- Bot Eat Brain
- Posts
- Google DeepMind's AlphaGeometry solves problems from IMO
Google DeepMind's AlphaGeometry solves problems from IMO
PLUS: Balls
TOGETHER WITH
Good morning, human brains, and welcome back to your daily munch of AI news.
Here’s what’s on the menu today:
AI pwns math chads 🤓 💻
DeepMind’s AlphaGeometry solves university level geometry problems.
TikTok’s new Depth Anything model 📸 📡
It’s a depth perception model that achieves state-of-the art results.
Sports. Sports. Sports. 🤖 ⚽️
IBM and Savilla FC launched Scout Advisor, a player recruitment tool.
From our partners
Discover Your Next Venture with Microns 🚀
Delivered to your inbox every week, Microns.io is a newsletter to help you discover profitable and bootstrapped startups for sale.
Join 6,000+ forward-thinking entrepreneurs like you and uncover the secrets behind great deals.
Learn firsthand from our success stories:
✅ Indie Space Store - View on X
✅ Remote Profile Newsletter - View on X
✅ MyArchitectAI SaaS - View on X
As seen on the Hustle. Trusted by 6,000+ entrepreneurs and investors.
MAIN COURSE
AI got that math rizz 🤓 💻
Back in September, we reported on Google DeepMind’s AlphaMissence. It’s an AI tool that helps find the cause of genetic diseases.
In November, we covered Google DeepMind’s AlphaFold. It’s an AI program that predicts the 3D structure of proteins, which directly leads to new drugs.
Yesterday, we reported on Google DeepMind’s AlphaFold new breakthrough. It identified new psychedelic molecules to potentially use in new antidepressants.
What did those geeks do this time?
Last week, Google DeepMind and NYU researchers developed AlphaGeometry. It’s an AI model that solves university-level geometry problems.
Who cares?
AlphaGeometry solved 25/30 geometry problems from IMO 2000-2022. This surpasses previous AI capabilities and matches the average human gold medalists.
How is it better than ChatGPT?
Models like ChatGPT rely on visual-spatial reasoning and have issues with translation. AlphaGeometry combines neural language models for enhanced problem-solving and logical proof verification, which more resembles fast and slow thinking processes in the human brain.
What’s under the hood?
It’s trained on 100 million unique geometry problems and proofs, which allows expansive knowledge development without relying on limited human expert data.
So it’s a math nerd, big whoop.
The methods used in AlphaGeometry could be adapted to other purposes. Google DeepMind plans to open-source its code and models in the future.
RECOMMENDED READING
Here’s a newsletter we recommend for entrepreneurs, investors, operators, and creators. Anyone interested in making more money, and becoming a better manager of people should read Sweaty Startup’s weekly newsletter.
SIDE SALAD
AI goes deep 📸 📡
Back in May, we reported on TikTok’s new chatbot Tako. It’s a chatbot that you can chat with, ask for video recommendations, and more.
On Friday, researchers from TikTok shared a paper on Depth Anything. It’s a new AI-driven depth perception model that outperforms previous state-of-the-art models.
That’s deep, bruh. 😤 🤟
It’s a TikTok thing?
It’s an AI model developed by researchers from TikTok, The University of Hong Kong, and Zhejiang Lab.
What’s so great about it?
It achieves unmatched depth estimation, which means it understands object distances from images. It was trained with 1.5 million labeled images and over 62 million unlabeled images.
‘Yuge.
What’s the point?
It accurately measures depth in even challenging environments like low-light and underwater. It’s highly scalable and adaptable, so it can be used to enhance TikTok’s green screen effect, autonomous driving, and more.
A LITTLE SOMETHING EXTRA
Balls 🤖 ⚽️
In August, we reported on Amazon’s football AI features. It provides game insights, highlights, graphics, and more to its Thursday Night Football streams.
In September, we covered an AI drone that beat the world champion FPV drone racer. It was reportedly the first time AI outperformed humans in a physical sport.
Sports. Every AI nerd’s favorite topic.
Yesterday, IBM and Sevilla FC announced an AI scouting tool, Scout Advisor. It’s a tool to improve the efficiency of the player recruitment process.
Sevilla FC?
It’s a Spanish football club. They launched Scout Advisor, an AI tool developed by IBM. It uses IBM’s Watsonx and natural language processing to help scouts identify and evaluate potential recruits.
How does Scout Advisor work?
It accesses over 200,000 scouting reports and searches for quantitative metrics and qualitative assessments. It summarizes player details, tracks performance stats, generates lists of recommended players based on specific criteria, and more.
“Our team of scouters and analysts is committed to finding the best players to support our team’s success and have worked tirelessly to build an impressive database of player reports, and now, with the support of IBM, we can utilize this asset to its full potential.”
YOUR DAILY MUNCH
Tools
Bonkers — easy-to-use, intuitive text-to-image platform.
Verisoul — Cybersecurity platform that protects against threats, bots, and more.
Holly — AI-powered hiring platform that helps you find candidates efficiently.
GetWebsite.report — audits your sites and gives AI-powered insights.
Think Pieces
How AI is helping the human immune system. It helps predict the outcomes of surgery, design vaccines, and more.
How to use AI to skyrocket your business’s success. What currently dormant sectors will receive a resurgence with AI?
AI doomsday is just hype. An MIT study about how many jobs AI will replace and when.
Startup News
NVIDIA is hosting an AI developer contest. Winners could receive a GPU, tickets to its GTC conference, and more.
ElevenLabs raised $80 million in Series B funding. It’s a startup that creates voice creation/editing tools.
Qdrant raised $28 million in Series A funding. It’s a Berlin-based startup that offers an open-source, vector search engine/database.
Research
Rambler — an AI interface that takes lengthy, fluffy speech and turns it into easily readable text.
ActAnywhere — a video generation model that creates backgrounds that align seamlessly with motion.
MEDUSA — accelerates LLMs by incorporating multiple decoding heads for parallel token prediction.
MEMES FOR DESSERT
TWEET OF THE DAY
The Berkeley humanoid got some flack for being slow. “My grandma could walk faster.” It’s still pretty cool, though.
Tag us on Twitter @BotEatBrain for a chance to be featured here tomorrow.
AI ART-SHOW
Until next time 🤖😋🧠
What'd you think of today's newsletter? |