• This “backpack” gives you six extra arms…
  • A “constitutional” artificial intelligence
  • The Allen Institute for AI enters the generative AI space

Dear Reader,

Was it another rapid unscheduled disassembly?

We’re not sure, but it definitely didn’t go as planned…

Presidential candidate and current Florida governor – go Panthers! – Ron DeSantis attempted to do something historical yesterday around 3 p.m. on Twitter Spaces. For those that aren’t familiar with Spaces, it’s a relatively new feature similar to Clubhouse that enables anyone to open up a “room” and have a conversation with as many people as you’d like.

Typically, a limited number of people are named speakers, and sometimes there is a moderator, and the rest of the “audience” is on mute but has the ability to raise their hand with a question. It is a modern, digital version of a town hall, which is why it makes so much sense for Twitter. After all, that has been Elon Musk’s goal all along, to preserve freedom of speech and provide a platform for everyone to be heard.

It appears that he was too successful in that mission.

Around 3:15 ET yesterday, there were already about 580,000 listeners to Ron DeSantis’ Presidential campaign launch. It was the largest in history for Twitter, or any other audio meeting platform for that matter.

The demand was so heavy, Twitter Spaces crashed. It just couldn’t take the load.

Naturally, the pundits were all over the event claiming it was a failure for both Musk and DeSantis; but from my perspective, that’s a very narrow view that misses the bigger picture.

I’ve worked in high tech my whole career, and I’ve never not seen technical issues as platforms scale. And when platforms are scaling quickly like Twitter is right now, the chances of technical issues are even more likely.

Google, Facebook, Netflix, Amazon, Apple, etc. have all had their moments with technical problems. All of them, several times. No matter how well these great companies plan for various scenarios, there is always a breaking point. And Twitter found it. Next time will be better. 

The more interesting shift at play here is that a presidential candidate chose to launch their campaign on a digital town hall as opposed to a more traditional broadcast medium. The reality is that trust in mass media and journalists who work at mass media outlets is at all time lows. All of the big players are suffering because people just aren’t listening to them much at all anymore.

Alternative media platforms are quickly filling that gap. Twitter is now a perfect example.

Rather than taking the risk of having an announcement or interview with a mass media outlet where the video will likely be edited in post-production, in an effort to frame someone in a certain light or remove comments that a media outlet feels don’t fit a political narrative… a digital town hall allows anyone to present their thoughts and ideas in a cohesive, complete manner.

Robert F. Kennedy, Jr. has been doing much the same. He’s been very active not only on Twitter, but also on a wide range of alternative media outlets. Despite being a Democrat, many of his thoughts, ideas, and sharing of scientific research are counter to the current political narrative. He has effectively been shadow banned by the Democratic Party despite being intelligent, well spoken, and speaking the truth about a wide range of matters. His voice should be heard unfiltered, and so should DeSantis’.

That’s how a free and democratic republic is supposed to work.

We’re midstream in an exciting upheaval of media and content distribution. This is a cultural shift as much as a societal shift. “We the people” want to hear what others have to say. And we don’t want their voices to be edited or censored. We want it straight from the horse’s mouth – good or bad.

And once these alternative media platforms have scaled to support 1 million, 5 million, even 50 million listeners, there will be little use for mass media at all. 

What could we do with six extra arms…?

A group of roboticists from the University of Tokyo developed a device that presents an entirely different vision of robotic human assistance from anything we’ve seen before.

The device is called “Jizai Arms.” The name means “freely moving” or “free motion” in Japanese. And it’s a strap-on backpack that can hold as many as six robotic arms – each with a somewhat dexterous hand at the end.

Check this out:

This is wild. The person in this video is demonstrating the arms in something of an artistic display, but the potential applications are almost limitless.

In truth, I found this to be a little disconcerting at first. It just looks… odd.

But then I realized – this could be an incredibly useful device.

The first thing I think about is augmenting human labor with these extra arms. I imagine workers in a warehouse or industrial setting whose job requires manual labor. Jizai Arms could certainly help.

For example, what about somebody working on picking and sorting products? Their job is to pick specific products out of different bins to organize the warehouse or get the products ready for shipping.

Well, with this device, a human worker could potentially have eight arms to work with… which means they could pick 16 different products out of bins at the same time. How’s that for a productivity boost?

Another good application would be surgery.

As it stands now, surgeons rely on nurses to hand them the various instruments they need throughout a surgical procedure. Jizai Arms could certainly make that process more efficient.

And then I have to think – what if we extrapolate a little bit here?

At the moment, the arms are controlled with a wireless controller. Not very sophisticated. But what happens if these robotic arms are augmented with generative artificial intelligence (AI) and computer vision? Suddenly the arms could “see” their environment. And they could “understand” what the human user was doing and what they need to get done.

In that scenario, the Jizai Arms could determine for themselves the best ways to assist the person using them. That’s an interesting thought.

Finally, what if we add a brain-computer interface (BCI) to the human user? This would enable anyone to control the robotic arms with their thoughts. That would allow quadriplegics to have functional arms and hands once again. The BCI would allow them to control the arms using their thoughts. The same would of course be true for anyone using the arms for productivity improvements.

So as strange as the Jizai Arms device is, I can see a lot of utility here. And it’s certainly a very different vision for robotics compared to the humanoid robots we’ve been researching.

Anthropic’s new and improved generative AI…

Anthropic just announced a big upgrade to its generative AI Claude. The company has improved the AI’s “short-term memory” in a way that has major implications…

We first highlighted Claude back in January. As a reminder, it is a direct competitor to OpenAI’s ChatGPT.

One of the ways Anthropic differentiated Claude initially is by referring to it as a “constitutional AI.” The team claims they wrote a series of principles that the AI will adhere to. This will arguably be one of the toughest hurdles for the industry in the years ahead. With a technology as powerful as generative AI, how will we use it in a way that is morally and ethically responsible?

Anthropic’s “constitutional AI” sounds like a step in the right direction. But Anthropic hasn’t released these principles to the public yet. So we can’t make any informed judgment on them.

But it seems constitutional principles aren’t the only differentiating factor. Anthropic just announced that Claude can ingest up to 75,000 words into its short-term memory. This is something ChatGPT cannot do yet.

Anthropic calls this “expanding the context window.” It means that Claude can take in vast amounts of information to work with.

In fact, they tested this out by feeding Claude the entire novel The Great Gatsby. In a matter of seconds, Claude was able to summarize the book and answer any questions on it.

That’s interesting. But it’s just scratching the surface…

What this means is that Claude can now take in huge amounts of information so that we don’t have to.

For example, I imagine many companies will feed Claude their technical manuals and product documentation papers. These things are incredibly boring to read, and usually we just need a few specific bits of information… but the AI could ingest all the information in seconds. Then it would be an expert on each product… and it would have perfect recall ability. We could imagine how useful that would be for customers looking to get answers on a technical question.

Or, for those of us working in financial services, what if we had Claude pour through reams of financial statements and SEC filings? Or for those in the medical research field – Claude could pour through mounds of scientific papers. Then it would be able to summarize everything for us and answer very specific questions.

And what about legislation?

We know that most pieces of legislation contain hundreds of pages of legalese. And often, there are hidden items buried in there that are materially important. I don’t think many congresspeople even read these things.

Well, Claude could ingest every piece of legislation in seconds. Then it could summarize everything, highlight what’s most relevant, and root out the contradictions buried within.

Simply put, an AI like this has immense utility. This is something that will be adopted rapidly.

Claude is still in closed beta testing right now. But I have to think it will be available to the general public soon. When that happens, it’s going to spread like wildfire.

The value of this hasn’t been missed by the venture capital community. Anthropic just raised $450 million at a $5 billion valuation. Incredible for a company that is just about two years old.

The organization setting up to achieve what OpenAI was supposed to…

We’ve been tracking the progress of the Allen Institute for AI for a few years now. It was founded by the late Paul Allen, who was also a founder of Microsoft. The organization has been a key player in the industry for producing AI-related research as a non-profit entity. And it just introduced its own generative AI open language model. They call it AI2 OLMo. It will operate much like Claude and ChatGPT.

But here’s the thing – the Allen Institute plans to make its generative AI openly available to the entire scientific community. For free.

What’s more, the Institute will release everything – the code, the model, all training curves, evaluation benchmarks, training data, you name it. They are making everything open source and fully transparent.

Sound familiar?

This is remarkably close to what OpenAI originally set out to achieve in its early days as a non-profit. But somewhere along the lines, OpenAI established a for-profit entity and shifted its focus towards those efforts. This is what caused Elon Musk to leave the organization that it originally founded.

The crucial distinction is that the Allen Institute has stayed true to its non-profit roots. Meanwhile, OpenAI is now primarily a for-profit company.

This new contribution from the Allen Institute for Artificial Intelligence will be an invaluable resource for researchers and computer scientists. They can leverage this work and use the generative AI just as it is. Or they can dig into the code and modify it to suit their specific applications.

This is great to see.

The Allen Institute has a reputation for high-quality work. This will be one of the most credible license-free AI models in the world. While some of the major AI-related semiconductor companies like NVIDIA, Graphcore, and Cerebras have already done something similar, the Allen Institute will be neutral in terms of computing platform. And I’m sure it will be employed for a wide array of research projects and services.

So this is another big development in the generative AI space. And the Allen Institute expects AI2 OLMo to be ready early next year. It’s only seven to nine months away, and it will empower governments, enterprises, as well as individuals to have free and open access to some of the most advanced artificial intelligence available. With technology releases like this, there are almost no barriers to entry for building AI.


Jeff Brown
Editor, The Bleeding Edge