IA
Note sull'episodio
The AI that scores 94% on a PhD physics exam
will probably write the most robotic email
you've ever read. Welcome to the AI paradox of 2026.
ChatGPT Plus and Perplexity Pro — same $20/month.
But they work in completely opposite ways.
In this episode:
• Why Gemini tops every benchmark — but finishes
dead last on real tasks (GDPval Elo: 1317 vs 1633)
• GPT-5.4 Thinking: the only AI that shows you
its blueprint before it answers
• 1M tokens = 1,500 pages: when does context
window actually matter?
• 2x more Deep Research for the same $20 —
who gives more and why
• ChatGPT Memory vs Perplexity Spaces: one
remembers YOU, the other remembers your documents
• New Finance Focus Mode in Perplexity: pure
SEC data, zero Reddit noise
Parole chiave
ChatGPTChatGPT PlusGPT-5.4 ThinkingClaudeClaude SonnetGeminiGemini 3.1 ProAI comparison 2026AI benchmarksAI limits