• AI Valley
  • Posts
  • Meta Robotics + Mind Reading AI

Meta Robotics + Mind Reading AI

PLUS: Perplexity launches 'deep research'

Author

Barsee
17 Feb

Together with

Howdy. It’s Barsee again.

Happy Monday, AI family, and welcome back to AI Valley.

In today’s edition:

  • Meta just achieved AI mind-reading with 80% accuracy

  • Grok 3 is set to launch today

  • Plus trending AI tools, posts, and resources

Ready, set, go…

META AI

Meta just achieved AI mind-reading with 80% accuracy

Meta has collaborated with international researchers to develop Brain2Qwerty, an innovative AI model capable of reconstructing sentences from brain activity with up to 80% accuracy.

Here's what you need to know:

  • The research builds on previous work from French neuroscientist Jean-Rémi King, which focused on decoding visual perceptions and language from brain signals.

  • Unlike traditional brain-computer interfaces that require invasive implants or surgery, the team used non-invasive techniques such as MEG and EEG to record brain signals from 35 healthy volunteers as they typed, training an AI to reconstruct their sentences.

  • When tested on new sentences, Meta claims that MEG-based decoding accurately predicts up to 80% of characters, with an average error rate of 32% (some subjects achieved 19%), compared to about 8% for professional transcribers and below 5% for invasive systems like Neuralink.

  • Beyond decoding thoughts into text, Meta’s AI helps to analyze how the brain turns abstract ideas into movements. By capturing 1,000 snapshots per second of the brain, it pinpoints the exact moments when thoughts transform into words, syllables, letters, and even the precise finger movements used for typing.

Why it matters: 

Meta's new AI is not just a technical breakthrough but a potential lifeline for those with speech or motor impairments. Although it’s still early days and the technology faces challenges like real-time use and hardware limits, this research brings us closer to a future where everyone can communicate freely and be understood.

TOGETHER WITH FLOW AI

Work at the speed you think with Wispr Flow

Meet Wispr Flow: the dictation tool that sounds just like a more polished version of yourself—minus the "ums" and "uhs." Whether it's an email, doc, or a prompt, Flow transcribes your words instantly. It's 3x faster than typing and automatically cleans up mistakes. That's time back in your day—and fewer embarrassing typos.

SIDE UPDATES

Meta has launched a new division within its Reality Labs to develop both hardware and software for robots designed to perform household chores. Instead of creating its own branded robots, Meta is partnering with companies like Unitree Robotics and Figure AI to manufacture and sell them. The company also plans to integrate data from its AR and VR products to further advance its robotics capabilities. While prototypes are in the works, it may take years before these platforms hit the market.

Perplexity has launched Deep Research, an advanced tool designed to conduct iterative searches, analyze hundreds of sources, and generate in-depth reports in minutes. It significantly outpaces OpenAI’s Deep Research, which takes 5 to 30 minutes per task, and has outperformed competitors like Google’s Gemini Thinking in benchmark tests. The tool is free with a daily query limit, while a premium plan offers up to 500 queries per day for $20/month.

Sam Altman and Elon Musk co-founded OpenAI in 2015, but their partnership unraveled in 2018 when Musk departed after a power struggle. The rivalry escalated further when Musk launched xAI, directly competing with OpenAI. Most recently, he made a hostile $97.4 billion bid for OpenAI’s controlling nonprofit entity. This deep dive explores their turbulent relationship through insights from those close to the duo.

Anthropic is set to release its latest AI model, possibly Claude 4, in the coming weeks. Described as a "hybrid model", it can switch between deep reasoning and rapid responses. It is said to outperform OpenAI's o3-mini-high on certain programming tasks and excels at analyzing large codebases and business benchmarks. The new model will also feature a "sliding scale" function, allowing developers to manage computing costs associated with its advanced reasoning capabilities.

Sam Altman recently hinted at major upgrades to GPT-4o, and users are already noticing the difference. The model now not only seems smarter but also shows greater personalization and emotional understanding. Its previously rigid persona has been replaced with more engaging character that speaks in a slightly spicy tone, making conversations feel more real.

GROK 3

Grok 3 is set to launch today

Elon Musk announced via X his plans to launch its latest chatbot Grok 3 via live demonstration on Monday, at 8 PM Pacific Time, describing it as "the smartest AI on Earth".

Here's what you need to know:

  • Musk first revealed Grok 3's launch plans at the World Government Summit in Dubai on Thursday, highlighting its superior reasoning capabilities, claiming it surpasses all competing AI models released so far.

  • The Grok 3 model will feature a "Chain Of Thought" reasoning capability, allowing it to process complex tasks step-by-step like human, improving its ability to handle intricate queries and provide logical responses.

  • It is built on a mix of synthetic and real-world data, allowing it to achieve logical consistency by repeatedly reviewing data and correcting its own mistakes.

  • Musk also hinted that Grok 3 will see major improvements in reasoning, programming, and multimodal capabilities like text and image analysis.

  • Although initially planned for release by the end of 2024 with 100k H100 cluster training and 20x the compute of Grok 2, the launch was delayed but is now set for today's live demonstration.

Why it matters: 

The launch of Grok3 comes as global competition in the AI chatbot market intensifies, with companies like DeepSeek gaining traction. The imminent release of Grok3 not only represents the significant step for xAI in the AI field but also marks a new starting point for Musk in his rivalry with competitors.

TRENDING TOOLS

  1. Talo > Real-time AI translator for video calls.

  2. Tough Tongue AI > Use AI Agent to generate, share, and rehearse difficult conversation scenarios.

  3. Learn Copywriting > Get your copywork scored by an LLM.

  4. Github Copilot Agent Mode > Coding autonomously.

  5. Rabbithole > Turns questions into interactive mind maps of connected discoveries.

THINK PIECES / RESOURCES

CONTENT CORNER

1/

2/

3/

4/

5/

THAT’S ALL FOR TODAY

That’s all for today’s issue, folks.

💡 Help me get better and suggest new ideas at [email protected] or @heyBarsee

👍️ Like what you see? Subscribe here

Thanks for being here.

REACH 100K+ READERS

Acquire new customers and drive revenue by partnering with us

Sponsor AI Valley and reach over 100,000+ entrepreneurs, founders, software engineers, investors, etc.

If you’re interested in sponsoring us, email [email protected] with the subject “AI Valley Ads”.