Just started a side project. @InfoKase_com — a marketplace for premium intel. The internet used to lack information. Now it’s drowning in noise. AI is only making the flood worse. InfoKase connects buyers and sellers of premium research and private intel across all
RESEARCH
-

Tina: Open Source AI Project with Training Logs and Models
By
–
Notion Blog: https://
shangshangwang.notion.site/tina
Code: https://
github.com/shangshang-wan
g/Tina
…
Training Logs: https://
wandb.ai/upup-ashton-wa
ng-usc/Tina
…
Models: https://
huggingface.co/Tina-Yi -

Tiny Reasoning Models via LoRA Achieve Strong AIME Performance
By
–
Apparently, you can build tiny reasoning models via LoRA.
Tina is a family of 1.5B models trained using LoRA-based RL—cost-efficient and performant.
The best Tina hits >20% gain in reasoning performance and 43% Pass@1 on AIME24,
with just $9 in post-training + eval cost. -
Understanding AI Model Performance Through Ablation Studies
By
–
It would be nice to see ablations to understand when this breaks and when it excels potentially
-
Consensus Methods for Variance Reduction in LLMs
By
–
The interesting question is why and when would this work? Intuitively this looks like variance reduction in a setup where variance dominates bias because the LLM has very high capacity. So consensus makes sense. Thoughts @roydanroy @ArnaudDoucet1 ?? It has vibes of online
-
Prompt Evaluation and Optimization: Continuous Iteration
By
–
5/ Evaluation & Optimization Prompts aren’t “set it and forget it.” They A/B test, evaluate, and refine. They track hallucinations, latency, cost, and safety. Prompt engineers treat prompts like software always in iteration.
-
Product Thinking: Designing Prompts for Real Use Cases
By
–
4/ Product Thinking They design prompts with real use cases in mind. AI isn’t just for demos. It powers: • Customer support
• Internal tools
• Research agents
• Creative workflows Prompts must work inside real products and systems. -

Isomorphic Labs Showcases AI Drug Design at ICLR2025
By
–
Stop by the Google booth at #ICLR2025 today at 5PM to chat with Agnieszka, an Engineering & Research lead from Isomorphic Labs. Learn about the company and its unique approach towards Drug Design with AI. https://
isomorphiclabs.com/articles/meet-
our-team-agnieszka-michael
… -

DeepMind AI Replicates Fruit Fly Behavior Using MuJoCo Physics Simulator
By
–
DeepMind built an AI that replicates how a fruit fly walks, flies, and behaves — using MuJoCo, its open physics simulator
— Rowan Cheung (@rowancheung) 24 avril 2025
The work will help scientists understand what drives specific behaviors in the fly, finding links that labs can't always measurepic.twitter.com/hziUoOGZWODeepMind built an AI that replicates how a fruit fly walks, flies, and behaves — using MuJoCo, its open physics simulator The work will help scientists understand what drives specific behaviors in the fly, finding links that labs can't always measure
-

Survey of AI Agent Protocols: Standards and Interoperability
By
–
A Survey of AI Agent Protocols https://
arxiv.org/pdf/2504.16736