- Chaos Theory
- Posts
- 🥟 Chao-Down #251 DeepMind's latest AI AlphaGeometry aces high-school math, FDA approves an AI product that predicts cognitive decline, MIT researchers use AI to identify new class of antibiotics
🥟 Chao-Down #251 DeepMind's latest AI AlphaGeometry aces high-school math, FDA approves an AI product that predicts cognitive decline, MIT researchers use AI to identify new class of antibiotics
Plus, how OpenAI is approaching the 2024 worldwide elections.
Another big breakthrough just came out of Google’s DeepMind. And this time it’s AI is coming for high school math.
Google DeepMind recently published a new AI system that solve geometry problems at the Math Olympiad level. The system named “AlphaGeometry” solved as many geometry problems as the average human International Mathematical Olympiad (IMO) gold medalist who is usually someone currently in high school.
It’s a big leap forward for AI in mathematical reasoning as solving geometry problems requires flexibility, creativity, and a solid understanding of the rules and theorems. Unlike other AI models, AlphaGeometry did not include extensive web data in its training regime. Rather, the system combines a neural language model with a symbolic deduction engine to validate ideas with mathematical rules. In order to train the model, DeepMind generated 100 million synthetic theorems and proofs.
The team tested AlphaGeometry on 30 challenging IMO geometry problems and the system successfully solved 25 of the problems within the 4.5-hour time limit! This surpassed the previous best system that could only solve 10 problems.
You can read more in-depth the results in their Nature paper as well as their open-sourced code on Github.
-Alex, your resident Chaos Coordinator.
What happened in AI? 📰
The FDA has reportedly approved an AI product that predicts cognitive decline (Engadget)
Using AI, MIT researchers identify a new class of antibiotic candidates (freethink.com)
Amazon launches generative AI tool to answer shoppers' questions (CNBC)
How OpenAI is approaching 2024 worldwide elections (OpenAI)
AI Startup Anthropic Is Working on Image Analysis for Chatbot Claude (Bloomberg)
AI models that don’t violate copyright are getting a new certification label (The Verge)
Always be Learnin’ 📕 📖
Psychology for UX: Study Guide (nngroup.com)
Measuring Developer Productivity: Real-World Examples (pragmaticengineer.com)
Web LLM attacks | Web Security Academy (portswigger.net)
Projects to Keep an Eye On 🛠
maybe-finance/maybe: Personal finance and wealth management app (Github)
data-drift/data-drift: Metrics Observability & Troubleshooting (Github)
VikParuchuri/surya: Accurate line-level text detection and recognition (OCR) in any language (Github)
The Latest in AI Research 💡
LLM Augmented LLMs: Expanding Capabilities through Composition (arxiv)
AMIE: A research AI system for diagnostic medical reasoning and conversations (Google Research Blog)
INTERS: Unlocking the Power of Large Language Models in Search with Instruction Tuning (arxiv)
The World Outside of AI 🌎
Costco is testing out a new system for entering stores (CNN Business)
Why Teslas and other electric vehicles have problems in cold weather — and how EV owners can prevent issues (CBS News)
Empty Nesters Own Twice As Many Large Homes As Millennials With Kids (Redfin)
China’s Worst Deflation Streak in 14 Years Puts Pressure on PBOC (Bloomberg)
Bottled Water Contains Previously Undetected Nanoplastics (Bloomberg)
Could psychedelics unlock blindness, stroke, anorexia, autism, and more? (Vox)