latent space

The 2025 AI Engineering Report — Barr Yaron, Amplify

Aug 1, 2025 · 12:33

Barr Yaron presents early findings from the 2025 State of AI Engineering survey at the AI Engineer World's Fair. The survey reveals that while titles vary widely, the community is broad and technical, with nearly half of seasoned software engineers having worked with AI for three years or less. Over half of respondents use LLMs for both internal and external use cases, with 3 of the top 5 models for customer-facing products from OpenAI. RAG is the most popular customization method (70%), and fine-tuning is surprisingly common, with 40% using LoRA or QLoRA. More than 50% update models monthly, and 70% update prompts at least monthly. A 'multimodal production gap' exists, with image, video, and audio usage lagging text. 80% say LLMs work well, but less than 20% say the same about agents, though most plan to use them. Evaluation is the top pain point.

Latent Space Paper Club: AIEWF Special Edition (Test of Time, DeepSeek R1/V3) — VIbhu Sapra

Jul 25, 2025 · 53:54

VIbhu Sapra presents DeepSeek's latest models and announces a new curriculum-based Test of Time Paper Club. DeepSeek's May 28th R1 update doubles reasoning tokens (12k→25k) and matches O3/Gemini 2.5 Pro, improving AIME 2024 from 70% to 87.5%. A new distillation of R1 traces into Qwen 3 8b achieves performance on par with Qwen's 235b thinking model, proving reasoning models distill efficiently. The Test of Time Paper Club will cover 50–100 foundational papers over six months in San Francisco and remote, targeting core AI engineering topics from attention and scaling laws to inference techniques. Sapra emphasizes that pure RL (GRPO) on verifiable math and code unlocks emergent reflection and aha moments, shifting scaling from pre-training compute to inference-time reasoning.