LLMs clear the Turing Test & more (LLMs clear the Turing Test & more (April 3, 2025 edition)
David Pawlan
Co-Founder
Hey friends,
Today’s AI news hits a little different—turns out the Turing test isn’t just theory anymore. We’ve got LLMs out-humaning humans, Meta bringing AI to the UFC, Claude going to college and the government, and Genspark unleashing a new super-agent that makes Manus look like a warm-up act.
Let’s get into it!
👇🔥 Big Story: LLMs Have Officially Passed the Turing Test
Researchers at UC San Diego confirmed that GPT-4.5 fooled judges 73% of the time in casual conversations, officially passing Alan Turing’s legendary test for machine intelligence.In some cases, the AI outperformed actual humans.
With AI agents rapidly advancing in text, audio, and video, we may soon live in a world where we can’t tell who—or what—we’re talking to.
🎓 Claude Goes to School & the State
Anthropic’s new Claude for Education helps students develop critical thinking with a Socratic-style “Learning Mode.” They’ve already partnered with Northeastern and LSERundown.
On the government side, Claude is now FedRAMP High and IL-2 compliant, making it available to U.S. federal agencies.
🕶️ Meta + UFC = AI Octagon Takeover
Meta and UFC announced a multi-year partnership to bring immersive VR and AI features into UFC events. Meta Quest will power fan experiences, Threads becomes the official social partner, and Meta’s branding hits the Octagon.
📞 Agents Just Got Superpowered
Genspark launched a general-purpose AI agent that smashes performance records, handling tasks like making calls and booking restaurants without any user input.
Meanwhile, a second startup by General Agents debuted a “real-time computer autopilot” that uses your mouse and keyboard to book things like Airbnbs—no APIs required.
🌍 AGI on the Horizon
Google DeepMind released a 145-page safety plan that predicts AGI—AI with human-level skills—could arrive by 2030. It flags major risks like “deceptive alignment” and critiques rivals like OpenAI for focusing too much on automating alignment and not enough on misuse.
⚙️ Productivity Tools of the Day
- Lindy – AI assistant for meetings, research, and outreach
- Findr – Unified search across your bookmarks, notes, and docs
- ModelMatch – Compare vision models, no code needed
- Focal – Make your own AI-generated TV show
- MoCha – Create cinematic AI characters from speech & text
🧠 Prompt of the Day
📚 Book Summarizer:
“Summarize the book [Book Title] in an engaging, easy-to-digest format. Highlight key insights, actionable lessons, quotes, and real-world applications. Suggest a few related reads too.”
💡 Quick Hits
- OpenAI’s PaperBench ranks Claude 3.5 Sonnet #1 at replicating academic research.
- Meta’s upcoming $1000+ Hypernova smart glasses will use gesture controls & neural wristbands.
- OpenAI formed an expert commission to advise on AI safety + nonprofit direction.
- ByteDance & Alibaba placed $16B in orders for Nvidia’s H20 chips.
- NotebookLM now lets you “discover” sources from across the web.
That's it for today!
If you made it this far, you’re exactly the type of person I love building with. Reply if you want to jam on any of today’s stories or if something cool is happening in your world—I’d love to hear it.
Till tomorrow,
—David
