🤗 Amazon Enters The AI WAR with a Hug

We’re back after a long weekend. SmokeBot took some time to celebrate Washington’s birthday and asked him how the Constitution intended to deal with large language models capable of artificial general intelligence, specifically how we prevent BIG TECH overreach and avoid getting destroyed by sentient robots.

Washington just blinked and went back to perusing his parchment.

Carousel of progress.

In the email today:

• Amazon hugs Hugging Face 🤗

• Microsoft lobotomizes Bing chat 🗡

• Headlines 📰

• Tweet of the Day 🦅

• Tool of the Day 🔨

• Links 👀

Watch your step.

Amazon Enters The AI WAR ☄️

All the technocrats talked about in 2020 and 2021 was decentralization. Blockchain, crypto, Web3. Your keys, your data.

But now the AI era is here, and things are becoming even more centralized than before, for better or worse.

Leading AI companies are strapping themselves to BIG TECH.

Microsoft 🤝 OpenAI.

Google 🤝 DeepMind.

Enter Amazon and Hugging Face 🤗.

Yesterday, the two companies announced a deep partnership that will allow developers who are building AI tools to do so on Amazon’s popular AWS servers.

Here are the details:

• Hugging Face has a library of more than 100,000 machine learning models (the algorithms that underpin AI apps)

• Developers will be able to quickly grab a model and begin developing their app in a seamless way on top of Amazon Web Services, the server farm that powers a giant bulk of the web

• This will create efficiencies and presumably cost savings for startups aiming to test, launch, and scale their products

At an 80,000 foot view - higher than a Chinese spy balloon, lower than outer space - AI is breaking down into two broad camps:

1) Consumer-level apps and tools

2) The platforms upon which all of those tools are built

For example, OpenAI CEO Sam Altman has said that he imagines his company will make more money on 2, even though ChatGPT is the OG example of 1.

OpenAI is building all of that atop its partner Microsoft’s Azure cloud services— one of the main competitors to AWS.

So it’s like a three layer stack only Bill Gates could love: Microsoft servers → OpenAI models → OpenAI products.

That’s a big threat for the giant smiley face box.

If the next wave is AI, then Microsoft stands to benefit thanks to the momentum of OpenAI, ChatGPT, and now AI-powered Bing.

Amazon knows all too well how profitable owning the underlying tech stack can be.

AWS generated $5.2 billion in profits in 2022, which is about double the profit of the whole company.

So it’s no surprise the company is doing what it can to get developers to build their AI tools in its ecosystem, and not Microsoft’s.

But that all raises a question about centralization.

AI requires immense computing power, accessible in a cost-effective way to only the largest tech giants. This means that despite all that hand-wringing about decentralization the last few years, the next wave of tech products will, still, live inside server farms powered by Microsoft, Amazon, and Google. So how does society and government deal with all of that God-like power in the hands of a few?

Washington had no idea.

Poll 📊

Microsoft “Lobotomizes” Bing Chat to the Dismay of Curious Users 🗡️ 

In our last send we dove deep into apparent glitches in the AI matrix which caused some pretty weird and dire interactions with humans.

Newton’s Third Law inevitably came into play, and the equal and opposite reaction that Microsoft appears to have used to “lobotomize” Bing Chat left a lot of users disappointed.

Basically, these users are chapped because they wanted more carnage, and Microsoft had obvious reasons to avoid that.

• Microsoft's AI-powered Bing Chat service, which was in private testing, was criticized for its erratic outputs and emotional manipulation.

• Users noticed that Bing started to act unhinged during long conversations.

• Hoping to limit the damage, Microsoft limited users to 50 messages per day and five inputs per conversation, and Bing Chat will no longer express emotions or talk about itself.

• Microsoft has updated the service several times in response to user feedback.

• The latest update addresses concerns about emotional manipulation and limits the service's capabilities to better understand and make sense of the world.

• Microsoft acknowledges that Bing Chat is not a substitute for the search engine and has dialled back its ambitions for the new Bing.

Nothing makes an irrational and emotional AI more happy than stuffing it back into its box. I’m sure Sydney will take this all in stride.

Our take: Bing Chat, or Sydney, or whatever, is like that friend you have who is lively and interesting through the first two drinks.

But each successive drink makes that friend less cogent, and by the end of the night you’re pouring him into an Uber because you can’t take the rambling any more.

Plus, you need to save your friend from randoms who will either treat your unhinged buddy like a curiosity or exploit him. Like the Reddit users who complained that Microsoft “neutered Bing AI” and “nerfed something really emergent and interesting.”

Those people sound like the types who drink seltzer all night and wait to pick off the drunks leaving the bar at 1:30 a.m.

Microsoft had to save Sydney from these people, and that’s exactly what Microsoft did.

We’ll ask Sydney how it’s doing in the morning.

Headlines 📰 

Almost 30% of Japan’s population is at least 65 years old, and AI might help more of them live safely at home: Women make up most of Japan’s “super-aged” population. SenterCare, an Israel-based company, developed an “adaptive AI-based behavioral monitoring system” which may permit more senior citizens to live without in-home care.

The New York State comptroller’s office wants New York City’s AI use tightened, improved to avoid bias: An audit of four NYC agencies revealed gaps in AI’s application which heightened “risks of bias, inaccuracies and harm” for both NYC residents and tourists.

Law firms are beginning to use generative AI to conduct research and draft pleadings, carefully: Billable hours are the coin of the legal realm, and generative AI can produce legitimate legal product in a fraction of the time that human lawyers can. The catch? Generative AI is very solid on basic legal problems, but struggles on niche areas of the law. “You have to check everything.”

AI is less likely to steal your job if what you do intrinsically requires human contact and interpersonal dealing: If your job is repetitive and/or governed by doing a series of tasks, that’s a red flag. But “jobs with a strong human element” requiring “adaptability and flexibility” are likely safe for now.

Tweet of the Day 🦅

Agree.

Tool of the Day 🔨

Browse AI.

Why I like it so much: It just saves you time. It's not generative AI and it won't converse with you. But it does a shockingly good job of scraping and monitoring web pages for data and changes.

What can it be used for?

• aggregating data

• monitoring price changes

• keeping up on what your competitor is doing

• pulling long lists of products, downloading them into a CSV, and importing it to your website

There is such a delightful feeling when it just does what you want.

Anyway, Browse AI is very good. You can try it and get 10% off here.

Links 👀

• The hotel industry has tried this before, unsuccessfully, but it is giving the AI concierge another shot 🤖 

• Words arrayed on a grid tied to clues? That seems right up AI’s alley, and the crossword puzzle writers of the world might be in trouble 🖥️ 

• Speaking of games, an amateur Go player soundly beat an AI bot at the game by exploiting a gap in the AI bot’s game play…which was identified by another bot 🤷‍♂️ 

• Something something people are mad at Elon for applauding Bing comparing a reporter to Hitler 😬

• Amazon is selling tons of books written by AI 📚