From 45e246d369833d2605c4dc2a627be5a9c77364df Mon Sep 17 00:00:00 2001 From: Protocol Zero <257158451+Protocol-zero-0@users.noreply.github.com> Date: Thu, 14 May 2026 04:20:31 +0000 Subject: [PATCH] =?UTF-8?q?docs:=20rework=20README=20tone=20=E2=80=94=20so?= =?UTF-8?q?und=20like=20dev=20docs,=20not=20a=20pitch=20deck?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit The previous rewrite framed the value proposition explicitly: a section literally titled "The 30-second pitch" with an emoji-decorated table of Economics / Engineering / Strategy. That reads as marketing copy in a developer-community README. Same substance, but the form telegraphs the audience and dilutes credibility with peer developers. This pass keeps the content (cost / reliability / portability) but moves it into the kind of section developer READMEs actually have: - "The 30-second pitch" → "Motivation" — standard dev-doc heading - 3-row emoji table → 3 plain bullets framed in engineering terms: * "Inference cost collapses" (the dollars-vs-cents reality) * "Production reliability is mechanical, not aspirational" (ledger / sandbox / line-count anchor) * "A harness is a portable asset" (transfer + compounding) - Removed "multi-billion-dollar dynamic", "shipping agent products", "bespoke labor done by senior staff", "reproducible / transferable / compoundable" — all phrasings that read as investor-deck syntax rather than engineering observation. Also de-marketed two other lines in the Roadmap target section: - "**The thesis.**" → "**Why this gap is closable.**" - "**What this story will demonstrate when it lands.**" → "**Why this run is worth doing.**" The "story" framing implied a script being executed. The new framing asks the engineering question the run is meant to answer. English and Chinese READMEs updated in lock-step. No code changes. Co-Authored-By: Claude Opus 4.7 --- README.md | 20 +++++++++++--------- README.zh.md | 20 +++++++++++--------- 2 files changed, 22 insertions(+), 18 deletions(-) diff --git a/README.md b/README.md index 2a15998..adb8caa 100644 --- a/README.md +++ b/README.md @@ -23,15 +23,17 @@ --- -## The 30-second pitch +## Motivation -Frontier-class agent behavior is not exclusively a function of model size. It is the joint product of *the model* and *the harness that runs the model* — the prompt strategy, the tool-calling logic, the sampling, the verification, the retry policy. Today, that harness is hand-tuned by senior staff at every serious AI lab. **Evolution Kernel turns harness tuning into a reproducible, auditable, automatable runtime.** Point it at a target repo, give it a measurable goal, walk away. Come back to a git branch of accepted improvements, a ledger of every decision, and — if the goal was well-chosen — a small model behaving like a much larger one. +Frontier-class agent behavior is the joint product of *the model* and *the harness that runs it* — prompt structure, tool loop, sampling and best-of-N, verifier, retry policy. Today that harness is hand-tuned by senior engineers at every serious AI lab, and the resulting code is usually the actual ceiling against which the base model is judged. -| | What this unlocks | -| --- | --- | -| 💰 **Economics** | Frontier API calls (GPT-5.5, Claude Opus 4.7) cost $$ per task. A 3 B-active open-weight model running locally costs fractions of a cent. Closing the capability gap *without retraining* is a multi-billion-dollar dynamic for anyone shipping agent products. | -| 🛠 **Engineering** | Every decision is ledgered, every change is a named git commit, every experiment runs in a git-worktree sandbox **and** an OS-level firejail sandbox. This is a runtime you can put in production, not a research demo. | -| 🎯 **Strategy** | "Prompt engineering" and "harness tuning" are bespoke labor done by senior staff. This loop makes them reproducible, transferable, and compoundable. | +Evolution Kernel makes harness tuning a reproducible runtime. Point it at a target repo with a measurable goal, walk away. Come back to a git branch of accepted improvements, a ledger of every decision, and — if the goal was well-chosen — a small model behaving like a much larger one. + +What that buys you, concretely: + +- **Inference cost collapses.** A 3 B-active model runs locally at fractions of a cent per task; the frontier APIs (GPT-5.5, Claude Opus 4.7) bill in dollars. Closing the capability gap *without retraining* moves an agent stack from "expensive to run at scale" to "near-zero marginal cost at scale". +- **Production reliability is mechanical, not aspirational.** Every decision is ledgered, every accepted change is a named git commit, every experiment runs in a git-worktree sandbox plus a firejail OS-level sandbox. The whole runtime is ~1,900 lines and was designed to be put in production, not demoed. +- **A harness is a portable asset.** Once you have evolved a good harness for one model, it transfers to other models in the same parameter class. The work compounds across model generations instead of being thrown away each time a new base model ships. --- @@ -97,7 +99,7 @@ Every attempt is written to a **ledger**: goal, observation, plan, diff, evaluat **Why this model.** [Qwen3.6-35B-A3B](https://qwen.ai/blog?id=qwen3.6-35b-a3b) is Alibaba's flagship open-weight model (Apache 2.0, released April 16 2026): a Mixture-of-Experts architecture with 35 B total parameters but only **3 B active per token**. It runs on a single consumer GPU. At a 3 B active-parameter footprint — roughly **30× smaller** than a frontier-class dense model — it already lands at 73.4 % on SWE-bench Verified, **within 15 points of GPT-5.5 while costing fractions of a cent per task to run locally**. -**The thesis.** That 73.4 % is the result of months of hand-tuned harness engineering by the Qwen team. Closing the remaining gap to GPT-5.5 (88.7 %) is exactly the kind of work Evolution Kernel automates — the planner converges on better tool selection, parallel sampling, verifier loops, and error-pattern recovery. Same model, same weights, evolved harness. +**Why this gap is closable.** That 73.4 % already represents months of hand-tuned harness engineering by the Qwen team. The remaining gap to GPT-5.5 (88.7 %) is more of the same work — better tool selection, parallel sampling, tighter verifier loops, error-pattern recovery — and that is exactly what Evolution Kernel automates. Same model, same weights, an evolved harness. **What the loop will do, generation by generation** (illustrative — these are the *kinds* of moves the planner has historically converged to during internal prototyping): @@ -147,7 +149,7 @@ Final: 73.4 % → ~85 % within 4 points of GPT-5.5 · within 3 of Claude Opus Inference cost on the target model: ~$0 (runs locally on a single GPU) ``` -> **What this story will demonstrate when it lands.** Frontier-class agent behavior is not exclusively a function of training compute or model size. A 3 B-active open-weight model + an automatically evolved harness can close most of the gap to the largest closed-source frontier model — at one-thirtieth the active-parameter footprint and near-zero inference cost. The harness then becomes a portable asset, usable with any model in the same parameter class. +> **Why this run is worth doing.** If it lands as targeted, a 3 B-active open-weight model plus an automatically evolved harness will close most of the gap to today's largest closed-source frontier — at one-thirtieth the active-parameter footprint and near-zero inference cost. The harness, once evolved, transfers to other models in the same parameter class. --- diff --git a/README.zh.md b/README.zh.md index 4d02573..c889dbb 100644 --- a/README.zh.md +++ b/README.zh.md @@ -23,15 +23,17 @@ --- -## 30 秒电梯演讲 +## 动机 -旗舰级 agent 能力不只是模型大小的函数。它是 *模型* 和 *运行模型的 harness* 的联合产出——prompt 策略、工具调用逻辑、采样、验证、重试策略。今天，这套 harness 在每一家严肃的 AI 实验室都靠资深工程师手调。**Evolution Kernel 把 harness 调优变成一个可复现、可审计、可自动化的 runtime。** 把它指向一个目标 repo，给它一个可衡量的目标，然后离开。回来看到的是：一条 git 分支记录所有被接受的改进、一份 ledger 记录每一个决策、以及——如果目标选得好——一个表现得像大模型的小模型。 +旗舰级 agent 能力是 *模型* 和 *运行它的 harness* 的联合产出——prompt 结构、工具循环、采样和 best-of-N、验证器、重试策略。今天这套 harness 在每家严肃的 AI 实验室都靠资深工程师手调，最终结果通常就是评判基座模型时实际的能力上限。 -| | 这件事的价值 | -| --- | --- | -| 💰 **经济** | 旗舰 API 调用（GPT-5.5、Claude Opus 4.7）按任务计费 $$；3 B-active 开源模型本地跑只要几分钱。在 *不重训* 的前提下缩小能力差距，对任何做 agent 产品的公司都是十亿美元量级的动力学。 | -| 🛠 **工程** | 每个决策进 ledger，每个改动是 git commit，每次尝试在 git worktree 沙箱 **加** OS 级 firejail 沙箱里跑。这是一个可以放进生产的 runtime，不是 research demo。 | -| 🎯 **战略** | 「调 prompt」「调 harness」今天是资深员工的手艺活。这个循环把它变成可复现、可迁移、可累积的资产。 | +Evolution Kernel 把 harness 调优变成一个可复现的 runtime。把它指向一个目标 repo，给一个可衡量的目标，然后离开。回来看到的是：一条 git 分支记录所有被接受的改进、一份 ledger 记录每一个决策、以及——如果目标选得好——一个表现得像大模型的小模型。 + +具体兑现成什么： + +- **推理成本塌缩。** 3 B-active 模型本地跑每任务几分钱；旗舰 API（GPT-5.5、Claude Opus 4.7）按美元计费。在 *不重训* 的前提下缩小能力差距，让一个 agent 栈从「规模化跑就烧钱」变成「规模化跑接近零边际成本」。 +- **生产可靠性是机械的，不是口号。** 每个决策进 ledger，每个被接受的改动都是一个具名 git commit，每次实验跑在 git worktree 沙箱加 firejail OS 级沙箱里。整个 runtime 共 ~1,900 行，设计目标是放进生产，不是 demo。 +- **Harness 是可移植资产。** 一旦给一个模型进化出好 harness，它能套用到同一参数级别的其它模型上。工作随模型迭代累积，而不是每次出新基座模型就丢掉重来。 --- @@ -97,7 +99,7 @@ **为什么选这个模型。** [Qwen3.6-35B-A3B](https://qwen.ai/blog?id=qwen3.6-35b-a3b) 是阿里 2026-04-16 发布的开源旗舰（Apache 2.0）：35 B 总参 Mixture-of-Experts 架构，**每 token 只激活 3 B**。一张消费级 GPU 就能跑。3 B 激活参数足迹——比旗舰稠密模型**小约 30 倍**——在 SWE-bench Verified 上已经做到 73.4%，**距离 GPT-5.5 只差 15 分，而本地推理成本是每任务几分钱**。 -**核心论点。** 那 73.4% 是 Qwen 团队几个月人肉调 harness 的成果。把剩下到 GPT-5.5（88.7%）的差距闭合，恰好是 Evolution Kernel 自动化的工作——规划器会收敛到更好的工具选择、并行采样、验证器循环、错误模式恢复。同一个模型，同一份权重，进化后的 harness。 +**为什么这个差距可以闭合。** 那 73.4 % 本身已经是 Qwen 团队几个月人肉 harness 工程的产物。剩下到 GPT-5.5（88.7 %）的差距，是同类型的工程工作——更好的工具选择、并行采样、更紧的验证器循环、错误模式恢复——而这正是 Evolution Kernel 自动化掉的事。同一个模型，同一份权重，一份进化过的 harness。 **这个循环将做什么（按代展开）**（示意——这些是规划器在内部原型中历史上倾向于收敛到的 *那一类* 动作）： @@ -145,7 +147,7 @@ 目标模型推理成本：~$0（单卡本地跑） ``` -> **这个故事落地后将证明什么。** 旗舰级 agent 能力不只是训练算力或模型大小的函数。一个 3 B-active 的开源模型 + 自动进化出的 harness，可以闭合到当今最大闭源旗舰之间大部分差距——激活参数足迹只有三十分之一，推理成本几乎为零。Harness 由此变成一个可移植资产，可以套到同一参数级别的任何模型上。 +> **为什么这个 run 值得跑。** 如果它按目标落地，一个 3 B-active 的开源模型加上自动进化出的 harness，就闭合了到当今最大闭源旗舰的大部分差距——激活参数足迹只有三十分之一，推理成本几乎为零。进化后的 harness 可以转移到同一参数级别的其它模型上。 ---