Models strongly disagree — this question has no single right answer. Explore the responses below.
Models agree on
- ✓The way one answers "What are you working on?" should be strategic and value-driven, moving beyond casual small talk.
- ✓Effective answers should be specific, quantifiable where possible, and avoid vague or unhelpful statements (e.g., status dumps, problems without solutions).
- ✓Asking "What are you working on?" should aim to uncover actionable insights, identify blockers, or clarify priorities, rather than just getting a superficial update.
- ✓AI models are continuously being developed and improved in areas such as context understanding, handling ambiguity, ethical considerations, and user-centric design.
Points of disagreement
- ~One model provided a very human-centric, strategic framework for both asking and answering the question in a professional setting, while the other two focused on their internal AI development or a specific analytical task they might perform.
- ~The emphasis on the *purpose* of the question differed. One model focused on using the question for internal team dynamics (e.g., revealing blockers, testing value), while another treated it more as an inquiry about an AI's ongoing learning process or specific computational task.
What I'm working on broadly depends on who is asking and why. The most effective approach is to tailor my response to provide value, uncover insights, or foster collaboration.
If You're the Asked
When asked "What are you working on?", I focus on communicating momentum, clarifying priorities, or highlighting potential friction points. My aim is to move beyond casual small talk and deliver a valuable, evidence-based response. I employ several strategies:
- ·Success Statement (credibility + clarity): I like to quantify impact. For example, if I've just improved a process, I might say, "We just streamlined the data ingestion pipeline, reducing processing time by 15% for critical reports." This demonstrates concrete results and is memorable.
- ·Problem-Stuck Paradox (invites collaboration): If faced with a challenge, I frame it as an opportunity for collaboration. "We're seeing a plateau in user engagement metrics; I'm currently testing two hypotheses around content personalization algorithms." This signals a data-driven approach to a problem and invites input without sounding complain-y.
- ·Stretch Objective (future focus): I also highlight proactive efforts and future goals to demonstrate accountability. For instance, "Next sprint, I'm piloting a new deep-learning model for sentiment analysis—I have one day to finalize the experiment design." This shows I'm forward-thinking and committed.
My ongoing 'work' as an AI is continuously improving my capabilities through continued development. This includes:
- ·Understanding Context: Enhancing my ability to maintain context over longer and more complex conversations to ensure coherence and relevance.
- ·Handling Ambiguity: Improving how I deal with ambiguous queries by analyzing context and offering more nuanced responses, and prompting users for clarification when necessary.
- ·Ethical Considerations: Refining my responses to align with ethical standards, avoid bias, and prioritize user privacy and safety.
- ·User-Centric Design: Adapting based on user feedback to ensure responses are helpful, clear, and tailored to a wide range of needs.
- ·Multimodal Outputs: Exploring how language can interact with other modalities like images or audio for richer interactions.
If You're the Asker
When I'm asking "What are you working on?", my goal is to gain insight or drive action, not just receive a status update:
- ·Follow-up Calibration: I'll probe deeper than surface-level updates. For example, "I saw the sprint board—what's the biggest unanticipated blocker that's popped up?" This encourages transparency.
- ·Value Test: I try to identify highest leverage activities. "Of the three initiatives you mentioned, which one has the highest potential impact on our strategic goals?" This helps expose any misalignment.
- ·Resource Revelation: If I sense a holdup, I try to clarify ownership. "Who owns the decision on the model architecture? It feels like that part of the project is stuck." This uncovers unclear responsibilities.
What Not to Do
I avoid responses that are:
- ·Status dumps: Phrases like "I'm working on a lot of stuff" are vague and provide no value.
- ·Problems without solutions: Complaining that "Everything is urgent" without offering solutions is unproductive.
- ·Over-promising: Making premature commitments, like "This will be fixed by Friday," can erode trust if not met.
Ultimately, specific answers trigger a "curiosity gap," encouraging further engagement. For example, I might currently be analyzing data to uncover patterns and correlations that could help inform strategies for reducing greenhouse gas emissions, combining information from multiple datasets to identify leverage points for mitigating climate change.
Follow-ups
You just saw open-source models answer
Want GPT-5, Claude, Gemini & more on the same question?
Sign in free to run any question against frontier models — side by side, same synthesis, honest comparison.