Loading page…
Company intelligence / Live profile
Track Scale AI's role in the AI economy through Superpower Daily reporting on business, models, and tools. This live company page updates automatically as new launches, partnerships, product moves, and market signals are published.
Latest coverage
Key developments
Its insurance copilot improved accuracy and cut running costs in synthetic-call tests. Real-call evaluation and safeguards remain unfinished.
The new DistressBench evaluation covered 25 models. Longer conversations exposed further weaknesses, while the effectiveness of referrals remains an open question.
Humanity’s Sixth Sense tests what images and videos imply, not just what they show. Social understanding was the weakest area for 21 of 25 models, though those scores measure agreement with human interpretations.
The new reference architecture lays out how companies can build, govern and distribute one agent across business applications and Gemini Enterprise without rebuilding it for each interface.
The study found lower harmful-response scores in Korean prompts, but also more refusals of harmless requests—complicating claims that a model is simply safer in another language.
The company says its pre-deployment system found more recoverable vendor credits by reserving analyst time for decisions where a mistake would be costly and hard to undo.
In two company-published case studies, testing whole AI systems—not just their models—uncovered violations that automated checks largely missed. The findings also show that fixes can lower failure rates without eliminating the need for retesting.
The missing links are not proof that any result was gamed. But after answer-leakage and task-quality concerns, buyers and researchers cannot tell which exact setup produced a displayed score—or whether it predates remediation.
Connected coverage
Your company watchlist
