All episodes

The Mixed Bag

Future Feed — Useful assistant, questionable confidence

Oct 13, 2026 · 7 min 15 sec · Carlitos & Nicole

Carlitos and Nicole, hosts of The Mixed Bag

About this episode

Useful assistant, questionable confidence Resources & links Episode, sources and available recording script: https://themixedbag.media/episodes/b1be4b3c-91b6-48c3-a3a5-a6e6d787e8ff/ City Mix Blog: https://citymix.media/blog/ Connect & support Follow The Mixed Bag for new conversations. If this episode helped you, share it or leave an honest review in your listening app.

Transcript

Transcript assembled from the saved scripts for this recording. Untimed text; expression cues omitted.

Nicole: Welcome to Future Feed. I’m Nicole, your Tuesday host. Carlitos, I hear you managed to be late while writing the message explaining that you were late. How many drafts did the explanation need before anybody could actually read it?

Carlitos: I’m Carlitos. Three drafts, one deleted joke, and an explanation that had become a project. The useful message was very short. That is a good starting point for today: what does a tool actually help us do, and where do we still need our own judgment?

Nicole: Fair. Today we’re asking where generative AI can help with ordinary work, and where a person still needs to check the answer. I also tried an imaginary weekend itinerary so polished that it nearly distracted me from verifying the places and times.

Carlitos: I like beginning with a specific task. Organizing notes, drafting a message, or comparing information you already have gives us something to inspect. Saying this will change everything gives us considerably less to work with before lunch.

Nicole: The National Institute of Standards and Technology published a generative-AI risk profile in twenty twenty-four. It identifies confabulation among the risks: a system can produce false content in a confident form. Smooth writing does not settle whether the details are correct.

Carlitos: Which means a neat paragraph is not its own evidence. If the output contains a date, a quotation, a claim about a person, or a source, check the original. The confidence belongs to the presentation; our confidence needs something stronger behind it.

Nicole: Let’s try an imaginary City Mix assignment. We have three verified event notices and want a short draft listing. Giving the tool those notices can create a starting point. Then somebody compares the result with the notices, including the dates and venues.

Carlitos: And the somebody does not say it looked fine while checking only the headline. A draft might sound elegant and still change the useful detail. I would rather have an ordinary sentence that sends people to the right place than a glamorous one that doesn’t.

Nicole: We can also ask whether the tool did the task we actually wanted. Did it organize the information? Did it make a summary easier to read? Or did we spend the entire afternoon fixing changes that were never necessary in the first place?

Carlitos: That last one is the technology version of reorganizing a drawer and ending with everything on the floor. Sometimes a tool saves time. Sometimes the useful result is discovering that this particular job was already simple enough.

Nicole: The comparison should include the review work. Time to get an answer is one part; time to verify and finish it is another. In our hypothetical experiment, we could try the same small task both ways and note what changed.

Carlitos: We could also decide what success would look like before seeing the result. Fewer corrections? A clearer outline? A useful first draft? That gives us something to assess beyond whether the output made us say wow for five seconds.

Nicole: Wharton researcher Ethan Mollick gives us a useful expert perspective. In his September essay The Overhang, he argues that deep subject knowledge helps people recognize where AI succeeds and fails. That is his argument about working with AI, not a promise that a particular model will get your task right.

Carlitos: Here’s our practical application: use an assistant to organize questions for a DC itinerary, then check the museum’s actual calendar, the restaurant’s current information, and the transit agency’s alerts. Someone who knows the neighborhood can also spot an implausible route. Useful output still needs judgment about the task.

Nicole: There’s a local creative connection too. City Mix’s reporting on Revolutionary VP Studios in Laurel describes a company advertising a large LED production stage. That is a specific production proposal, with company specifications and people behind it, rather than a universal promise about the future.

Carlitos: It also reminds us that several technologies can be involved in making media. A virtual-production stage and a text generator do different jobs. We should identify which process we are discussing before saying technology is doing something for an entire industry.

Nicole: If we talk to a local creator, I’d want to ask what problem the equipment solves, what expertise it requires, and which parts of the job remain difficult. Those answers could give listeners a picture of the work rather than only a picture of the machine.

Carlitos: And whose work gets credited. Designers, technicians, performers, and people making creative decisions deserve more attention than an anonymous claim that a tool made everything. Technology reporting gets more interesting when there are actual people in the explanation.

Nicole: Here is a habit I’m trying to change: asking a very broad question, receiving a very broad answer, and then being surprised that the answer isn’t tailored to the job. Our fictional itinerary suffered from a small case of that.

Carlitos: A small case? You asked for the ultimate weekend. I would have started with Saturday afternoon, two verified notices, and how much time you had. The ultimate weekend sounds like a project that requires a marketing department and a weather guarantee.

Nicole: A fair revision. Then we keep anything personal or confidential out of a casual experiment unless we understand the tool and the permission to use that information. The easiest practice example is one we are comfortable treating as non-sensitive.

Carlitos: We also keep a person accountable for the final result. If we publish a wrong event time, saying the machine supplied it does not help the listener standing at a closed door. The editorial decision is still a decision somebody made.

Nicole: So our experiment has four questions: what is the task, what information are we using, how will we verify the result, and what would make it worth using again? That can fit on one page. Your future does not require twelve dashboards to begin.

Carlitos: And it can include the answer this tool did not help this time. Curiosity works better when we’re allowed to notice a limitation. That answer might save you more time on the next assignment than another enthusiastic demonstration would.

Nicole: We are also building something local: Around Town, our social platform in beta at City Mix. Its current public page explains that approved community posts appear in publication order, with the same feed for everyone, rather than popularity rankings. That is the concrete starting point for our algorithm-free approach.

Carlitos: Here algorithm-free describes the feed experience we are aiming for: community moments without a personalized popularity ranking deciding what you see. There is still software and moderation. Beta means we are learning and refining the experience. Visit City Mix’s Around Town page, explore the community moments, and tell us what makes a local platform useful to you.

Nicole: Bring us an everyday task you want explained, and tell us which part creates the most work. We’ll look for sources, actual examples, and people who understand the process. Future Feed is about a future you can ask useful questions about.

Carlitos: I’m Carlitos, with Nicole. Keep the curiosity and check the result. Next time I’m running late, I’ll send the useful message first. The director’s cut of my explanation can stay in the drafts. That’s Future Feed on Tuesdays.

Episode sources