Per the Fudan University researchers who built MOSS, it's not a true Chinese equivalent to ChatGPT. It's trained on a fraction of the data and performs poorly in Chinese. Interesting to see them still compare it to ChatGPT though. The hype is real. https://
txsun1997.github.io/blogs/moss.html
RESEARCH
-
MOSS: Not a True Chinese ChatGPT Equivalent Despite Comparisons
By
–
-
RLHF Insufficient for AI Alignment Problem Solutions
By
–
Retweeted this before but it seems even more relevant now. RLHF is not the way to solve the alignment problem. a lot of experience has now confirmed what @percyliang said below.
-
CarperAI trlX enables open-source RLHF model training at scale
By
–
CarperAI is doing great work lowering the barrier for RLHF training (i.e. training ChatGPT-like models). The latest release of their trlX library includes this great example, showing how to train RLHF models at scale with an open-source dataset! https://
x.com/synth_labs/sta
/synth_labs/status/1628544125120430081
… -
Can Technology Save the World? Nuclear to AI Solutions
By
–
Will tech save the world? If so, when? Talk new advances in nuclear energy, health care/medicine, AI, transportation, agtech, vertical farming, cryptocurrency, automation, microelectronics fab, etc and how that will all save us from this horrid economy.
-
First CFD Simulation on Wafer-Scale Engine Announced
By
–
Recently we announced the first-ever computational fluid dynamics simulation on our Wafer-Scale Engine. We joined our partners at NETL and PSC who made this achievement possible for a fire-side chat. Watch to learn more about how we are reinventing CFD!
-
Transformer Models: Introduction and Comprehensive Catalog Overview
By
–
Transformer models: an introduction and catalog A super concise and brief overview of almost all popular Transformer models. Written well and provide pointers for recent advances in Transformer model architectures. https://
arxiv.org/abs/2302.07730 -
Mathematics of Diffusion Models: Deep Dive Resource
By
–
On the Mathematics of Diffusion Models A great resource for those who want to dive deep into mathematics of diffusion models, the kind of models that revolutionized image generation. https://
arxiv.org/abs/2301.11108 -
Bing performs multiple sequential searches from one prompt
By
–
Example showing Bing can perform multiple sequential searches, with LLM response generation in between, from a single user prompt:
-
MIT Wearable Device Monitors Vocal Fatigue With Skin Sensors
By
–
With the help of opera singers and a speech-language pathologist, an MIT researcher helps develop first wearable device to quantify vocal fatigue.
— MIT CSAIL (@MIT_CSAIL) 22 février 2023
Combining skin-interfaced sensors w/a haptic feedback module, the device can monitor singers' vocal health: https://t.co/bkM3t56BNP pic.twitter.com/5JLoleetcAWith the help of opera singers and a speech-language pathologist, an MIT researcher helps develop first wearable device to quantify vocal fatigue. Combining skin-interfaced sensors w/a haptic feedback module, the device can monitor singers' vocal health: http://
bit.ly/3xHQaJr
