?? Wiser! #138: AI Pirates | Bitcoin Beats SEC | Tesla's AI Vision For The Future Of EVs

TL;DR: After The Atlantic exposed the unauthorised pirating of 170,000 books to train large language models, the question we should ask is "is that ok?" (It's not a simple answer!) Plus: a US judge has overturned an SEC decision on Bitcoin ETFs, and Elon Musk has given us a glimpse of the future of AI self-driving.

Welcome to Wiser!

Whats In Wiser! This Week

Google has unveiled an?pile?of new AI tools and capabilities at the kickoff of the company’s three-day?Google Next conference ?in San Francisco. I didn’t have time to digest them all and do them justice for this week’s Wiser! That will come in next week’s issue.

However, the three stories I focus on are all important for you to be able to talk about. First, there’s the story from The Atlantic about large language models being trained on published books without prior consent. It raises legal, moral and ethical questions that are central to the future development of generative AI.

Then’s there’s the Bitcoin story of the moment. I don’t write so much about crypto but this is important because it shows how both traditional finance and regulators are struggling to embrace digital currencies.

Finally, there’s an Elon Musk story. But not about TwitterX, this one is about Tesla and their latest approach to self-driving AI that could totally transform the way cars, robots, autonomous machines behave.

Chart of the week looks at the web1 BigTech giant Yahoo and their steady rise back to prominence. AI superpowers is a 2 for 1 bargin this week, with a free course and an excellent podcast to listen too. Plus, as always, there’s a dozen or so other interesting stories going on in tech.

Apologies to PLUS! members, there’s no long form article this week. I’m in the middle of shutting down the website and moving to a simpler landing page model. I’ve also started a free “how to get the best out of LinkedIn” course provided by LinkedIn, invitation only. Should be interesting given that I think LinkedIn’s latest algorithm changes are rubbish. I can’t wait to tell them!

ATB, Rick

P.S. REMEMBER: Insight and Information Gives You Leverage!

Be The Smartest Person In The Room...Get Wiser! Every Week

Join The Mailing List For Free ...And NEVER Miss An Issue.

w/Artificial Intelligence

The Atlantic Exposes Popular AI Models Use of Pirated Books for Training Data

The Atlantic just released a major investigative piece that proves that popular large language models, like Meta's Llama 2, have been using pirated books to train their models. The article claims that upwards of 170,000 books, the majority published in the last 20 years are in Llama's training data. The books are part of a data set called Books3 and was also used to train BloombergGPT, Luther AI's GPTJ, which is a popular open source model, and other unnamed LLMs.

Here’s The Thing You Need To Know: BigTech companies like Meta and Google have been profiting off the work of others for the past 15 years. The news industry missed a trick in the early 2000’s by allowing Google, Meta et al unfettered access to all of its work. Is the same about to happen with generative AI? Well, according to Rebecca Tushnet, a law professor at Harvard, the law is “unsettled” when it comes to fair use involving unauthorised material in this way. So the answer may well be yes. And the point is that once the LLMs have been trained on any data, it can’t be untrained. There’s no taking the eggs out of this cake! That’s why you’re most likely going to see licensing deals between the big AI firms and the big media firms. A second angle to this story is the question “why train on books?” It’s really quite simple, the large language models need to be trained on good writing. If they just scraped the open Internet they’d collect the vast majority of banal and meaningless fluff thats on the web. From misinformed blogs or corporate marketing speak, you can find tons of vacuous writing that reads well for the hard of thinking but leaves the curious none the wiser. Expect the debate on the legal, moral and ethical issues around generative AI to simmer along in the background for a long while yet! Link

w/Crypto

US Appeals Court Orders SEC to Reconsider Grayscale Bitcoin Trust Ruling

A United States federal appeals court has ruled that the SEC must reconsider its ruling on the Grayscale Bitcoin Trust. This was widely expected to be a game changer for crypto by giving less crypto savvy investors easy access to cryptocurrency via an ETF that more or less tracked the spot price of Bitcoin (as opposed to a futures ETF betting on what might happen). The issue was that the Bitcoin ETF wasn't an ETF in the eyes of the SEC and they denied permission to investment firm Grayscale to convert their Bitcoin Trust into an ETF. However, this ruling has told the SEC that they got it wrong in denying Grayscale. Remember, it’s not just Grayscale impacted by this ruling, there’s also Fidelity, BlackRock, Whismtree, Vanek, InvestCo, and others all have put in filings with the SEC to have a Bitcoin ETF.

Here’s The Thing You Need To Know: SEC chair Gary Gensler has been waging a war on crypto for a few years now. Each denial of the industry ends up in court and more often than not the regulator loses. He just doesn’t seem to want a crypto market in the United States and is pursuing a King Canute strategy that will fail as long as there’s life in Bitcoin. The point is that there is a market demand for a Bitcoin ETF. For anyone wanting exposure to Bitcoin without owning one, just as in the same way you want exposure to tech stocks but don’t want to actually buy Apple or Alphabet stocks, a Bitcoin ETF is the obvious vehicle. And, as Bitcoin’s credibility steadily increases and it becomes a viable and sustainable asset class, then an ETF just makes sense in a diversified portfolio, regardless of whether you believe in Bitcoin of not. Just looking at Bitcoin as an asset, its price is up nearly 25% year on year. More .

Further Reading: SEC chair Gary Gensler's court losses are piling up in crypto

w/Autonomous Vehicles

Elon Musks Latest Tesla Self-Driving Demo Doesnt Use Any Code

In another example of what AI will become, Elon Musk took to the streets of California in his own Tesla Model S and live streamed 45 minutes of him showing off the Tesla's full self-driving version 12 (remember this) on the streets. So what, you might be thinking? Well, this was different to all previous self driving experiences because this was self-driving that doesn't use a single line of code to pilot the car autonomously. Right now the way Tesla’s do full self-driving is they have eight outward facing cameras with computer vision. The cameras observe and record everything that goes on around the car. The cameras feed computer code that drives the car. There's code that tells the car to slow down for a speed bump, stop at a red traffic light or reduce speed if a person walks out in front of the car. That’s how self-driving works today, by code. But not tomorrow.

Here’s The Thing You Need To Know: Tesla is an AI company that makes cars. They have more data about human behaviour behind the wheel than anyone. Musk has given us a taste of what’s to come from AI and it’s going to transformation the future of self-driving, robotics, machine automation, everything. That’s because version 12 is based on observing how a human drives a car and learning from the human’s behaviour, not from the instructions of pre-programmed computer code. The AI learns in the same that any new driver learns. The AI observes human and how they interact with the world around them. For example, the AI will learn that when a human sees a person crossing the road they might slow down, but don’t necessarily stop, although some times they do. With a code driven car, the driving decisions are binary, stop/go. With a human driver there’s more subtlety and nuance. It’s the latter mode that the version 12 AI machine learning will mimic. The bigger picture is that this capability, powered by Tesla’s super computer called Doja, can be applied to robotics, General AI and fully autonomous robo taxis. Watch Musk’s Livestream .

Chart Of The Week

For the uninitiated, Yahoo was a Web1 version of BigTech way before any of today’s Web2 BigTech giants were even thought off. In?1994, the company's founders launched "Jerry and David's Guide to the World Wide Web", the first version of an online Internet directory. Yahoo was what we used to call a “web portal” and gave early WWW users the ability to search, send email, do shopping, get the news and more. In its dotcom bubble heyday, Yahoo was once valued at?$125 billion.

But Yahoo stumbled after the bubble collapsed and then Google, Facebook, Amazon came on the scene and ate its lunch. As the wave of Web2 washed over the Internet, Yahoo was increasingly left behind and bought by?Verizon?for?only $4.5 billion. Yahoo returned to private ownership with Apollo in 2021 and the brand is on a journey to reinvent itself. After all, it still has as many hits a year as wikipedia, twice as much revenue as TwitterX and it provides a reliable email service that I have used for over 100 years.

For more data driven insights like these.

AI Superpower Of The Week

1. AI: An Introduction For Business (19th September)

This week’s Wiser! superpower is not an AI tool, but an AI lesson. I can’t think of many better people better qualified than Charles Radclyffe to give a 101 introduction on artificial intelligence.

During this introductory session Charles will cover:

• What AI is and what it isn't - key areas of AI
• How is AI changing business as usual
• Key considerations for business leaders before jumping into AI
• How to start building an AI strategy - the types of strategies you could consider
• Interactive Q&A where you can air any questions you have about how AI may impact your business

If you're a business leader and unsure of how to approach to topic of AI, this is for you!

Best of all, it’s free!. I’m already registered and you can sign up too, right here .

2. Mustafa Suleyman: Will AI Save or Destroy Humanity

Mustafa Suleyman is the founder of Inflection.AI , the developers of Pi , my favourite Ai chatbot. He’s British with an English mother and Syrian father and now one of the leading silicon valley characters at the forefront of the AI gold rush. It’s worth a listen:

Watch Wiser! on YouTube

What Else Is Going On In Tech?

• OpenAI is reportedly on pace to hit $1 billion in annual revenue as more companies adopt ChatGPT, a huge leap fuelled by enterprise demand.
• China is set to approve its first public generative AI services , allowing its tech giants to compete with US-based models. Baidu's Ernie Bot is expected to be the first to win approval, after China's endorsement of the industry comes after implementing its first AI regulations.
• Google’s Duet AI can now write your emails for you. The company’s corporate $30/month subscription AI copilot has a range of new services and can?enhance Gmail’s existing smart reply feature ?by allowing users to draft longer, more personalised emails with a tap. More about Google’s new product announcements in next week’s newsletter.
• Microsoft has?unveiled ?a new AI training method called the "Algorithm of Thoughts" (AoT) , designed to make large language models like ChatGPT more efficient and human-like in their reasoning abilities. The new approach is the natural next step for the company, which has invested heavily AI, and particularly in OpenAI, the creators of DALL-E and ChatGPT.
• If AI becomes conscious: here’s how researchers will know .
• Amazon is rumoured to be developing its own?NFT?marketplace. The retail giant is already playing in the NFT space via its?Prime Gaming portal and are giving away NFTs?totally free for subscribers . If you pay for an Amazon Prime subscription, then there’s already a selection of free NFTs that you can claim and use in current blockchain games.
• Delivery robot company?Starship?has?said?that its rovers will be on 50 university campuses this autumn and now has over 2K delivery bots operating across the USA. Other bot companies are expanding, too: Coco is in Los Angeles and Austin, ex-Uber owned Serve Robotics?services over 200 LA restaurants, and Uber Eats?partnered with Serve in May to deploy 2K more bots. (Delivery robot rides through a crime scene. )
• General Motors is to use Google AI chatbot for its OnStar in-car concierge service. OnStar is already powered by intent-recognition algorithms that have been using Google Cloud’s conversational AI since 2022. The latest announcements will allow OnStar to provide drivers with responses to common queries like “help me, I’m lost” and “which way to McDonalds.” The move to up-skill OnStar’s capabilities is in line with the car manufacturer’s strategy to build a?$25 billion services subscription business by 2030 .
• This is funny, but also serious! While some experts?estimate ?that up to?90%?of online content could be AI-generated by?2026, the?Columbus Dispatch ?has halted its use, after an illegible AI written article detailing a high school football game went?viral.
• Those tuning in to highlight videos of the US Open tournament may encounter a prolific new commentator: IBM’s AI platform Watson. IBM’s AI is to provide “audio commentary and text captions on video highlight reels of every men’s and women’s singles match,” IBM said in this?blog post .
• High tech “NeuroSkin” trousers powered by artificial intelligence?are helping stroke survivors regain their walking ability .
• Goldman Sachs is working on an AI tool for a?“regularisation-based”?asset-hedging system . The system uses AI to predict the performance of a hedging portfolio relative to a certain asset it’s tracking. It’s another example of the verticalisation of AI as corporates expand their own AI capabilities.

Further Reading

About The Author

Rick Huckstep is a writer, podcaster and YouTuber with a passion for emerging technologies and the way they shape tomorrow’s world.

??? Check out rickhuckstep.com .

?? Sign up for the newsletter and get Wiser! every week (plus a free copy of The Utility Of Emerging Technologies.)

??? Follow me on YouTube .

??? Listen to the Big Tech Little Tech podcast.