- Machines vs Machines
- Posts
- Galactica, Invisibility Cloaks, and More!
Galactica, Invisibility Cloaks, and More!
Hey All! Today’s newsletter has as much flavor and spice as a bucket of finger-lickin' good chicken on a fine Saturday afternoon.
Let’s Dig Right Into It!
Galactica AI “pulled” after 3 days of launch
This Christmas, gift yourself with an Invisibility Cloak
Robots that can build themselves might soon become common
Agribots are becoming better and better
Meet CICERO - an AI that negotiates with people
AI may soon become more competent
MassRobotics announces its Healthcare Robotics Startup Cohort II
The lawsuit that can change the shape of AI generation
GPT-4 is coming!
Galactica AI has been pulled back after 3 days of launch because “the vibes are off”
Every Galactica can’t be as good as Battlestar, and Meta’s latest AI tool of the same name certainly proves it.
Meta announced the launch of the demo version of Galactica AI, a tool that would make it easier to write scientific papers.
But all didn’t turn out to be rainbows and sunshine when the demo was used by experts.
I asked #Galactica about some things I know about and I'm troubled. In all cases, it was wrong or biased but sounded right and authoritative. I think it's dangerous. Here are a few of my experiments and my analysis of my concerns. (1/9)
— Michael Black (@Michael_J_Black)
6:47 AM • Nov 17, 2022
Here’s all you need to know:
Galactica AI was reported to have been trained on more than 48 million scientific papers.
It was launched to make academic papers easier and more automated to write.
Despite extensive training, the tool ended up generating results that weren’t only scientifically inaccurate but also “dangerous”, “racist”, and “insensitive”.
According to experts, it generated pretty “convincing” tones of text suggesting antisemitism is good, the benefits of suicide, and the pros of eating crushed glass.

The AI tool’s own co-founder believes all of this is a gross exaggeration,
Galactica demo is off line for now.
It's no longer possible to have some fun by casually misusing it.
Happy?— Yann LeCun (@ylecun)
5:20 PM • Nov 17, 2022
But… wasn’t the tool promised to be an accurate tool for research papers rather than the academic version of DALL E?
Potterheads Assemble! Here’s the Invisibility Cloak You Wanted in Childhood
Remember when we all used to dream about having an invisibility cloak like Harry’s two decades ago? It might now be possible to have one.
Researchers at the University of Maryland have worked with Facebook to come up with a real-life invisibility cloak, or “sweater”.

Here’s what it could do:
The “sweater” has been trained on several algorithms opposite to datasets used by AI systems to recognize people.
The purpose of this is to create an impact that would make it harder for a person to get recognized in front of an AI system.
The sweater has basically been designed to delete a person from a machine’s vision.
With its efficiency and use still in question, Reddit has had its say with one commentator adding, “the sweater is so ugly that even an AI is refusing to look at it,”
While we agree with the user’s sentiment, here’s what Japan launched with the help of green screens a few weeks back:
Japanese scientists discovered invisibility
— How Things Are Manufactured (@fastworkers6)
11:06 AM • Nov 17, 2022
Much more interesting, no?
Looking at these new inventions, we’re assured all of this ‘might’ come in handy when robots actually start ruling over the human population.
Other News:
> Researchers at MIT are working on building robots that can efficiently self-assemble without needing human help. This is the next step in robot evolution, don’t you think?
> Agribots are becoming the next big thing in farming, we read this article with utmost interest while villages in most parts of the world still struggle with learning how to use tractors efficiently.
> Meta has announced CICERO, an AI that can actively compete and negotiate with people in games. The company suggests it can pave the way for intelligent assistance in the future.
> UCLA develops AI materials with “muscle memory” of their own to adapt to environmental changes. Hmm… will they also be forced to get out of their comfort zones to “learn and grow”?
> MassRobotics announces Cohort II of their Healthcare Robotics Startup Catalyst, with ABLE Human Motion, Mitsubishi Electric, and Robot on Rails, among other participants.
> The Github Copilot lawsuit, with claims suggesting the tool is infringing copyrighted codes, could change the game on generative AI making their solutions more lawfully streamlined.
Fundings
Cradle raises $5.4 Million to develop AI-designed protein machines and cell factories
Care.ai scores $27 Million in funding
Redbrick AI has raised $4.6 Million in funding led by Sequioa
Visionary AI gets video enhancement funding
Sapia.ai raises $17 Million in Series A funding
Meme of the Day!

Source: www.warandpeas.com