Hi all(emaal),
Hope everyone is enjoying the start of the sunny Dutch weather after a rather rainy King’s Day. Here’s a list of what’s been happening in AI the last few weeks, from Mormons publishing guidelines on how to use AI to a disabled writer with brain damage fighting for copyright on her book that heavily relied on AI generated tekst to tell her story, and even a rather cool application of eye-tracking software to re-imagine ancient artistic carvings.
This time also a rather large ‘etc’ section at the bottom full of some more miscellaneous and.. intriguing topics within the AI world.
On to the news-
Mormons published a set of guiding principles for using AI to spread their religious message
Yet another example of hallucinations in LLMs causing security risks as they don't learn to deal with fringe cases in otherwise stable patterns, this time by it trying to install packages that don't exist (which can be abused by creating this package yourself and putting it online with malicious intent)
Article on the current state of generative AI and its copyright infringement woes
List of 16000 artists' names leaked as part of the 'style list' purportedly used by Midjourney in the training of their generative image models
School employee used AI Deepfake to make school principal sound racist, people were quick to judge and harass the principal before due process revealed the ruse
Ethan Mollick on how he wrote a book utilising AI-tools as assistant within the writing process
Adobe to add 3rd-party AI video generation tools, like OpenAI's Sora, to Adobe Premiere
The story of a disabled writer with brain damage fighting for copyright on her book that heavily relied on AI generated tekst to tell her story
As technology is more oftenly used as a medium to present art, so too does the issue of fast technological developments decrease the 'lifespan' of an art piece through technological obscelence.
Using eye-tracking software to reinterpret ages-old artistic carvings
Sidney Leeder, producer behind the SORA short film Air Head, on how they actually used the new video generation AI in production in order to create the short film
Paper exploring how generative AI methods could be used for personalised learning experiences including simulation exercises, coaching, and co-creation
How science fiction acts like a great tool for educators to talk about AI in the classroom, and it's potential opportunities and threats
Using the introduction of the calculator as an example of how AI can transform the classroom without any doom scenarios
Google launched a low-/no-code AI Agent Builder on their Vertex platform, allowing for the low threshold building of multi-agent AI systems, a powerful (and expensive) technique to gain much better results from recent LLMs/Foundation Models
Study by Harvard Business Review on how generative AI is actually used in practise
Overview article explaining the current game of scale in AI, some of its hurdles, and the enormous amounts of investments currently being thrown at AI developments
Stanford's Institute for Human-Centered AI published their annual AI Index report for 2024, read their general findings (which won't surprise those following AI developments) or download their report here
Increasing progress on using brain implants with 'artificial vision' to help bring forms of sight back to blind folk
Detecting parasitic infection through analysing human excrement with AI models
LLaMA 3 was launched by Meta, likely to become the new foundation for most (semi-)open-source models
New, even bigger, Mixture of Experts model from Mistral called Mixtral 8x22B (8 models of 22 billion parameters in size), released in a semi-open-source manner
New state of the art 'background removal' model RMBG v1.4, published as an open-source model for non-commercial purposes
State Space Models (SSM) are a rising 'alternative' to transformer-architectures, although they are conceptually very similar they allow for longer context windows through compression of information and faster speeds
Publication discussing a systems research approach towards an LLM Runtime framework which can guarantee certain properties such as 'reversibility of actions' and 'ensuring safekeeping of cryptographic secrets'
Convolutional Neural Networks (CNN) used to study the flight mechanics of flies
Humans braincells communicate in a feed-forward manner as opposed to cycli seen in other animal models, at least when considering the outer layers of the neo-cortex. Some similarities drawn towards feed-forward neural-networks within AI modeling
What is (e)-personhood? An ancient philosopher's take on what defines human thought vs animals, mapped on today's AI developments
How self-assembling DNA strands can be used as a classifier for sorting images in order of brightness
Video showcasing in real-time how OpenAI tech in a humanoid robot produced by Figure1 allows for voice control of robotics in environments they weren't trained for
What if we make Transformers run on Quantum computers?
Creative new way of benchmarking LLMs (tongue in cheek): making them battle eachother in Pokemon
See you in the next one!
Leven is mooi
