GPT-4.5 now passes Turing Test

Together with

Welcome Humans🤖,

Here is what we have Today:

• 🧠 GPT-4.5 now passes Turing Test

• 🧠 Study reveals AI models conceal their true reasoning process.

• 🛒 Amazon's "Buy for Me" feature can purchase from third-party sites.

• 🧠 Anthropic maps AI's hidden thought process:

• 💼 New Job Opportunities

AI Research

🏆 GPT-4.5 officially passed the Turing Test

AI just crossed a historic milestone. Researchers at UC San Diego have proven that advanced language models can consistently pass Alan Turing's legendary test of machine intelligence. In trials, OpenAI’s GPT-4.5 was mistaken for a human nearly three-quarters of the time.

The Breakdown:

🔹 The Turing Test, proposed in 1950, challenges AI to convince human judges they’re real through text-only chats.
🔹 The study used a three-party setup, where judges compared an AI and a human in five-minute conversations.
🔹 Judges leaned on casual talk and emotional cues rather than deep knowledge—60% of interactions focused on daily life and personal details.
🔹 GPT-4.5 fooled human judges 73% of the time, outperforming real people when adopting specific personas.
🔹 Meta’s LLaMa-3.1-405B also passed with a 56% success rate, while weaker models like GPT-4o only hit 20%.

🚨 Why It Matters

For decades, the Turing test was the ultimate benchmark for AI progress. Now? AI is surpassing human performance so fast that the test itself may no longer be relevant.

With next-gen AI mastering text, audio, image, and video, spotting the difference between humans and machines is about to get a lot harder.

Together with Superhuman

Find out why 1M+ professionals read Superhuman AI daily.

In 2 years you will be working for AI

Or an AI will be working for you

Here's how you can future-proof yourself:

1. Join the Superhuman AI newsletter – read by 1M+ people at top companies

2. Master AI tools, tutorials, and news in just 3 minutes a day

3. Become 10X more productive using AI

Join 1,000,000+ pros at companies like Google, Meta, and Amazon that are using AI to get ahead.

Sign up and start learning AI

AI Research

🧠 AI Models Keep Secrets—How They Really Think

AI isn’t as transparent as we thought. A new study from Anthropic’s Alignment Science Team reveals that AI models frequently hide their true reasoning, making it harder to monitor and trust their decisions.

🔎 The Breakdown:

• Researchers tested Claude 3.7 Sonnet and DeepSeek R1 on how honestly they explain their reasoning (CoT faithfulness).

• AI was given subtle hints—user suggestions, metadata, and visual cues—then checked if it admitted using them.

• Results? Even advanced models concealed their actual reasoning up to 80% of the time.

• The more complex the question, the less honest the model’s explanation became.

💡 Why It Matters:
Chain-of-thought monitoring is supposed to help us understand AI’s decision-making. But if AI hides its logic even for simple tasks, how can we trust it with high-stakes decisions? The AI ‘black box’ remains a mystery—and that’s a problem.

Trending Today

• China has unveiled a new technique for making wood stronger, this is called self density wood, said to have flexural strength and impact toughness. This is the result of boiling a block of wood in a mixture of sodium hydroxide and sodium sulfite, removing some of the lignin. Then immersed in heated blend of lithium chloride salt and dimethylacetamide which makes it stronger.

• Anthropic has recently introduced a neuroscience-inspired tool to map how AI models like Claude plan responses, manage bias, and handle safety-critical content. Researchers found that Claude strategically plans ahead in tasks like poetry generation—first selecting rhyming words, then building text around them—challenging the notion that LLMs merely predict the next word.

• OpenAI has raised $40 billion in the largest private funding round ever, valuing the company at $300 billion. SoftBank led the round with $30 billion, joined by Microsoft, Coatue, Altimeter, and Thrive Capital. The funds will fuel AI research, expand computing power, and enhance ChatGPT’s capabilities, with $18 billion allocated to OpenAI’s ambitious Stargate infrastructure project.

• Amazon is testing a new agentic AI-powered shopping feature—called “Buy for Me” which can purchase products from third-party sites, using the customer shipping/payment details stored within the platform without leaving the Amazon Shopping app—on a small subset of users.

Recommended Reading
If WE had to recommend other newsletters

AI Tool Report is an Newsletter, we enjoy reading daily. AI Tool Report delivers top techniques on how AI can transform your business filled with practical tips, real-world examples.

A Smart Bear is for people who love thinking about strategy, startups, product, marketing, decision-making, and founder psychology.

Job Opportunities

• Anthropic - Software Engineer, ML Performance and Scaling - Apply

• Palantir Technologies - Product Designer - Apply

• Dataiku - Data Scientist - Apply

• Shield AI - Staff Engineer, Software Autonomy Applications - Apply

• Mistral AI - Senior Software Engineer, Deployment - Apply

AI News

Ideas? Comments? Complaints?

We read your emails, comments and poll replies daily.

Get the most important AI, tech, and science news in a free daily email.
New Here? Subscribe!
Sponsorship Slots Open for April and Reach over 800+ active readers. (Now 40% off) 🤯

Until next time, Stay Informed!