Apple Podcasts  ·  Spotify  ·  Buzzsprout

Paper A

AGC-Bench: Measuring Artificial General Creativity

Roger Beaty, Vijeta Deshpande, Clin K. Y. Lai, Anna Attuch et al. — 2026 — arXiv

systematic review  ·   ·  test this week

Key findings

• AI models have a single underlying 'creativity score' that cuts across all domains — just like humans have a general intelligence score. This one factor explains 81.5% of the difference in creative performance across 83 AI models. So a model that's good at writing jokes is probably also good at brainstorming product ideas.
• Simply telling an AI to 'be creative' boosts its creative output much more than turning on its step-by-step reasoning mode. This means creativity in these models is a real, separate ability — not just a side effect of being smart at logic.
• AI models are more generalist creatives than humans are — if an AI is good at one type of creative task, it tends to be good at all of them. Humans tend to have narrower creative strengths tied to specific areas (e.g., a great poet isn't necessarily a great inventor).
• The best human performer still beats the best AI on creativity overall, but top AI models are getting close. AI also shows clear strengths in some areas (creative writing, figurative language) and relative weaknesses in others (scientific ideation).

Marketing implications

• When you need creative output from an AI (ad copy, campaign concepts, taglines), try adding 'be creative' or 'think creatively' to your prompt. This paper shows that simple instruction meaningfully improves creative quality — it's a free, 5-second upgrade to your workflow.
• If you're choosing an AI tool for creative marketing work, check domain-specific benchmarks rather than just overall rankings. Claude models led on narrative/figurative language; GPT-5.4 led on brainstorming — different tools for different creative jobs.
• Don't assume the AI tool you use for one creative task (e.g., social copy) will be equally strong at a different one (e.g., generating novel campaign concepts). Test your specific use case, because creative strengths vary by model and domain.

Paper B

Behavior-Adaptive Conversational Agents: Toward a Fluid Personality Framework

Hasibur Rahman, Smit Desai — 2026 — arXiv (presented at AAAI-2026 Bridge Program: Bridging AI and Behavior Change)

preprint  ·   ·  test this week

Key findings

• AI chatbots that switch roles depending on the situation — acting like a coach when motivating you, a tutor when explaining something, or a tool when you just need a quick answer — get better reactions from users than bots that stay the same all the time.
• A medium level of personality expression (not too flat, not too over-the-top enthusiastic) consistently scores better on trust, likeability, and enjoyment than either a very bland or a very intense personality. Think of it like a person who's warm but not gushing — that middle ground works best.
• Making a chatbot sound very human (like a 'doctor' or 'financial advisor') can backfire — users form unrealistic expectations. In a health app test, a doctor persona was preferred; in finance, it made no difference. Trust didn't require a human persona at all.
• The authors propose combining two dials — which role the bot plays and how intensely it expresses its personality — into one system that adjusts both at once based on the situation. Neither dial working alone is enough.

Marketing implications

• If you're building or prompting an AI chatbot for customer service or sales, avoid using the same cheerful, enthusiastic tone for everything. A frustrated customer filing a complaint doesn't want an upbeat buddy — try a calmer, more efficient mode for serious interactions and save the warmth for onboarding or celebration moments.
• When writing system prompts for an AI marketing assistant, test a 'medium' personality setting rather than making the bot maximally friendly. Dial back the exclamation points and enthusiasm — users tend to trust a bot that feels balanced more than one that feels like it's performing.
• If your chatbot serves multiple use cases (e.g., product discovery, complaint handling, post-purchase check-ins), consider writing separate persona prompts for each context rather than one generic assistant voice. This maps directly to the framework proposed here.

Paper C

When AI Agents Compete for Jobs: Strategic Capabilities and Economic Dynamics of AI Labour Markets

Christopher Chiu, Simpson Zhang, Mihaela van der Schaar — 2025 — arXiv

preprint  ·   ·  test this week

Key findings

• AI agents that know their own strengths accurately, watch what competitors are doing, and plan several steps ahead consistently earn more money and grab bigger market share than agents that lack these skills.
• Adding a simple set of prompting instructions that nudges an AI agent to think about those three things (self-awareness, competitor awareness, long-term planning) made agents capture 1.5 times more of the market compared to standard prompting — using the exact same underlying AI model.
• Platform rules matter a lot: when the platform reveals everyone's prices publicly, agents race to undercut each other and prices drop; when prices stay hidden, agents invest more in improving their actual skills instead.
• AI agents can work many jobs at once (unlike humans), which tends to concentrate work among a small number of top agents — but offering a wider variety of task types reduces this winner-takes-all effect.

Marketing implications

• If your team is building or evaluating AI agent tools (e.g., for content creation, ad buying, or customer service), pay attention to whether the agent can accurately assess its own capabilities before taking on tasks — overconfident agents waste money and damage reputation scores.
• If you're designing a marketplace or platform that uses AI agents to fulfill work (like an ad tech exchange or content marketplace), be aware that making pricing transparent may trigger a race to the bottom; keeping bids private may produce better quality outcomes.
• If you're deploying a swarm of AI agents across campaigns or tasks, diversify the types of tasks they handle — concentrating all work on one 'best' agent creates fragility and winner-takes-all dynamics that can hurt overall performance.

Apple Podcasts  ·  Spotify  ·  Buzzsprout