---
type: "commonplace"
title: "Exhibit 7 (Case 3:24-cv-05417-AMO, Dkt. 553-8)"
description: "Jared Kaplan's Concise Research Roadmap for Anthropic 7/2021"
author: "Trey Causey"
date: 2026-10-21
slug: "2026-10-21-i-smitop-com-concise-research-roadmap-pdf"
url: "https://www.treycausey.com/commonplace/2026-10-21-i-smitop-com-concise-research-roadmap-pdf/"
tags: []
sourceTitle: "Exhibit 7 (Case 3:24-cv-05417-AMO, Dkt. 553-8)"
sourceUrl: "https://i.smitop.com/concise-research-roadmap.pdf"
---

> See clear signs of life on search, printf debugging, and other valuable tasks, evaluate whether tool use and RL are dissolving the limitations of current large LMs

<div class="commonplace-commentary">

Six months after Anthropic was founded, Jared Kaplan produced this concise research roadmap (just about 2 pages, excluding comments). 

It is **remarkable** how stable this has remained. What sticks out to me is the early focus on building models that are good at software engineering as the path to AGI.

It's not quite as simple as Tesla's famous <a href="https://www.tesla.com/secret-master-plan">Master Plan v1 and v2</a>, but I'd argue that Anthropic has stuck to the plan much more closely. 

</div>