized model can accelerate inference speed if it is optimized for 8bit Integer supporting hardware.”
arxiv.org/abs/1910.06188
Beautiful explanation of Riemann surfaces
youtu.be/4MmSZrAlqKc
Great job @welchlabs made my day :-)
ChinniKrishna retweeted
Using gradient descent for everything xitter.cf/danhett/status/1176116…