• Synthetic
  • Posts
  • New AI 'Brain Scanner' Begins to Reveal AI's Mysteries 🧠

New AI 'Brain Scanner' Begins to Reveal AI's Mysteries 🧠

Plus, What Fully Automated Firms Will Look Like ⚙️

Author

Steve Brown
April 03, 2025

In partnership with

Subscribe to Synthetic

The week’s most interesting and relevant AI news and analysis

This Week in AI

Source: Nature Neuroscience, 2025

Researchers at the University of California, Berkeley and San Francisco built a brain-computer interface (BCI) to enable a 47-year-old stroke victim to speak. The patient, who suffered a brainstem stroke almost two decades ago, now has a vocabulary of over 1000 words thanks to deep learning networks that monitor her sensorimotor cortex and translate silently ‘spoken’ words into audio. The new research was published in Nature Neuroscience. 🧠

While Apple is (rightly) being criticized for being slow to the AI game, it is rumored to have several big AI projects underway inside its secret labs. One project, codenamed Project Mulberry, is a significant AI-powered expansion of the Apple Health app and may be sold as Apple Health+. The AI agent will collect data from devices and wearables and recommend ways to improve your health. Sources indicate that Health+ will use your iPhone’s camera to assess your workout. 🏋️‍♂️

This week, Anthropic launched Claude for Education, a specialized version of their leading-edge chatbot tailored for higher education. Anthropic is inking campus-wide access agreements for their AI services and has launched a new “Learning Mode” for Claude that guides students to reason through problems and develop critical thinking skills rather than providing answers. 🎓

Synthetic’s Take: Educators must adapt, embrace change, and encourage their students to use AI tools or risk being made irrelevant. Education hasn’t changed much in 250 years. At a time when education is too expensive, inaccessible to too many, and is burdening students with decades of debt, we must reimagine education for the AI era. More of Synthetic’s thoughts on the future of education can be found here.

Quick Hits

Video: Vinod Khosla’s Predictions on our AI Future 🔮

This week, Synthetic is showcasing another interview with Vinod Khosla, this time speaking with Peter Diamandis at the Abundance 360 Summit. The 30-minute video includes several short ads from Peter Diamandis that you should fast forward through. ⏊

Interesting quotes, insights, and predictions from this video include:

  • One billion bipedal robots by 2040, and that’s likely an underestimate. These robots will do more work than the entire human population today, and will work 24/7.

  • Robots could cost $10,000 and be as essential as a smartphone.

  • Bipedal robots will be a bigger business than the auto industry in 20 years.

  • Expertise will be free. AI tutors, doctors, engineers, and accountants will work 24/7.

  • In the next 5 years, each person should be supervising 5 AI interns to expand their impact.

  • Ads will become less about emotional persuasion and more informational to appeal to AI buying agents.

  • A helpful prompt for an AI: “How would Steve Jobs answer this question?”

  • “In the future, instead of humans having to learn computers, computers will learn humans.”

  • Most cars in most cities will be autonomous by 2050.

  • By 2030, nobody will debate whether fusion energy is economically viable. Net energy production will be demonstrated by 2027.

  • Old coal plants will be retrofitted and repurposed with fusion boilers rather than coal.

AI Tech and Innovation

One of the main challenges of large language models (LLMs) is that even though humans made them, we don’t fully understand how they work. In new research, Anthropic has developed a ‘brain scanner’ to trace the ‘thoughts’ of LLMs and explore how they work. Their insights might surprise you and show why LLMs are so terrible at math, how they rhyme poems, and why they hallucinate. The research also revealed that LLMs aren’t just predicting the next word in a sentence.

Joseph Coates was diagnosed with POEMS syndrome, a rare blood disorder. He was told there was no cure and was headed for hospice. Rare diseases affect millions worldwide but often lack approved treatments. An AI model developed by Dr Fajgenbaum suggested a unique cocktail of existing drugs that turned his health around. Drug repurposing—using existing approved drugs for new diseases—is a promising approach due to the lack of pharmaceutical interest in developing new therapies for rare conditions. Could AI help us find cures for diseases already hiding in plain sight?

Google’s AI research lab, Google DeepMind, has been holding back some of its world-leading AI research. Historically, DeepMind and Google Brain published all of their research, including Google’s seminal paper on transformers that kicked off the generative AI age. Researchers grumble that their careers are harmed if they can’t publish research and get recognition. But….Google is in a race with well-funded competitors (OpenAI, xAI, Anthropic, Amazon, Meta, and DeepSeek) and can no longer afford to give away their research, so they have adapted their strategy.

Synthetic’s Take: This is a natural evolution for Google DeepMind and has been discussed internally for some time. Many researchers come from academia, where publishing research is the lingua franca of the trade and success is measured by the volume and quality of their research papers. However, different rules apply in business, especially at a competitive, high-salary company like Google. While there is ongoing debate over open-source collaboration versus home-grown innovation, a balanced approach makes sense—share strategically but hold secret sauce close.

AI Insights

❝

“The ability to turn capital into compute and compute into equivalents of your top talent is a fundamental transformation.”

Dwarkesh Patel, Podcaster

Imagine taking the top ten most experienced and talented people in your company and cloning them to fill every other role. How would that change your growth trajectory and competitiveness? Dwarkesh Patel argues that once human-level AI arrives, it will profoundly transform companies, not just because it affords the potential for everyone to have a super smart 24/7 AI assistant but because digital employees (agents) can be copied, distilled, merged, scaled, and evolved in ways humans can’t. What might a fully automated company look like where every employee and every manager is an AI?

Synthetic’s Take: By 2030, perhaps companies won’t be limited by finding rare talent but by how much computing power they can afford.

A study of 30,000 manufacturing firms found that early adopters of AI experienced a short-term productivity drop due to disruptions in established practices. However, these firms eventually outperformed in sales growth, productivity, and employment, suggesting that AI adoption can lead to long-term success.

❝

AGI is a ‘country of geniuses in a data center’

Dario Amodei, CEO and co-founder of Anthropic

Google DeepMind is calling for urgent AGI (human-level, artificial general intelligence) safety research with a focus on four main areas:

Google DeepMind’s areas of concern for AGI safety

In their 145-page paper, Google DeepMind calls for urgent conversation and regulations to limit risks and harm. You can learn more here.

Toolkit for the Future

This end-to-end, AI-powered hiring platform automates everything from drafting job descriptions and advertising jobs to screening, shortlisting, scheduling, interviewing, and hiring candidates. Megan AI is the first human-quality AI recruiter and hiring manager fully integrated with the entire hiring process. Make hiring decisions with 78% less busy work.

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.

Focus on the email that matters. Deep Clean automatically identifies old, unimportant emails so you can delete them. Snooze emails until a better time and easily track emails that need a response. Automatically filters unimportant emails into specific folders. It works across every device where you check your email. Get to inbox zero with the award-winning SaneBox.

Recording, editing, and refining audio demands more time than most entrepreneurs can spare. Generate voices, high-quality music, and sound effects efficiently and cost-effectively with a single click. Create multilingual voice tracks in up to 29 languages to reach new markets. Try it for free.

Find phone numbers, emails, and company org charts to reach the people you need to. Fill information gaps in your contacts, auto-generate follow-up emails, and identify new target markets. Get 50 free credits and discover why over 500,000 companies use Seamless to grow their business.