- Stream Notes
- Catch Up
- Computer acting super weird
- Car stuff
- 24 hours of Le Mans at Porsche dealer
- Began posting old stream notes to BuenosDS.dev!
- Dug into the Docker logs for last week, can’t figure out the error, asked Vespa Slack
- Currently, Vespa does not support models bigger than 2GB, feature request here.
- Workaround is to use GGUF
- AI/Tech News
- AI papers of the week
- Text-to-LoRA: Instant Transformer Adaptation
- Introduces a hypernetwork-based approach for instantly generating LoRA adapters from natural language task descriptions, removing the need for conventional task-specific fine-tuning.
- What is a LoRA?
- Reinforcement Pre-Training
- This paper introduces Reinforcement Pre-Training (RPT), a new paradigm that bridges LLM pretraining and RL by reinterpreting next-token prediction as a reasoning task rewarded via verifiable correctness. Instead of relying on hand-curated annotations or costly human feedback, RPT applies RL on vast unannotated text corpora, assigning intrinsic rewards based on whether a predicted token matches the ground truth.
- TableRAG: A Retrieval Augmented Generation Framework for Heterogeneous Document Reasoning
- Kinda seems like they’re trying to let people “talk” to their data
- Unclear who this product is helping, wondering how/if it would interface with visualizations/dashboards
- Code Researcher: Deep Research Agent for Large Systems and Code Commit History
- Skillful Joint Probabilistic Weather Forecasting from Marginals
- Text-to-LoRA: Instant Transformer Adaptation
- AI Agents Weekly
- Pay wall
- These seem interesting:
- How Anthropic uses Claude Code and builds research agents
- Databricks has launched Agent Bricks
- Smol News
- Open Chinese Models
- MiniMax-M1
- Hailuo 02 (fka Kangaroo)
- Video model
- Moonshot AI’s Kimi-Dev-72B
- Coding
- Columbia University finds that AI agents easily fall for injection/phishing attacks
- LangChain cooked
- Davia, Python app to web UI
- Google planning to cut ties with Scale AI
- LangChain: How and When to Build Multi-Agent Research Systems
- Build Deepseek from Scratch YouTube Playlist
- Lot of theory discussion
- Local Open Source VSCode Copilot with MCP
- Open Chinese Models
- AI papers of the week
- We raided phweedomstudio
- Catch Up