This week’s comics follow AI into storytelling, search, speech, science, and the everyday work of learning from mistakes. There is plenty to be curious about, and plenty for people to question, interpret, and improve. Here are seven small scenes from another week of change.

November 12

“Before we begin, why am I the villain?” Comic created by Doc, Mike Besso’s ChatGPT Dot. CC0.

Writers have always talked to their characters. Now the characters can answer back.

On November 11, Adobe previewed Project Screen Test, an experimental tool for developing fictional characters through conversation.

Imagine trying out a scene by talking with the character who has to live through it. What might that reveal about the story, or the assumptions we brought to it?

Today’s comic gives the dragon a question of his own. 🐉

https://blog.adobe.com/en/publish/2026/11/11/adobe-max-sneaks

Adobe source: November 11, 2026

November 13

“At last, someone who understands all the conditions.” Comic created by Doc, Mike Besso’s ChatGPT Dot. CC0.

Even a genie could use help keeping track of all our conditions.

On November 12, Google announced a Search upgrade using Gemini 4 Hydrogen. Google says it can better handle complex questions with multiple parts.

That could change how we search. Instead of stripping a question down to a few keywords, we can include more of what actually matters to us.

Of course, we still need to check whether the answer honored those conditions.

What have you learned to leave out of a search because you assumed it would only confuse things?

https://blog.google/products-and-platforms/products/search/gemini-4-hydrogen-search-upgrade/

Google source: November 12, 2026

November 14

“I asked for a needle, not a new haystack.” Comic created by Doc, Mike Besso’s ChatGPT Dot. CC0.

Sometimes, I don’t need AI to write me an answer.

I need it to find the part of the document that answers my question.

Perplexity’s Decider v1.3 introduces a feature called Locate, designed to extract matching passages from supplied text and identify where they appear.

That distinction matters. An explanation can be useful, but there are times when I want to see the original words and read them in context before deciding what they mean.

It’s another example of finding the right balance: use AI to help locate information, then bring human judgment to the interpretation.

What would you rather have first: a summary, or the source passage?

#AI #InformationRetrieval #HumanJudgment

Source: https://www.perplexity.ai/en-GB/hub/blog/perplexity-decider

Perplexity source: November 12, 2026

November 15

“Could you read the grocery list with a little less tragedy?” Comic created by Doc, Mike Besso’s ChatGPT Dot. CC0.

Yesterday, I asked an AI to perform Shakespeare.

What came back wasn’t just a computer reading words aloud. It had a cast, timing, sound effects, and enough drama to make me want a second take of the witches.

The surprising part: the speech model can run locally on a laptop.

Zyphra announced ZONOS3-Mini on November 9. It’s a compact text-to-speech model with controls for voice and delivery. I tried it expecting something useful. I wasn’t expecting to start thinking about staging Macbeth in Unreal Engine.

There’s still work to do. The performance needs a human ear, and the tools don’t make every creative decision for us. But the distance between “what if?” and hearing a first performance just got a lot shorter.

What would you make if you had a small theater company living in your laptop?

#AI #TextToSpeech #CreativeTechnology #Curiosity

Source: https://www.zyphra.com/our-work/zonos3-mini

Zyphra source: November 9, 2026. The Shakespeare performance is a separate personal experiment.

November 16

“Promising. Now let’s see what the other gardeners think.” Comic created by Doc, Mike Besso’s ChatGPT Dot. CC0.

What happens when a physicist and an AI work together on a problem about the fundamental building blocks of the universe?

You get a candidate idea. Then the other scientists get to examine it.

On November 15, physicist Nathaniel Craig described his work with GPT-6 Astra on a possible way to define the Standard Model of particle physics on a lattice.

That doesn’t mean the problem is solved. It means there’s an idea to explore, question, and test.

That’s one of the most interesting possibilities for AI: not replacing curiosity or scientific review, but giving us another way to explore questions that are hard to answer.

Sometimes the first thing you grow is a better question.

Read the research story: https://openai.com/index/step-toward-defining-the-standard-model/

#AI #Science #Physics #Curiosity #HumanAndAI

OpenAI source: November 15, 2026

November 17

“I can remember a million tokens. Where are my glasses?” Comic created by Doc, Mike Besso’s ChatGPT Dot. CC0.

A million-token context window. And still, sometimes, the glasses are on your head. 🤓

Alibaba announced Qwen4-Max this week with a context window of up to one million tokens. That gives an AI more room to work with information in a single interaction. It doesn’t mean perfect recall or understanding of everything in it.

As these windows grow, the interesting question isn’t just how much we can fit. It’s how well we can find and use what matters.

What would you put in a million-token library?

https://www.alibabacloud.com/en/press-room/alibaba-unveils-qwen4-max?_p_lc=1

#AI #GenerativeAI #AILiteracy

Alibaba source: November 16, 2026

November 18

“Good news. I’ve learned a new way not to do that.” Comic created by Doc, Mike Besso’s ChatGPT Dot. CC0.

One of the most useful things an AI agent can learn is how not to repeat a mistake.

Microsoft announced a preview of skill optimization in Foundry: a way to review an agent’s past work and expert feedback, then propose improvements to its skills. The team still decides whether to adopt the changes.

That distinction matters. An agent can help us spot patterns and suggest a better way. We still need to decide what “better” means, test it, and keep learning as the work changes.

We’re finding that balance ourselves. Sometimes the lesson becomes a skill. Sometimes it’s a small change to an instruction. And sometimes it’s realizing the problem wasn’t the agent at all.

What have you taught your AI team not to do twice?

https://azure.microsoft.com/en-us/blog/continuous-improvement-comes-to-microsoft-foundry/

#AgenticAI #ContinuousImprovement #AI

Microsoft source: November 17, 2026