• A new “world’s largest semiconductor campus” takes shape
  • A new AI unicorn is born
  • Facebook quietly scraps its support for NFTs

Dear Reader,

In all of the excitement around generative artificial intelligence (AI) that’s happening right now, there’s a lot of nuance that is missed concerning this incredible technology.

It’s no one’s fault. 

After all, the sheer simplicity of being able to converse with a powerful large language model (LLM) like GPT-4 through a simple interface like ChatGPT is fantastic. It’s a feeling that most of us haven’t had since the first time that we picked up an Apple iPhone. It’s novel, intuitive, very cool, and immensely useful.

And yet, it has limitations.

LLMs are trained on a massive body of knowledge, most of it from the internet. There is a lot of work for the industry to do here…after all, garbage in, garbage out. 

The very best LLMs will have curated training data of the highest quality information, preferably the kind without political bias of any kind.

Another nuance is that LLMs aren’t trained on real-time data. Said another way, they aren’t “up to date” in the way that we think of the internet being a real-time source of information.

Also, LLMs are designed to output information only in the form of text. That can come in the form of language, or in the form of computer code written to achieve a desired outcome.

But this begs the question, how do these powerful generative AI models like GPT-4 make the leap into the real world? Rather than just outputting useful text or code, how ca we enable these LLMs to perform functions in order to get stuff done for us?

Days ago, OpenAI made yet another major announcement around its support for ChatGPT plugins. To most of us that won’t sound like a big deal, but it is…

An easy way to think about what this means is to imagine plugging in specific real-time data into a powerful generative AI like GPT-4. GPT-4 doesn’t train on the real-time data, but it can understand, optimize, and act based on that real-time information.

Here are some simple examples of software applications that OpenAI announced having plugins for ChatGPT:

  • Expedia can be used to book travel

  • Instacart can be used to order groceries from a local store

  • OpenTable can be used to make reservations and recommendations

  • Shopify can be used to search and find products from millions of options

Yes, we can use and enjoy these applications now; but what this means is that ChatGPT will be able to interact and eventually use these applications on our behalf. 

It will be empowered to do the searching, analysis, and work for us and present solutions… all in a matter of seconds.

For example, just imagine feeding ChatGPT with a list of five dinner recipes and ChatGPT will do all the shopping for us through Instacart. Or perhaps we’ll give it the task of finding flights and hotels on certain dates with any key parameters and ChatGPT will figure it all out for us in seconds.

And these latest developments aren’t just a limited functionality designed for a handful of useful software applications. OpenAI has structured this in a strategic way. It’s trying to position itself in as dominant of a competitive position as possible with this app store-like model for ChatGPT plugins.

The evidence speaks for itself. A few days ago a developer managed to hack the ChatGPT API and discovered a long list of plugins that haven’t yet been announced.

It includes a list of certainly soon-to-be-released plugins for the NBA, Slack, Vogue, Crypto Prices, TODO, and so many others. I’m sure that the below logos are already out of date as well.

Likely Future Plugins for ChatGPT

Source: Twitter (@rez0_)

We should consider this as just an indication of what’s to come…. an absolute avalanche of thousands of plugins for just about anything we can imagine.

This is how an artificial intelligence crosses the bridge between intelligent, useful output to being able to perform functions based on real-time data in the real world.

All of this is of course software based. We can’t see it, feel it, or touch it; but it will appear like magic.

But what comes next will be something that we can see, feel, and touch. GPT-4 and its competitors are becoming multi-modal. That means that the AIs can receive inputs like images and video streams.

And that means that when we use computer vision and feed a real time video stream into an artificial intelligence, this bleeding edge software will be perfectly suited to “run” on hardware that operates in the real world – robotics.

This could be something as simple as a cleaning robot in our home or office and something as advanced as a humanoid robot along the lines of what Tesla is building with Optimus.

And it’s coming fast. 

We’ll start to see deployments of this kind of technology for commercial applications in the second half of this year.

I know it’s hard for us to get our heads around what this all means. It is changing and advancing on a daily basis right now, making a firm understanding nearly impossible. 

But for those who might want to get a glimpse of the future and think some more about how the world will look different, I recommend a classic – The Robot Series of books from Issac Asimov.

Samsung’s timing wasn’t a coincidence…

South Korean conglomerate Samsung just made a major announcement.

Most people know Samsung as a consumer electronics company. I’m sure many readers own one of Samsung’s smartphones or televisions.

But this company is much more than just a consumer electronics player. One of its many subsidiaries is one of the largest and most important semiconductor companies in the world.

And Samsung just revealed that it is building a $230 billion semiconductor manufacturing campus in South Korea. It will consist of at least five separate production facilities.

Once complete, this will be the largest semiconductor campus in the world.

And the timing of the announcement isn’t a coincidence.

If we remember, last year Intel announced a $100 billion plan to build out a semiconductor hub in Ohio which would have been the largest semiconductor manufacturing location on the planet. This was very much driven by the $52 billion “CHIPS Act” passed by the U.S. Congress last year. The announcement was positioning for large government subsidies from the CHIPS Act to offset part of its costs.

Samsung clearly wasn’t going to sit on the sidelines. It went bigger, by more than twice the capital commitment.

This is certainly a reflection of South Korea’s intention to at a minimum to continue to be one of the top semiconductor hubs for the foreseeable future. And given the scale of this investment, it suggests that Samsung has a target to overtake TSMC as the world’s largest semiconductor foundry company.

As we know, TSMC is headquartered in Taiwan. But there are a lot of concerns about TSMC’s stability and independence given the geopolitical aspirations of China with respect to Taiwan’s sovereignty.

In fact, this geopolitical risk prompted Warren Buffett’s Berkshire Hathaway to sell almost its entire stake in TSMC earlier this year after holding for just a matter of months.

This was a smart move by Samsung given the current environment regarding Taiwan. Depending on what happens, Samsung may be given a window of opportunity where it may have a stronger competitive advantage.

And with the U.S. government offering subsidies for bringing production back on shore, U.S. semiconductor companies have shifted their manufacturing strategies to benefit from the CHIPS Act.

Not surprisingly, Samsung’s announcement was coordinated with South Korea’s own government subsidies plan to support the industry. And additional 360 billion won will be provided to support chip packaging, and 100 billion won in electricity and water infrastructure that are required to support this massive increase in manufacturing capacity.

This is all good news for consumers and companies around the world. After all, we experienced a global semiconductor shortage during the COVID-19 pandemic. Many types of semiconductor simply didn’t have enough manufacturing capacity and there were chokepoints throughout the industry.

And incredibly, despite the pandemic being over, there are still shortages in certain sectors, like the automotive industry. Major investments in stable markets are absolutely critical to ensuring that doesn’t happen again.

Assuming that Intel can get its act together and execute on its plans, Intel, Samsung, and TSMC will be the three largest semiconductor manufacturing companies for the next decade. As a reminder, chip giants like NVIDIA, AMD, and Qualcomm have almost all of their semiconductors manufactured by TSMC.

As consumers, all of this investment and competition will ensure two things. The torrid pace of material improvements in semiconductor performance will continue year after year. And the costs for leading edge semiconductors will remain price competitive. And for commodity semiconductors like memory products, our price per Gigabyte (or Terabyte) will continue its historical rate of decline.

Adept AI is now worth $1 billion…

Adept AI is just had a massive funding round. The company raised $350 million in Series B round.

As a result, Adept is valued at $1 billion. It’s now a unicorn.

As a reminder, we first highlighted Adept AI last May. That’s when the company first came out of stealth.

What really caught my eye about this company is the management team. Adept AI was founded by a former vice president of OpenAI as well as some researchers from the Google Brain division. This is the department that pioneered Google’s work on a form of artificial intelligence (AI) called deep learning.

What makes Adept AI stand out is the company’s unique approach. Unlike other AI companies, they’re focused on using this powerful technology to make people more productive with the help of an AI-based digital assistant.

Get this – they train their AI to understand and use popular software platforms that people use every day. This includes social media platforms, customer relationship management platforms, and even spreadsheet software like Microsoft Excel.

To be more specific, the AI learns how to use the software itself. And once the AI knows how a given software works, it can be instructed to perform tasks that rely on using that underlying software.

For example, imagine we’re a cybersecurity company and we want to find Chief Information Officers (CIOs) from companies of a certain size on LinkedIn. Let’s assume that this is the profile that best fits our go to market strategy for business development.

Normally, this kind of thing would take a person a long time to do. They would have to go through profiles, find the right people, and then make a list of their top targets. Company names would have to be correlated with annual revenues. It’s a lot of tedious, manual work.

That’s where Adept’s AI comes in. It could handle a task like this in just a matter of minutes. And if given a proper e-mail template, the AI could reach out to each CIO on behalf of the company as well with customized language suitable for each target company.

In just the same way, Adept’s AI could help automate tasks within virtually any software platform. It’s just a matter of training the AI.

Some AI companies are more focused on consumer targeted digital assistants which will help us with tasks throughout the day. Adept is taking the same concept just focused on enterprise use cases. And it takes things one step further by studying the software that enterprise organizations use every day, and then making its AI functional with all those software platforms.

This is a powerful approach with an absolutely massive market opportunity. So we’ll certainly be tracking this company very closely going forward.

Adept AI is part of a new breed of AI companies that are set to enhance human productivity. That, in turn, will transform the workplace in a way that we haven’t seen since the internet rolled out.

I can’t help but be a bit disappointed though. When I reflect back on the last three decades of work, I can remember how many tasks over the years were tedious, manual, and very time consuming; but required for me to get my work done. And now, these tasks will all be done in seconds…

Meta’s latest move is quite telling…

Meta, the company formerly known as Facebook, just made a big announcement. The company is giving up on its plan to enable non-fungible token (NFT) functionality across its social media applications.

Regular readers may remember that this plan has been in the works for over a year now. It centered around Meta’s work to build NFT functionality directly into its digital asset wallet.

The idea was that both Facebook and Instagram users would be able to show off, create, and sell NFTs on both platforms. And this made perfect sense.

Social media is all about showing off to friends, family, and strangers. And during the most recent NFT boom, people loved to show off their unique NFTs. They became a status symbol in some cases, as well as a way to signal to people how you wanted to be seen.

So on the surface it’s surprising to see Meta drop this initiative. After all, most of the work is likely done by now. And the combination of digital collectibles and social media makes a lot of sense.

To explain the decision, CEO Mark Zuckerberg said Meta was shifting its focus to other areas. He declared 2023 as the “year of efficiency” for Meta.

But there’s a lot more to this story.

Remember, NFTs are digital assets which in many cases have material value. They can be bought and sold with cryptocurrencies or a stablecoin. This puts them in the same category as the larger crypto industry.

And right now the U.S. government is more hostile than it’s ever been towards cryptocurrencies and digital assets. They’ve gone so far as to take legal action against major crypto companies like Coinbase, Kraken, and Ripple Labs. These are three of the most buttoned-up and regulatory compliant companies in the entire space.

So Meta saw the writing on the wall here.

Launching NFTs into the current environment would bring about too much unwanted regulatory scrutiny. I wouldn’t be surprised if someone explicitly told Zuckerberg to back off. And this looks to be something of a pattern for Facebook.

Longtime readers may remember that the company attempted to launch their own cryptocurrency, backed by a basket of currencies, called “Libra.” The coin was formally announced in 2019.

But as the years went on, Facebook lowered its ambition for Libra—rebranded as “Diem” in December of 2020. Eventually, Facebook gave up and sold the project to Silvergate, a crypto friendly bank, in January of last year.

I can’t say for sure, but I suspect Zuckerberg was told to “stay in his lane” by several people in the government.

This latest development demonstrates just how bad the current regulatory environment is. There’s no reason why enabling NFTs on social media applications like Facebook and Instagram should be a sensitive subject.


Jeff Brown
Editor, The Bleeding Edge