Gustavo’s Corner: AI News for CFOs - #7
This edition of Gustavo’s Corner: AI News for CFOs captures a moment where AI progress is accelerating across consumer tools, enterprise platforms, and core infrastructure, while underlying risks and costs are becoming more visible. From real-time translation and next-generation models to research gaps and shifting corporate strategies, the news reflects how AI is moving from experimentation into everyday business reality.
Google rolls out real-time translation for any headphones. Google Translate is being upgraded with Gemini-powered, real-time audio translation that works with any pair of headphones. The beta feature is designed to preserve a speaker’s tone, emphasis, and cadence, making translated conversations sound more natural and closer to how people speak.
AI struggles to separate facts from beliefs. Researchers have found that even advanced language models often fail to distinguish objective facts from subjective beliefs, frequently correcting false beliefs instead of simply recognizing that a person holds them. This behaviour could introduce real risks in sensitive domains such as healthcare, law, and mental health. Using a new benchmark called KaBLE they tested 24 AI models on 13,000 questions and observed sharp accuracy drops when models dealt with first-person false beliefs. They conclude that modern AI systems exhibit a strong “corrective bias,” driven by training methods that prioritize factual accuracy above all else. Researchers argue that before AI is deployed in high-stakes, human-centred settings, future models must better separate truth from belief rather than reflexively correcting users.
OpenAI fires back at Google with GPT-5.2. The launch of GPT-5.2, its most advanced model to date, offers three variants; Instant, Thinking, and Pro, targeted at developers and professional users. OpenAI highlights major improvements in coding, reasoning, long-context comprehension, and tool use as competition with Google’s Gemini 3 intensifies. GPT-5.2 is a consolidation of recent upgrades that now lead many reasoning benchmarks, but the model also brings higher compute costs. OpenAI’s growing infrastructure spending raises questions about long-term sustainability as it tries to keep pace with Google’s deeply integrated AI ecosystem.
Oboe raises $16M from a16z to scale its AI learning platform. It offers a generalized learning platform that turns virtually any subject into a structured educational experience. Oboe combines tailored visuals and clear explanations to break down complex technical topics into accessible, step-by-step lessons. It recently raised $16M to continue expanding and refining its platform, signaling strong investor confidence in its approach to AI-powered education and its potential to reshape how advanced concepts are taught.
Meta faces internal confusion as its AI strategy shifts. Its earlier focus on open-source Llama models has fragmented as the company pivots toward a new proprietary system called Avocado. The model has been delayed until early 2026, raising doubts internally following leadership changes and the weaker-than-expected performance of Llama 4. Meta is sharply increasing spending, including a reported $14.3B hiring deal for Alexandr Wang and higher capital expenditures planned for 2025. These moves reflect Mark Zuckerberg’s urgency to catch up with OpenAI, Google, and Anthropic, even as investors question the company’s direction and potential returns. Its employees are feeling the strain as teams face long hours, reorganizations, and layoffs while products like Vibes lag behind competitors. Meta’s absence from top-model discussions has intensified pressure on Avocado to deliver a meaningful breakthrough and restore confidence in the company’s AI roadmap.
Why this news is important to CFOs and their teams:
These developments reinforce the need to approach AI with both ambition and discipline. Breakthroughs from Google and OpenAI point to productivity gains and new use cases, but also to rising vendor dependency and compute expenses that demand tighter financial oversight. Research showing AI’s difficulty in separating facts from beliefs highlights governance, compliance, and reputational risks if systems are used in sensitive finance, legal, or people-facing processes. At the same time, Oboe’s funding and Meta’s strategic uncertainty illustrate the uneven returns and execution challenges behind large AI investments. Finance leaders must therefore guide their organizations to invest selectively, set clear guardrails, and build internal literacy so AI adoption delivers measurable value without exposing the business to unnecessary
—
/Gustavo
