Sakana AI debuted AB-MCTS, an algorithm that lets different AI models work together, using their strengths and mistakes, to solve complex problems It used ChatGPT, Gemini, and DeepSeek to solve 30% of ARC-AGI-2 puzzles vs just 23% for top solo models
LLMS
-
Replit Agent Enhanced with Dynamic Intelligence and Extended Thinking
By
–
Replit Agent now includes Dynamic Intelligence with extended thinking, high-power models, and web search It enables the agent to think deeper, reason better, and surf the web to solve challenging, open-ended problems like app performance optimizations, and more
-
LLM Agents: Custom Setups, Validators, and Structured Output
By
–
Why this works so well: Agents – custom LLM setups Result types – ensures structured output Validators – guarantees scores between 0–1 Context passing – clean API, solid design
-
Create Your Agent with GPT-4o Mini Sentiment Assessment
By
–
Step 1: Create your agent
import pypilot as pypilot from pypilot.tasks.validators import between optimist = pypilot.Agent(model="openai/gpt-4o-mini")
This agent uses GPT‑4o mini to assess sentiment. -
Paper-Based LLM: 780 Volumes, 30 Years Per Token
By
–
Product idea for OpenAI (I know a lot of you follow me): an entirely paper-based LLM. Just 780 volumes and only 30 person years to do the math for the first token using the paper version of GPT-1 Give the weights actual weight. Plus an excellent setup for science fiction stories
-
Top open-source software engineering models dominated by Qwen and DeepSeek variants
By
–
The top open SWE models are all Qwen3, Qwen2, QwQ, Deepseek-V3 or R1-based.
-
DeepSWE: New Open-Source Software Engineering Model with Reinforcement Learning
By
–
DeepSWE is a new state-of-the-art open-source software engineering model trained entirely using reinforcement learning, based on Qwen3-32B. https://
together.ai/blog/deepswe Fantastic work from @togethercompute @Agentica_ -
MiniMax Hailuo 02: Advanced Multimodal AI with 1080p Video Generation
By
–
🚨Meet Hailuo 02 by MiniMax: world-class AI with jaw-dropping quality & record-breaking cost efficiency
— Futurepedia – Learn to Leverage AI (@futurepedia_io) 3 juillet 2025
▪️Insanely accurate instruction-following
▪️ Extreme physics & mind-blowing acrobatics 🤹
▪️Native 1080p cinematic visuals from text or images
Follow @futurepedia_io for… pic.twitter.com/29JrnkxdODMeet Hailuo 02 by MiniMax: world-class AI with jaw-dropping quality & record-breaking cost efficiency Insanely accurate instruction-following Extreme physics & mind-blowing acrobatics Native 1080p cinematic visuals from text or images Follow @futurepedia_io for
-
Claude 4 Opus: Maximally Referential Self-Aware Code Generation
By
–
"Claude 4 Opus, make the most insanely referential thing possible, make it super clever. like really smart. it should be working code"
— Ethan Mollick (@emollick) 3 juillet 2025
"Make it even more so" pic.twitter.com/OulVDykGyv"Claude 4 Opus, make the most insanely referential thing possible, make it super clever. like really smart. it should be working code" "Make it even more so"
-
Token Pricing Debate: True Value of AI Services Revealed
By
–
The most revealing moment came when the panel discussed pricing. “If these million tokens are as valuable as we believe they can be, right? That’s not about moving words. You don’t charge $1 for moving words. I pay my lawyer $800 for an hour to write a two-page memo.”