Hi all(emaal),
Small re-upload, something seems to have gone wrong with the scheduling of post publishing. News moves fast, but last month’s updates should very well still be relevant. Hope you can again find something of value here!
April’s issue should launch as scheduled (properly this time) next week.
Now without further ado-
Musk requesting USA court to decide whether GPT-4 reached human intelligence
New non-profit company Fairly Trained checks whether companies used copyrighted data in the training of their AI models
Article talking about LLMs (like ChatGPT) being put in robots, and its potential repercussions
Digging deeper into empathy within AI systems, its potential and certain fallbacks for faking empathy
How prompting an LLM that it's in star trek was most effective at getting the AI to do maths (or: how rudimentary our usage of LLMs really is)
First time a cartoon created through heavy use of Generative AI aired on Chinese state television
EMO, new model that is able to generate convincing videos of portraits singing, speaking, etc, using only a singular portrait image as reference, either realistic, historic, or fantastical (like anime)
AI performing well on divergent thinking tasks, which is being written about as it showing its creative potential vs humans
One of the biggest educational resource providers in the US bought up Writable, an AI tool build on ChatGPT to support teachers in their grading and other tasks
Results of survey on 1500+ USA kids between 12-17 years old, on what they (want to) understand about AI
Sal Khan, person behind the big non-profit organisation Khan Academy, on why he thinks AI will have an important role in education's future
Another study looking at GPT4 (the model underlaying ChatGPT) in medicine, finding it to function at expert+ level in handeling questions about eye medical topics
New Machine Learning model developed for simulating chemical reactions in organic material
Blog by OpenAI's head of engineering, talking about the 5 biggest scaling obstacles, potentially interesting for both the 5 topics mentioned as for what's left out
Toyota developing a 'soft' robot
Humanoid robots are getting better and more 'natural' in their movements fast, see this video of the new Unitree H1
Singapore in talks to invest directly into OpenAI
Inflection.AI released their Pi model's new version 2.5, focussing on longer personal conversations
Anthropic released their new model Claude 3, which is performing better at a wide scale of much-used benchmarks
First LLM build using federated learning (i.e., trained by combining the resources of different organisations without needing to share their private data with eachother)
Stability.AI released the technical report for Stable Diffusion 3
Research article looking at Github Copilot's effect on programming efficiency and performance
Finetuning ('teaching') LLMs to divide arithmatic tasks into smaller chunks seems to vastly improve its arithmatic potential
New research on 1-bit LLMs for improving efficiency
See you in the next one!
Leven is mooi
