One Shot Writing; New User Guides; Jon Stokes on Context

Here is a short list of recent notable AI developments that we think professional communicators and publishers should be paying attention to:

The Illusion of Thinking. Apple released a very important paper that shreds the view that reasoning models are working in a similar manner to human thinking. TL;DR: Reasoning models handle some specific problems very well, but can not handle general problem solving the way that humans do, and anthropomorphizing AI may be doing it no favors.

News Sites Getting Crushed By SEO. We have been saying in this newsletter for months that SEO is going to zero. TL/DR: So....Google trained its models without consent or payment on Publisher content, then took away Publishers traffic, and is now attempting to sell its models back into those same Publishers. No one who accedes to this has a right to complain later.

WaPo, Opinion and AI.The Washington Post has a new strategy to aggregate opinion columns, including non professional writers who may be assisted by AI tools provided by the Post.

And here's what's happening in Symbolic:

One Shot AI Writing For Multi-Section Drafts

Symbolic last week introduced a much anticipated newsletter function, allowing drafts to be created by Sections. Each Section can carry its own research targeting, and its own format. Now, once each Section has been set up, a single "Write All Sections" button allows you to generate the entire document with a single click. Once the draft is generated, a user can accept the entire text, reject the entire text and try again, or go through Section by Section and accept or reject at that level. This saves time and effort, and gets to the end result faster.

New Step by Step User Guides

Symbolic has a new set of user guides: https://symbolic.ai/guides. These guides take a step by step approach to some common workflows on the platform, from organizing Research to creating custom Voice and Format instructions, to writing a Sectioned Draft. More guides will be added over time, and if there is something that you would like to see as a reference, please reach out directly to us.

Did Claude Code Lose Its Mind, Or Did I Lose Mine?

Our team at Symbolic uses generative AI tools intensively in multiple phases of the software development lifecycle, from feature development, to refactoring, to bug fixing. Jon Stokes, Symbolic’s CTO, recently wrote about his experience with Claude Code — one of the more popular AI “copilot” development tools from Anthropic. Specifically, he wrote about the dangers of not having a process that you stick to when doing AI-assisted coding.

When working on part of an upcoming feature for the platform, Stokes observed that Claude Code’s output quality seemed to be degrading. At first it seemed to degrade subtly, but eventually it became unusable. Often the model providers will constantly tweak and tune their flagship models, so Stokes assumed that Anthropic’s constant fiddling with Claude was causing it to lose its mind.

But then he noticed something — in the coding session where he got the worst output, he had departed from his usual practice of breaking the problem up into smaller bites and feeding them to the model one at a time. In fact, he had beens steadily drifting away from this practice for a few weeks, and the more he drifted the worse the tool’s output got. When he went back and followed the process he had described in earlier newsletter posts, the tool’s output quality went all the way back up.

Stokes drew some lessons from the experience about the importance of ensuring that you have a process for dividing up larger tasks into smaller ones, and that you actually stick to the process. He also connected these sessions to Symbolic’s recently announced newsletter function, which is specifically designed to encourage users to break up larger drafts into smaller sections with more focused AI inputs.

That's it from us this week. We are always working to improve Symbolic and make your experience more efficient and enjoyable. We would love to hear your feedback and any suggestions you may have for future updates. Don't hesitate to reach out to team@symbolic.ai.