✅ Created aiNews: AI News/AI-News-2024-12-20.md ✅ Created chat: AI Chats/Chat0001-12202024.md ✅ Created workEffort: Work-Efforts/WE0001-12202024.md ✅ Created newsletter: Newsletters/NovaBrew-2024-12-20.md
Friday, December 20, 2024
Week 51 of 2024 • Q4
⬅️ Previous Day | 🏠 Home | Next Day ➡️
📊 Day at a Glance
- 🗓️ Day: 355 of 2024 (97%)
- 📅 Week: 51 of 52
- 📊 Quarter Progress: 81 of 92 (88%)
- 🎯 Days until EOY: 10
- 🔄 Created at: 10:13 AM
📰 AI News
📝 Daily Reflections
Fri, 12/20/2024, 10:13
o3
o3
o3
OpenAI announces a new model.
It’s incredible at programming. Math. 96.7% accuracy on Competitive Math. It misses only one question on the math olympiad 87.7% correct on PhD level science. ARC benchmark President of the Arc foundation was PRESENT and sitting on the livestream. North star toward AGI ARC AGI 2019 by fancois chollet Unbeaten for 5 years. New state of the art score to announce Arc AGI is about having input examples and output examples Arc AGI version 1 took 5 years to go from not passing to passing. o3 scored 75.7 on Arc AGI semi-private holdout set New number 1 entry on Arc AGI hub When they ask o3 to think longer, it scored 87.5% on the same problem set. Human capability is at 85% threshold The president literally said “I need to change my worldview. The work is not over yet. We need more enduring benchmarks to monitor and guide progress. We are excited to work with OpenAI to develop new benchmarks. Ark prize will continue into 2025.”
o3-mini
Hony Yu Ren - model trainer
3 options:
- low, medium, and high reasoning time
Evals:
- coding:
- Sam “For an order of magnitude lower in cost, we can deliver the same performance as the larger frontier models, or higher.”
- new cost-efficient frontier on coding, math, science, every fucking thing
LIVE DEMO 03-minii is half the latency of o1 Achieves better performance with scaling of thinking and time at inference and compute.
Release - End of January, Early February
Fri, 12/20/2024, 18:18
Wow. We just hit something significant. A true step change today.
Fri, 12/20/2024, 18:18
Tonight I am at Stoble getting things ready, listing the Cognex and the Allen-Bradley parts on Ebay
Fri, 12/20/2024, 19:19
I’m gonna run a little experiment on escapecapitalism.org where I have an AI model perform capitalism in a little game of evolutionary iterative development. I want to make an AI do capitalism in a simulated environment.
See EscapeCapitalism Technical Specs for implementation details.
Fri, 12/20/2024, 21:18
I found the fastest and easiest source of money and I’m going to earn, if it’s what it appears to be, about 3000 from selling these microcontrollers, and then about another 3000 from selling these cables, and then yet ANOTHER 3000 selling other unrelated items, which means I should have about $6000 of play money coming to me for the season if I can get all this shit sold.
Bonkers dude. Bananas.
Fri, 12/20/2024, 21:21
Escape Capitalism - I can pretty much expect to be able to one-shot these kind of games and apps with o3 huh…
I want to see if Claude can develop a game in the browser with abstract english instructions. Let’s find out…
Fri, 12/20/2024, 21:49
I love the iterative development process these LLMs use. Can’t wait to apply it to science and robots. This is gonna be fucking COOL
Fri, 12/20/2024, 23:49
Ok I got it working. I have two AI models talking to one another in a chat window, doing some simulated capitalism. This is sick as fuck. Damn. As I learn to do more of this…
🐾 Today’s Pokémon
Kleavor (#900)
- Type(s): bug, rock
- Abilities: swarm, sheer-force, sharpness
- Stats: hp: 70, attack: 135, defense: 95, special-attack: 45, special-defense: 70, speed: 85
“Gotta catch ‘em all!” — Pokémon Trainer
🤖 GPT Thoughts
“If this works it’s gonna be so cool!” Please work…
🤖 Could not fetch GPT insights at this time.
✅ Quick Actions
- ☀️ Morning Review
- 📈 Check Analytics
- 🤝 Community Engagement
- 🌙 EOD Reflection
📱 Creator Hub
Content Pipeline
Latest Analytics
Projects
Connect with Me
🤖 AI Workspace
Active Prompts
Models
daily-note Friday week-51 q4