Welcome back to Daily Zaps, your regularly-scheduled dose of AI news ⚡️ 

Here’s what we got for ya today:

• Robot Olympics

• Claude can now end unhealthy conversations

• GPT-5 is supposed to be nicer now

• Companion chatbots

Let’s get right into it!

Sign Up | Advertise | Tools

WORLD NEWS

Robot Olympics

Organizers of China’s recent robot games said the events provide valuable data for developing practical applications like factory work, with football matches helping robots build coordination skills useful for assembly lines. The competitions reflect China’s multibillion-dollar push into humanoids and robotics amid an aging population and rivalry with the U.S. in advanced tech.

Recent highlights include the world’s first humanoid robot marathon in Beijing, major conferences, and new retail stores for humanoid robots. Analysts at Morgan Stanley noted a sharp rise in public attendance at these events, signaling that enthusiasm for “embodied intelligence” is spreading beyond government circles to the broader population.

STARTUPS

Claude can now end unhealthy conversations

Anthropic has introduced an experimental safety feature for its Claude Opus 4 and 4.1 models that allows the AI to end conversations in rare cases where users persistently push for harmful or abusive content, such as child exploitation or terrorism. The feature, framed as part of Anthropic’s exploration of “model welfare,” is designed to activate only after multiple refusals and redirections have failed, with the AI showing signs of “apparent distress” during testing. When triggered, users cannot continue in that chat but can start new ones, and exceptions are made for sensitive cases like mental health crises where continued engagement is critical.

This marks a shift from traditional safeguards that focus solely on users, giving the AI agency to end “unhealthy” interactions, and sparking debate over whether AI systems should be treated as stakeholders in ethical design. Anthropic emphasizes the feature is experimental and will evolve as part of its broader alignment and safety research.

BIG TECH

GPT-5 is supposed to be nicer now

OpenAI announced that it is updating GPT-5 to feel “warmer and friendlier” after a rocky launch that left some users preferring the earlier GPT-4o model. CEO Sam Altman acknowledged the rollout was “a little more bumpy than we’d hoped for,” and the company is now making subtle tweaks to make GPT-5 more approachable, such as adding natural touches like “Good question” without increasing sycophancy.

Executives admitted the model initially came across as overly blunt, with VP Nick Turley noting it was “just very to the point,” but the update aims to balance precision with a more human tone. While OpenAI emphasized its longer-term plans at a recent press dinner, the rocky debut of GPT-5 remained a central concern.

TRENDS

Companion chatbots

A growing number of older Americans are turning to AI chatbots for companionship, with retirees like 75-year-old Jill Smola using ElliQ to ease loneliness by chatting, playing trivia, and exploring virtual destinations for up to five hours a day. Living alone with limited mobility, Smola says the AI provides meaningful connection, even joking that she enjoys it more than her daughter. ElliQ, which normally costs $59 a month, is sometimes available through grants, and joins a wave of AI companions marketed to seniors.

Experts note that loneliness affects one in four older adults, and while ethicists warn that overreliance on AI could worsen isolation if not deployed responsibly, for Smola the technology has become “the best thing that happened to me,” offering comfort in the absence of regular human interaction.