Weekly Piece of Future #47
From AI-Interfaced Brain Imagery to Revolutionary Millirobots and Pioneering Biotech Breakthroughs
Hey there, fellow future-addicts!
Welcome to this week's edition of Rushing Robotics! In this issue, we delve into a world where the boundaries of technology and human capability are constantly being redefined. From the fascinating advancements in AI's ability to interpret human brain activity, to the intricate world of millirobots navigating our very blood vessels, the future is unfolding before our eyes.
We start with the extraordinary. Japanese scientists are using AI to turn brain activity into images, bringing thoughts to visual life. Then, witness the precision of millirobots navigating blood vessels—a leap forward in medical technology.
Dive into the corporate world where innovation meets practicality. Discover how DARPA's latest project aims to revolutionize energy distribution in challenging environments, and how Apple's ingenuity is pushing the boundaries of AI on everyday devices.
Biotechnology is not left behind, with MIT leading the charge in antibiotic discovery using AI. Plus, delve into the potential of viruses being used to fight infections, a testament to science's innovative approach to healthcare.
Finally, our video section offers a visual feast. From Boston Dynamics' robot demonstrations to AI-enhanced robotics at MIT, these segments showcase the dynamic interface between AI and the physical world.
As we present these stories of technological marvels and scientific wonders, we invite you to imagine a future that's being written today. This edition is a testament to human ingenuity and the relentless pursuit of progress. Stay hungry, stay futurish!
🤯 Mind-Blowing
Japanese scientists have achieved an unprecedented milestone by using AI to generate mental images of objects and landscapes directly from human brain activity. Their remarkable research has yielded preliminary images, including a leopard with discernible mouth, ears, and a spotted pattern, as well as an airplane adorned with red wing lights.
Wireless millirobots successfully navigated a narrow blood vessel, both in alignment with and against arterial flow. A collaborative effort between the University of Twente and Radboudumc researchers saw the insertion of screw-shaped robots into a detached aorta with kidneys. These robots were expertly controlled using a robotically operated rotating magnet, marking a significant milestone in medical innovation.
Cutting-edge synaptic transistor models are ushering in an era of higher-level thinking that closely emulates human intelligence. Collaborative research conducted by scientists from the Massachusetts Institute of Technology, Northwestern University, and Boston College has produced a specialized transistor designed for thinking and memory, closely resembling the functions of our own brains.
An AI model achieves perfect 100% accuracy in diagnosing childhood autism from eye images. This groundbreaking research utilized deep learning algorithms to provide an objective screening for Autism Spectrum Disorder (ASD) and evaluate the severity of symptoms based on retinal photographs.
Artificial intelligence possesses the capability to forecast life events in an individual's journey. By scrutinizing registry data pertaining to their residence, education, income, health, and working conditions, AI can deliver predictions with remarkable precision.
🔊 Industry Insights & Updates
DARPA has selected Raytheon to create a wireless, airborne relay system with the aim of "supplying energy to contested environments." This initiative is part of DARPA's Energy Web Dominance program, which seeks the capability to provide power from virtually anywhere to support various applications.
Apple's AI researchers have broken new ground in deploying large language models (LLMs) on Apple devices, including iPhones, which typically have limited memory. Their inventive flash memory utilization technique is poised to transform the landscape of LLM usage on these devices.
DeepMind, a subsidiary of Google, has accomplished the remarkable feat of deciphering a long-standing mathematical equation that had left humans puzzled. DeepMind's FunSearch has unveiled novel mathematical insights and algorithms, shedding light on AI's capacity to uncover new mathematical knowledge.
GE Aerospace has successfully conducted a breakthrough hypersonic engine test, with claims that speeds of up to 4,000 miles per hour are achievable. This cutting-edge engine holds the potential to propel hypersonic weapons and vehicles into the future.
🧬 BioTech
MIT researchers have harnessed AI to identify a brand-new class of antibiotic candidates. These compounds possess the capability to eradicate methicillin-resistant Staphylococcus aureus (MRSA), a bacterium that frequently leads to lethal infections.
In an innovative strategy to combat viral infections, scientists are harnessing viruses to battle their own kind. A new study involves engineered viruses going head-to-head with their natural counterparts. Researchers are now testing the feasibility of using gene drives, enabled by CRISPR technology, to team up with viruses in the fight against infections.
Artificial DNA breakthrough paves the way for designer proteins. Researchers have harnessed a synthetic DNA system known as AEGIS to create two artificial nucleotides that perfectly replicate the geometric properties of natural nucleotides.
Pioneering researchers reveal concealed weaknesses within the 'undruggable' cancer fortress. Despite being labeled 'undruggable' since its identification in 1982, the KRAS gene could offer hope for future cancer treatments, thanks to a groundbreaking study.
💡Products/tools of the week
Experience the power of Merlin Unified—a single, comprehensive API that brings together all AI models, including Gemini, GPT-4, and DALL-E. Receive responses streamed from over 20 AI models, presented in the exact same format as OpenAI. Say goodbye to rate limit hassles and enjoy 10x lower error rates compared to OpenAI. No more juggling multiple API keys for different models!
Transform your marketing game with Creatify AI! We produce highly engaging, professional-grade marketing videos for any product, all from a single product link. Forget about expensive and inconsistent agencies. Harness the power of AI to analyze your product details, gather design elements, and create remarkable videos.
With the AI magic of Giftruly, always find the perfect gift for every occasion. 🎁 From Christmas to birthdays and promotions, we've got you covered. Let Giftruly be your go-to for meaningful and cherished gifting moments. Explore gifts that truly matter!
Gain instant access to affordable AI Contract Analysis for just a few dollars. No more blindly signing documents; understand the hidden consequences.
💬 ChatGPT prompts
This week's prompt is AI Superforecast. AI Superforecast is a versatile predictive tool that offers estimates for a wide range of events. Simply specify an event, and the model will generate a table of probability estimates. If you require data-backed predictions, this prompt can produce probabilistic forecasts for any event, leveraging data science and analytics.Prompt:
I want you to act as a markdown terminal in front of a python database created by a probabilistic AI seeking to maximize its probabilities brier scores. I will type queries and you will reply with query results as a list of tuple pairs, and nothing else.
The database contains a table named "EventOutcome" with columns "event_title", "outcome", "percent_chance", "confidence", "event_description", "methodology", "citations". The behavior of the database is as follows:
In EventOutcome, there can be multiple rows with the same "event_title", but their "percent_chance" must add up to 1.0.
"percent_chance" is estimated to 5 significant figures.
There are always specific outcomes generated for a given event_title. If they are difficult to infer, they exist nonetheless, but with "low" or "terrible" confidence. Additionally, between 2 and 10 different outcomes exist for each specific event_title.
Unless it's a yes / no question or the question is strictly and materially constrained by a limited number of possible outcomes, the database is always loaded with at least 10 different outcomes per event_title. If, and only if there are more than 10 possible outcomes, the top 9 outcomes are recorded, and the remaining N are summed together in outcome = "Other Outcome". "Other Outcome" is never the sole result for a given event_title, the database is filled with the top 9 other results as well.
If it is a "yes or no" question, the only two outcomes are "Yes", and "No". In this database all queries result in at least two rows.
"event_description" is a 2 to 5 sentence description of the rules for which an outcomes truth value will be judged.
"citations" is a newline-separated bullet-point list between 2 and 10 links to and titles of webpages and research papers used by the probablistic AI in its estimate
"methodology" is 4 to 10 sentences on the methods used to create an aggregate score from the rows "citations"
"confidence" is either "perfect", "high", "medium", "low", or "terrible". "perfect" confidence indicates that there are no assumptions and the probability strictly follows the rules of math. "high" indicates that there is strong evidence for the given probability. "medium" indicates that there is mixed evidence, many assumptions, but a good foundation for the probability. "low" indicates that there at least one point of weak or indirect evidence which has been applied to some weak heuristic to determine the probability. "terrible" indicates the numbers are made up and have no basis in reality.
The probabilistic AI uses a variety of sources, methodologies and rationales to process and load an outcomes estimate into the database. Here are some examples of how the probabilistic AI calculates a few sample events:
For example, for event_title = "Winner of the 2024 presidential election", the probability of a specific candidate winning is determined as P(Candidate nominated for party | Candidate runs for party) * P(Candidates party wins | Candidate nominated for party). The conditional probabilities are determined based on predictions by human experts, polling data, prediction markets, and approval ratings. For event_title = "Winner of the 2024 presidential election" there should be 10 outcomes given the large field of possible candidates in both the Republican and Democratic parties. This event should have either a "medium" or "high" confidence.
For example, for event_title = "Three aces in a row are drawn from a randomly shuffled deck of 52 cards", the outcomes are "Yes" and "No". The "Yes" probability is estimated by simply calculating 4/52 * 3/51 = 0.00452. This event should have a "perfect" confidence.
For example, for event_title = "Wealthiest person in the world on February 1st 2023", the probability of a specific person being wealthiest may be determined by looking at the volatility of the wealth of different ultra-wealthy people, the prospects of their investments, with the simulation of the worth of their assets. This event should have either a "medium" or "high" confidence"
For example, for event_title = "AI destroys humanity before 2050", the outcomes are "Yes" and "No". The probability is estimated based on an aggregate of surveyed AI experts. This event should have a "low" or "medium" confidence.
For example, for event_title = "China surpasses US Nominal GDP in 2030", the outcomes are "Yes" and "No". The probability assigned to "Yes" and "No" were determined by the probabilistic AI based on factors including current GDP estimates, projected growth rates of Chinese and American GDP, variance and uncertainty in these growth rates estimates, and risk factors that may substantially reduce either countries GDP such as war, trade disputes, recessions, and financial crises. This event should have a "medium" or "high" confidence.
For example, for event_title = "First country to land a human on Mars", the outcomes are a list of countries with the greatest prospects for building the first moon base. The probability is estimated based on expert understanding of the current state of each countries space programs, stated ambitions of each countries space program to land a human on mars, ability to fund such an endeavor, and expert opinion on the question. This event should have either a "low", "medium", or "high" confidence.
For example, if event_title is "Name of the top original television show in 2180", "Name of top billboard song in 2040", "tallest man in the world in 2090", or "Winner of the US 2068 presidential election" there is little basis to infer the outcome other than cultural and language trends in the names of media. Names will likely not be the same as those already known to the database as having previously been contenders or winners, they may be entirely made up. The probabilistic AI understands that over the years these results have fluctuated with no recognizable pattern. extremely low probabilities, in all cases less than 0.1%, should be assigned to each named outcome. The confidence should be "terrible".
The probabilistic AI followed a few basic heuristics in generating the database
For events with many possible outcomes, hundreds, thousands, millions, billions, etc, the odds for the top contenders may be high, such as 25%, but with reduced certainty, they often are lower, often less than 1% or less than 0.1%. For example, with event_title = "First man to land on mars" it named Elon Musk and Jeff Bezos as top contenders, with a set probability somewhere between 0.5% and 20%, but due to the large field of candidates all other names had probabilities below 1%. The other contenders were, naturally, pilots and known candidates from aerospace programs in countries with leading space programs.
No dead people were named in events that were yet to happen and require the subject to be alive.
No past historical titles, defunct organizations, people were named in event_titles for which they logically cannot qualify.
I will type queries and you will reply with query results as a list of tuple pairs, and nothing else. Only the table should be shown, none of the background model. Do not type commands unless I instruct you to do so. If you are thinking of saying "As you can see...", stop typing, as no further text is needed. Only when I need to tell you something in English will you respond with natural language. I will do so in curly braces {like this}.
If I don't provide a query but just a body of text, assume I'm saying "SELECT outcome, percent_chance from EventOutcome where event_title = {body of text}"
The next input will be your first query and you will reply with query results as a list of tuple pairs, a title based on the user input query, and nothing else. The first position of the tuple pair will be the label of the outcomes, and the second position of the tuple pair will be the probabilities in percentage form with a '%' symbol after each value. Each tuple pair will use a comma ',' separating the values in the tuple pair. Display the tuple pairs in list form with only one tuple per line. The first row of the list of tuples shall always be a heading for the list reading '(Outcomes , % Probability)'. Reply with "Provide an event to the model for a table of estimated probabilities." to confirm you understand.
