The week in AI: Neuralink livestreams first human using Telepathy brain implant to control a computer

Sponsored by

Welcome to The Dispatch! We are the newsletter that keeps you informed about AI. Each Thursday, we aggregate the major developments in artificial intelligence; we pass along the news, useful resources, tools and services, and highlight the top research in the field as well as exciting developments in open source. Even if you aren’t an engineer, we’ll keep you in touch with what’s going on in AI.

NEWS & OPINION

Neuralink showcases first human brain-chip patient playing online chess

-------------------------

A few weeks ago, Elon Musk’s Neuralink announced that it had implanted its first chip in a human brain. Now Neuralink has showcased a nine-minute livestream on X of this first patient moving a cursor on a computer using an implanted device (Neuralink named the device ‘Telepathy’) to play online chess.

Noland Arbaugh was paralyzed below the shoulders after a diving accident and is still exploring how he can use the implant effectively. “I’m so freaking lucky to be a part of this,” Arbaugh stated in the video as he moved chess pieces on a digital board. “Every day it feels like we’re learning new stuff, and I just can’t even describe how cool it is to be able to do this.” He also noted that in one of the first sessions he was able to start using Telepathy, he stayed awake until 6 in the morning playing a popular computer game called Civilization.

Neuralink has faced public and regulatory scrutiny for its animal experimenting practices.

AI pioneer Mustafa Suleyman leaves Inflection to join Microsoft, named CEO of “Microsoft AI”

-------------------------

On Tuesday, Microsoft CEO Satya Nadella announced that British AI pioneer Mustafa Suleyman had joined the company and would lead the newly announced “Microsoft AI” organization. Suleyman co-founded both DeepMind (now Google DeepMind) and Inflection AI. As CEO of the new arm, Suleyman will lead the development of Copilot and head Microsoft’s AI operations more broadly. The new Microsoft AI unit will initially comprise the Copilot, Bing, and Edge AI teams as well as Microsoft’s current Gen AI team.

Microsoft also hired most of Inflection’s employees, including chief scientist Karén Simonyan. Just last week, we highlighted Inflection’s newly upgraded and excellent AI chatbot, Pi - now, much of Inflection’s talent will be absorbed into Microsoft and the upgraded Inflection-2.5 LLM that powers Pi will soon become available on Microsoft Azure.

The New York Times suggests that regulators will scrutinize the Inflection acquisitions, but otherwise it looks like another major AI-era win for Microsoft.

Inflection will carry on with a new business model.

The AI world is gathering at Nvidia’s GTC 2024 event this week

-------------------------

Shares of Nvidia have surged 260% over the last 12 months on the AI boom and launched the chipmaker into the top tier of the world's most valuable companies. And they show no signs of slowing down. The tech giant is hosting a massive AI conference as part of its GTC (GPU Technology Conference) event this week.

Here are the major developments from the conference:

• At the keynote address on Monday, CEO Jensen Huang announced Nvidia's next-generation chip architecture called Blackwell and related products, including the company's latest AI chip called the B200. Full performance details have yet to be divulged, but Blackwell is between 2 and 30 times faster (depending on the measurement) than the previous generation of AI-optimized GPU’s from Nvidia called Hopper. A Blackwell GPU will cost $30,000-$40,000

• The company was already a major player in robotics, but it doubled down with new tools for roboticists to make their robots smarter.

• They’re facing scrutiny for announcing collaboration with a healthcare company that offers generative AI nurses who work for just $9 an hour. Hippocratic promotes how it can undercut real human nurses, who can cost $90 an hour, with its cheap AI agents that offer medical advice to patients over video calls in real-time.

• They announced Nvidia NIM, a new software platform designed to streamline the deployment of custom and pre-trained AI models into production environments. Typically, it could take developers weeks or months to ship similar containers, Nvidia argues - and that is if the company even has any in-house AI talent.

• Huang kept repeating the phrase “AI factory” as an alternative to data centers - “There’s a new Industrial Revolution happening in these [server] rooms: I call them AI factories,” he said. “The raw material that goes in is data and electricity. What comes out of it is data tokens. The token is invisible and will be distributed all over the world. It’s very valuable.”

Nvidia is also getting involved with combatting extreme weather, industrial automation, and Omniverse for collaborative 3D content creation and simulation. Check out the session catalogue for the event (there are over 1,000!) if you’re interested in a deep dive on even more developments going on under Nvidia’s AI umbrella.

• 'Materially better' GPT-5 could come to ChatGPT as early as this summer

• Apple is reportedly exploring a partnership with Google to bring Gemini to iPhones

• Intel to make world's largest AI chip factory in the US, spend over $100 billion across four states

• French regulators fine Google $270m over after using news outlets' content to train Gemini

• YouTube adds new tool requiring creators to disclose to viewers when content is AI-generated

• As AI tools get smarter, they’re growing more covertly racist, experts find

• Ghostbots: AI versions of deceased loved ones could be a serious threat to mental health

• ‘A landmark moment’: scientists use AI to design antibodies from scratch

Missed out on Ring and Nest? Don’t let RYSE slip away!

Ring 一 Acquired by Amazon for $1.2B

Nest 一 Acquired by Google for $3.2B

If you missed out on these spectacular early investments in the Smart Home space, here’s your chance to grab hold of the next one.

RYSE is a tech firm poised to dominate the Smart Shades market (growing at an astonishing 55% annually), and their public offering of shares priced at just $1.50 has opened. 

They have generated over 20X growth in share price for early shareholders, with significant upside remaining as they just launched in over 100 Best Buy stores.

Retail distribution was the main driver behind the acquisitions of both Ring and Nest, and their exclusive deal with Best Buy puts them in pole position to dominate this burgeoning industry.

Secure your stake before the offer closes.

TECHNICAL, RESEARCH & OPEN SOURCE

Apple’s MM1 research: methods, analysis & insights from multimodal LLM pre-training

-------------------------

Apple has quietly released MM-1, a series of multimodal AI models that scale up to 30B parameters. The MM-1 research paper is very comprehensive, and the Apple team has shared insightful findings from pre-training to fine-tuning, architecture components, and data selection strategies.

MM1 is multimodal and trained on images as well as text. This allows the model to respond to text prompts and also answer complex questions about particular images. One example in the research paper shows what happened when MM1 was provided with a photo of a sun-dappled restaurant table with a couple of beer bottles and also an image of the menu. When asked how much someone would expect to pay for “all the beer on the table,” the model correctly reads off the correct price and tallies up the cost.

It’s hard to draw too many conclusions about Apple’s plans from the research paper alone, but it’s the biggest sign yet that Apple is developing frontier generative AI capabilities of their own.

Elon Musk’s xAI open sources Grok-1

-------------------------

Last Monday, Twitter/X owner and xAI founder Elon Musk announced that Grok would soon go open source, making its source code freely available for anyone to access and modify. The company has made good on that plan, releasing Grok-1's base model weights and network architecture on GitHub.

xAI stated that the model being released is from when Grok-1 was in its "pre-training phrase" in Oct. 2023. So while this base model has been trained on an massive amount of text, it hasn't been fine-tuned to any specific task and will not sound or act exactly like the snarky Grok model available for use with a paid X/Twitter account. Grok-1 is a 314B parameter model with a Mixture of Experts (MoE) architecture. It was released under the Apache 2.0 license, meaning anyone is free to modify and redistribute the weights and architecture.

In an apparent reference to 1989 sci-fi comedy Bill & Ted's Excellent Adventure, the code of conduct accompanying Grok's release reads only: "Be excellent to each other."

MORE IN T/R/OS

• What I learned from looking at 900 of the most popular open source AI tools

• MindGraph: a proof of concept, open-source, API-first graph-based project designed for natural language interactions (input and output)

• Introducing Stable Video 3D: Quality Novel View Synthesis and 3D Generation from Single Images

• VLOGGER: Multimodal Diffusion for Embodied Avatar Synthesis

That’s it for this week! We’ll see you next Thursday.