ChatGPT Groupchats, System Prompts, and Daily LLM Use Cases | Sharp Tech with Ben Thompson

Summary

• ChatGPT group chats are an additional AI-native work surface, not a WhatsApp replacement. Ben’s premise is that technology usually layers onto existing behavior: “Something new is layered on top of what came before.” Shared research and drafting could be valuable, but weak notifications, poor mention navigation, absent personal memory, and the possibility that other apps could replicate the feature make its durability uncertain—not “the deepest moat of all time.”
• Ben currently prefers deliberate AI sessions inside ChatGPT over AI embedded everywhere. He sees AI interaction as “a very purposeful, explicit thing,” even using ChatGPT’s terminal connection while keeping the interface outside the terminal. He might be wrong—OpenAI has said it wants ChatGPT incorporated across apps, and Sam Altman is among those who disagree with Ben—but Ben currently views AI work distinctly.
• Gemini’s strategic opening is expanding the AI audience through multimodality rather than displacing committed ChatGPT users. Ben argues that images, Veo video, and dynamic UIs could reach “vastly more people” because they are more compelling and approachable than text chat. His risk case is that ChatGPT becomes Twitter: intensely valuable to deeply engaged users, yet difficult for the mass market to unlock.
• Hallucinations remain an adoption barrier. Andrew initially says he uses ChatGPT mostly on his phone for personal questions and hesitates professionally because it has “hallucinated and burned me,” especially where he lacks enough China or tech expertise to catch errors. Andrew’s rule is that specific statistics require a deterministic database or original source, not an LLM; he later also praises the Mac app for spectrum-related work.
• Domain expertise can increase an LLM’s usefulness because the user supplies direction, depth, and error detection. ChatGPT helped provide context on satellite spectrum, after the numbers were checked against original sources, while Ben’s antitrust knowledge elicited a much richer precedent discussion than a novice received. “The more you know about something, the more useful it is.”
• The durable adoption threshold may be behavioral: AI shifts from occasional novelty to a default tool for unresolved questions. Ben uses it for fog formation and turkey smoking, while Andrew cites NVIDIA’s meaning of “offtake,” Christmas-tree troubleshooting, and TV sizing. Ben warns against outsourcing thought; the winning posture is to “make it your assistant instead of your boss.”

Deep dive

1. Group chats add AI to collaboration without replacing messaging

• Ben rejects the replacement frame: “Something new is layered on top of what came before.” A shared ChatGPT thread let him and Andrew plan a podcast TV setup without repeatedly copying AI output into another chat.

• Andrew’s lawyer test is stronger: associates drafting separate parts of a brief could exchange research with the model present. He questions why other apps could not replicate the feature and notes that group chats do not bring in personal history or memories. Ben says it is not “the deepest moat of all time”; its significance depends on how people work and could help shape future workflows.

• The product remains rough: users need notifications for mentions, there is no easy way to find where they were mentioned, and long AI answers need collapsing. WhatsApp’s quote-jumping and tools for managing one’s position in a chat are the benchmark.

2. Purposeful AI work may deserve its own application

• Ben currently “view[s] AI work distinctly,” so collaborating with Andrew belongs in the AI app rather than inside a word processor. He explicitly hedges that this “might be totally wrong”; OpenAI has discussed incorporating ChatGPT across apps, including through its API and “Login with ChatGPT.”

• His jobs-to-be-done analogy reaches back to Facebook’s 2013 phone effort, a heavily skinned version of Android that was also available for other Android phones. Facebook argued that phones should be organized around people rather than apps, but Ben says the app paradigm won because it fit—or perhaps shaped—the way people work: many activities are about accomplishing tasks, not contacts.

• Likewise, Ben prefers a ChatGPT connection to the terminal over putting ChatGPT directly inside it, because the terminal is already inscrutable and embedding AI there could make it harder to see what is happening.

3. Gemini can widen the market through richer formats

• Ben’s strategic concern is that ChatGPT might become Twitter—indispensable to a deeply engaged minority without reaching everyone else. Gemini’s image generation, Veo video, and dynamic UIs could open AI to “vastly more people,” even without replacing heavy ChatGPT users.

• Andrew uses Twitter’s “Gemini vibes” as an onboarding analogy: the valuable, early discussions come from accumulated specialist accounts, not Google’s official feed. Ben adds that a normal user instead encounters “a mess of takes,” just as chatbot newcomers may lack the mental model needed to extract real value.

• Andrew agrees that a more compelling graphical interface could simply make AI easier and more approachable than text-first chat.

4. LLMs provide context; sources provide verification

• Andrew initially says he uses ChatGPT exclusively on his phone, mostly for household or random questions. Professional hallucinations have “burned” him: he can catch NBA errors, but lacks equal depth in China and tech, so he is cautious about professional use and mainly cites understanding CCP bureaucracies for Sharp China. Later, he calls the Mac app critical for work involving spectrum, SpaceX, and satellites.

• Andrew’s pushback is blunt: “Why are you using ChatGPT to ask for statistics? That’s a terrible idea.” Deterministic questions need deterministic databases or original sources. ChatGPT helped provide context on satellite spectrum, including uplink and downlink, while the numbers were checked against original sources.

• Their disagreement sharpens the use case. Andrew sees less need where he already knows basketball; Ben says expertise makes AI more valuable because informed prompts force depth. Ben’s antitrust discussion surfaced cases and precedents where a novice received generic output—after which he still checked the actual cases.

5. AI becomes powerful when curiosity turns it into a habit

• Ben asks ChatGPT about fog formation with his son and keeps an ongoing thread about smoking a turkey. Andrew cites asking what NVIDIA meant by “offtake” in an earnings-call quote, troubleshooting Christmas-tree lights, and deciding on a TV size for the podcast setup.

• Ben’s default is “everything I don’t know,” extending a lifelong habit of learning how systems work and connecting them. His custom prompt asks ChatGPT to answer from its own knowledge before searching, remain terse while preserving relevant substance, and write “as if you were 2 standard deviations smarter.”

• The broader warning survives the enthusiasm: users can outsource thinking completely, but an agentic user can “make it your assistant instead of your boss.”