Inside Artificialis - #12
Jan 31, 2023 12:43 pm
Hey, . Welcome to the monthly version of the Artificialis newsletter!
A summary of what happened during the month, the best blogs of our Medium publication, latest news of our Discord server events, and recent news in the world of Artificial Intelligence and Machine Learning!
So seat back, take a cup of coffee, let's go over what happened this month.
A little bit from me and the server:
This January, we introduced two main things:
- Our custom Discord bot, Visionary, has now entered its 2.0 version, only supporting slash commands. We'll be developing more advanced commands, including GPT and DALL-E integrations!
- We introduced the Distinguished role: this will be given to exceptional people who constantly help others in the server and who's active in the community (it can be here in the server, or our medium publication writers) and it'll come with perks (access to special Visionary's commands, more priviledges, giveaways, etc)
From our medium publication:
- From Sound to Sight: Using Vision Transformers for Audio Classification, Alessandro Lamberti
- How to choose a deep learning architecture to detect aircrafts in satellite imagery? - Jeff Faudi
- BERT model for Classification Task. Ham or Spam Email? - Iva Vrtaric
AI in the world
Google’s Deepmind fine-tuned an AI model similar to ChatGPT to answer medical questions
It’s called FlanPaLM, and it has 67.6% accuracy on the US Medical Licence Exam questions. To pass, you need to have 60% accuracy.
Yes, this AI could pass the test to be a licensed US doctor.
Microsoft is looking into integrating OpenAI’s chatGPT into Bing, its search engine
Microsoft Corp is in the works to launch a version of its search engine Bing using the artificial intelligence behind OpenAI-launched chatbot ChatGPT.
Microsoft could launch the new feature before the end of March, and hopes to challenge Alphabet-owned search engine Google [...]
Microsoft’s new AI can simulate anyone’s voice with 3 seconds of audio
Last week, Microsoft released VALL-E, a new text-to-speech AI designed to closely match the quality of any person’s voice if given a three-second audio sample. The system builds off an audio compression technology known as EnCodec to analyze and break down a person’s voice into its fundamental components, which it then uses to create other phrases with a similar tone and sound. VALL-E can also mimic the environment of the audio sample, such as making it sound like it comes from a telephone. Microsoft has not opened the software for user testing, acknowledging the risks it could pose by impersonating speakers or manipulating audio. [...]
Deep Learning Algorithm Can Hear Alcohol in Voice
La Trobe University researchers have developed an artificial intelligence (AI) algorithm that could work alongside expensive and potentially biased breath testing devices in pubs and clubs.
The technology can instantly determine whether a person has exceeded the legal alcohol limit purely on using a 12-seconds recording of their voice. [...]
An AI chatbot trained to predict mental health disorders gained medical device status in the UK, with a 93% accuracy. At the same time, the peer support platform Koko used GPT3 + humans to provide emotional support to 4.000 people.
Messages composed by AI (and supervised by humans) were rated significantly higher than those written by humans on their own. Response times went down 50%, to under a minute.
MusicLM: Generating Music From Text
We introduce MusicLM, a model generating high-fidelity music from text descriptions such as "a calming violin melody backed by a distorted guitar riff". MusicLM casts the process of conditional music generation as a hierarchical sequence-to-sequence modeling task, and it generates music at 24 kHz that remains consistent over several minutes. Our experiments show that MusicLM outperforms previous systems both in audio quality and adherence to the text description. Moreover, we demonstrate that MusicLM can be conditioned on both text and a melody in that it can transform whistled and hummed melodies according to the style described in a text caption.
Tip of the month
This month I wanted to share something I recently found:
The tuning playbook:
This document is for engineers and researchers interested in maximizing the performance of deep learning models.
Our emphasis is on the process of hyperparameter tuning. We touch on other aspects of deep learning training, such as pipeline implementation and optimization, but our treatment of those aspects is not intended to be complete.
'Till next month, you can find everyone here:
Have a fantastic month, !
If you'd like to support our community, and have access to millions of amazing articles and tutorials, consider subscribing to Medium's program via our link, we'll receive a small portion of your fee.
All of the money will be used to sponsor our Discord's events prizes!
