11d • 🤖! AI Updates !🤖
⚠️WHY CHATGPT WONT SUFFICE⚠️
WARNING: a bit nerdy
A lot of people in here using ChatGPT as their base AI and this is why it won’t be enough in the future
a 27M parameter ai model just beat gpt-5 on reasoning benchmarks!
yes, 27M parameters.
that's 4x smaller than gpt-1 (117M).
why?
because singapore startup sapient intelligence built a brain-inspired architecture that thinks like humans do: two modules working together - a slow strategic planner and a fast tactical worker.
what hrm (hierarchical reasoning models) achieved:
– 40.3% on arc-agi (beats claude 3.7's 21.2%)
– near-perfect on extreme sudoku (gpt-4: 0%)
– optimal 30x30 maze solving (state-of-the-art: 0%)
– trained in 2 gpu hours with just 1,000 examples
gpt-5 today:
– billions of parameters
– $17-20 per complex reasoning task
– requires massive pre-training
– still uses chain-of-thought "thinking out loud"
translation: it doesn’t matter what model you use, it’s what the model does that matters for what you do that will make it effective.
funny thing is gpt-5 just launched with all the hype, yet a model 100x smaller solves puzzles it can't even attempt.
2
5 comments
⚠️WHY CHATGPT WONT SUFFICE⚠️
powered by
skool.com/ai-agents-infrastructure-2975
Automate, save time, and maximize results with Artificial Intelligence.
The power of AI in your hands.
Suggested communities
Powered by