Fetching from the wire…
01
02
03
04
05
06
07
08
Research2026-06-12 · source-backed
This is my favorite research finding of the day. GEPA, an ICLR 2026 oral now shipping as dspy.GEPA, optimizes prompts by having an LM reflect in natural language on an execution trace, what went well, what failed, then evolving a tree of candidate prompts. Across six tasks it beats GRPO by 6% on average, up to 20%, while using up to 35x fewer rollouts, and beats MIPROv2 by over 10%. The reason it converges fast: it consumes domain-specific text feedback instead of only a scalar reward. If you've been hand-tuning prompts, this is the automated path that doesn't require a reward model.
Each link below shares sources, entities, or timing with this story.
Here's the one you can act on today. DSPy shipped 3.3.0b1 with a new ReActV2 module, and if you're still hand-tuning prompt strings, you're doing manual labor a compiler should do. You declare a task as a typed Input to Output signature, pick a reasoning strategy, and MIPROv2...
Released August 3, experimental dspy.Flex moves program structure into the optimizer's search space: given a signature, GEPA rewrites predictors, control flow, and the Python/LM call balance against your metric, with optimizer-authored source always running inside a CodeInterp...
PR #10305 fixes trace-capture evaluation so a crash on one example assigns failure_score instead of dropping the row, which previously caused an indexing error or silently shifted every later result. The release ties that to missing and misaligned validation results reported a...
NPO iteratively revises a prompt using a teacher model and rollout feedback, with no elaborate search (arXiv 2608.27266). It matches or exceeds GEPA at lower rollout cost, and its advantage widens with stronger teachers, which suggests teacher reasoning substitutes for optimiz...
New batching algorithms enable ~7x, up to 12x+, longer-context GRPO training with no accuracy or speed penalty versus optimized FA3 and chunked-loss setups (Unsloth Docs). Qwen3-8B GRPO reaches 110K context on one 80GB H100 via vLLM plus QLoRA. For solo builders doing reasonin...
dspy.LocalInterpreter runs generated Python in a persistent subprocess using the current interpreter, giving ordinary Python compatibility without Deno while separating the worker's memory, stdout and lifecycle from the DSPy process. The notes say plainly it is not a security...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.