Hi all! Easter has just ended, but it was there when I was still in the middle of synthesising this newsletter. Next to that, I had to be careful about the latest news that was published on April 1: April’s Fool Day (Check how Gary Marcus is excited about the GPT-5 preview). This week’s newsletter includes thoughts on the ongoing build-up to really helpful writing tools. In the news roundup, we have new AI battles, personal agents, LLM houses. And robots. A paper on the City as a Licence. And there are also some events to attend or track.
Triggered thought
Grammarly’s new generative AI features are intriguing yet intrusive, raising questions about the balance between suggestion and takeover. Writing remains the signature case for generative AI, just as music was for recommender systems, with Spotify Discover Weekly as the prime example. In the realm of creative writing, the ideal tool acts as a buddy, writing coach, and background researcher, but it has yet to be perfected. Some tools, like Grammarly, focus on improving grammar and now aim to inspire better writing through generative AI. ChatGPT and Claude promise to help build stronger arguments by easily incorporating background sources. However, there is a tension: the writing produced by these tools, especially ChatGPT, can be cliché-ridden and uninspiring. Lex, a tool that has undergone iterations to enhance creativity in writing, may be worth revisiting, so I did.
While it doesn’t yet add references, it is still necessary to construct narratives by prompting different services. I invited Lex to rewrite my first version, and it did a nice job, I think. I tried the rewrite function with Claude Opus and Sonnet, and with GPT-4. The latter creates rather formal speech and removes all personality. Funny enough, Claude Sonnet took a different standpoint, creating an observing piece: “the author wonders, etc…” So the above argumentation is built with the support of Claude Opus. And Grammarly, that keeps suggesting…
The question remains: when will we reach a point where a ghostwriter can start researching based on triggered concepts and collaboratively build a case without taking over the writing process entirely? One aspect is touched upon in the robotic facial expressions (see the news item below): in creating a natural feel of interaction, the AI must predict human behaviour to respond on time. But that is maybe something to dive into more at a later time and thought.
Read the notions of the news, paper for this week and events to track, via the newsletter. You can also subscribe for weekly updates on Tuesday 7 am CEST.
target is new 