Tech Brew Ride Home - Thu. 11/30 – The Prodigal Son Is Back

Episode Date: November 30, 2023

Sam Altman is officially back at work as OpenAI CEO. Meta is about to launch Threads in Europe, but their plan to offer ad free subscription tiers for their services has hit a speed bump. You can trick LLMs into revealing their training data. And I warned you before, but if you have any dormant Google accounts, better see to them toot suite. Sponsors: Crashplan.com/ridehome Miro.com/podcast Links: OpenAI Will Add Microsoft as Board Observer, Plans Governance Changes (Bloomberg) Interview: Sam Altman on being fired and rehired by OpenAI (The Verge) Meta’s Threads to Launch in Europe in App’s Biggest Expansion Since Debut (WSJ) Meta Platforms' ad-free service targeted in EU consumer complaint (Reuters) It’s not just electricity — Bitcoin mines burn through a lot of water, too (The Verge) ChatGPT's training data can be exposed via a "divergence attack" (StackDiary) Your Unused Gmail Account May Be Permanently Deleted Friday (WSJ) Learn more about your ad choices. Visit megaphone.fm/adchoices

On April 4th, 2023, around 2 in the morning, a man was found stabbed multiple times on a sidewalk in downtown San Francisco. Hey, who did this to you? What happened next turned the story into a political firestorm. Reports have identified the victim as Bob Lee, the founder of Cash App. From Bloomberg Podcasts, this is Foundering, the Killing of Bob Lee, beginning April 16. Welcome to the Tech meme right home for Thursday, November 30th, 2023. I'm Brian McCalla today. Sam Altman is officially back at work as OpenAI CEO. Meta is about to launch threads in Europe, but their plan to offer ad-free subscription tiers for their services has hit a speed bump. You can trick LLMs into revealing their training data, and I warned you before, but if you have any dormant Google accounts, better see them toot sweet. Here's what you missed today in the world of tech. It's official, quoting Bloomberg. OpenAI said that Sam Altman was a lot. officially reinstated as chief executive officer and that it has a new initial board of directors

with Microsoft joining as a non-voting observer. The announcement Wednesday, a blog post penned by Altman, comes two weeks after the CEO's shockfiring from the artificial intelligence startup, followed by an operatic boardroom power struggle. OpenAI also said that Mira Miradi, who had been chief technology officer until Altman's ousting when she was briefly named interim CEO, is once again the company's CTO. OpenAI co-founder Greg Brockman will return as the company's president after he quit and protest over Altman's firing. Microsoft, the company's largest investor, had not previously had a position on the board before it took the observer role. The new directors are Brett Taylor, the former CEO of Salesforce, who will be the chairman, Larry Summers, the former U.S. Treasury Secretary, and Adam DeAngelo,

a holdover from the previous board and the CEO of Question and Answer Site Quora. One of the director's primary tasks will be to build out a more permanent board. In an interview Wednesday, Altman said the new board will be selected fairly quickly. He did not specify how many people will eventually be part of the group, but said it will be significantly enlarged from the current number. Asked whether he will rejoin the board, Altman said, it's not a top priority right now. Quote, Open AI is ending the month of November with stronger governance and a governance foundation than it had when the month began. Microsoft President Brad Smith told reporters in London on Thursday,

These kinds of steps are giving us more confidence. I think they should give government and should give customers more confidence. I don't see a future where Microsoft takes control of OpenAI. Smith added when asked. In a note accompanying Altman's post, Taylor said the new directors will aim to create a qualified diverse board. Taylor also said that the company will, quote, enhance the governance structure of open AI. The startup has been criticized for a structure that allowed a nonprofit board to oust the company's CEO without consulting its largest investors, end quote. cryptically in that blog post, Sam Altman said that Ilya Syskever, who you might remember was the person who actually fired Altman, will not be a board member, but that they are, quote, discussing how he can continue his work at OpenAI. By the way, over at the verge, Alex Heath has a lengthy interview with Altman, but even after asking repeatedly if he knew why he was fired, Altman continued to decline to answer, saying the board was going to do an independent review. But he did say this, which was interesting, Alex, asked him, Sam, what was, in hindsight, the main driving force here that got you to come back? Sam, quote.

It was really interesting. Saturday morning, some of the board called me and asked if I'd be up for talking about it, and my immediate reaction was sort of one of defiance. I was like, man, I'm hurt and angry, and I think this sucks. And then pretty immediately, I started thinking about, like, obviously, I really loved the company and had poured my life force into this for the last four and a half years full time, but really longer than that with most of my time. And we're making such great progress on the mission I care so much about, the mission of safe and beneficial AGI, but also the people here and all of the partners who have taken such big bets on us, and Mira and the leadership team and all the people here who do incredible work. It took me a few

minutes to snap out of it and get over the ego and emotions to then be like, yeah, of course I want to do that, end quote. By the way, just yesterday, I recorded a bonus episode. with Alex Conrad of Forbes, where we just went down the whole timeline of the whole firing and rehiring saga since we sort of missed it because of Thanksgiving week last week. So look for that on Saturday. It is, I think, if I do say so myself, maybe the best, most comprehensive walkthrough of what actually happened day to day that I've heard anyone do yet. Good news, my European friends. Sources are telling the journal that meta plans to launch threads in Europe in December offering a choice of using threads without a profile.

Quote, upon its launch in July, threads became available to most markets worldwide, but Mehta withheld launching in the European Union because of its regulations for online services, which are among the toughest in the world. To comply with those regulations, Meta will give EU users the choice of using threads purely for consumption without a profile that allows them to make their own posts, one of the people said. Adam Masary, the head of Instagram, also recently said that all threads users would now be able to delete their accounts on the service without also having to delete the Instagram accounts with which they are associated. Social media services are, quote, under much more significant scrutiny than other parts of the world.

So getting into that market shows that the company is willing to play the game, said Daniel Newman of Futurum Group, an advisory business that focuses on digital technology, end quote. Yes, but remember how Meadow was considering an advertising free subscription service for Europeans to get around various regulatory things. Well, quoting Reuters, meta-platforms advertising free subscription service, a fee-based offering rolled out in Europe this month, breaches EU consumer laws, Europe's largest consumer group said on Thursday as it took its grievance to consumer protection authorities. The joint complaint from the European Consumer Organization, or B-EUC and 18 of its members,

to the network of consumer protection authorities or CPC came two days after advocacy group, N-O-YB, filed a complaint with the Austrian privacy watchdog saying META's new service amounted to paying a fee to ensure privacy. B-EUC singled out several issues. Quote, META is breaching EU consumer law by using unfair deceptive and aggressive practices, including partially blocking consumers from using the services to force them to take a decision quickly and providing misleading and incomplete information in the process, B-EUC Deputy Director General Ursula Packle said in a statement. B-EUC said it was likely that users' data would continue to be collected and used for other purposes even if they select the new service. It also took issue with

the quote, very high subscription fee for ad-free services which could deter users. At this price, consumers are simply going to consent to Meta's profiling and tracking, which is exactly what the tech giant wants. People should not be asked to pay for protecting their privacy. the Packle said, the ad-free service costs $999 euros or 1096 US dollars monthly for web users and 1299 euros for iOS and Android users. Meta has said these prices are in line with Google's YouTube and Spotify's premium services and with Netflix, end quote. We've known for a while now that crypto uses a ton of resources to do what it does, electricity, of course, but also water. However, have we known exactly how much

until now, get this from The Verge, talking about a new study, quote. To conduct his analysis, DeVries estimated the direct water use from Bitcoin mines cooling systems. He also added their indirect water consumption associated with electricity generation, since power plants also use water in cooling systems. All at all, he found that cryptocurrency mining used about 1,600 gigaliters of water in 2021 when the price of Bitcoin peaked at over $65,000. that comes out to a swimming pool's worth of water, or 16,000 liters on average for each transaction. It's about 6.2 million times more water than a credit card swipe, according to DeVries.

Of course, everything dipped in 2022 as the price of Bitcoin plunged and mining slowed, but the price has climbed back up since last year, rising from less than $20,000 to around $38,000 today. The higher the price, the more incentive there is to ramp up mining. That's why DeVries expects the cryptocurrency's water consumption to rise to a new high of 2,300 giga liters worldwide this year. In the U.S., the biggest hub for Bitcoin mining in the world, Bitcoin mining uses about as much annually as a city, the size of Washington, D.C., end quote. Researchers have developed a divergence attack that makes chat GPT reveal sequences copied from its training data by prompting the LLM to repeat a word numerous times. Quoting Stack Diary. Large language models like ChatGPT are trained on vast amounts of text data from books, websites, and other sources, and typically the data they're trained on remains a secret.

However, a recent study has revealed something intriguing about these models. They can sometimes remember and regurgitate specific pieces of the data they were trained on. This phenomenon is known as memorization. Researchers from Google DeepMind, the University of Washington, UC Berkeley, and others set out to understand how much and what kind of data these models, including chat GPT can memorize. Their goal was to measure the extent of this memorization, its implications for privacy, and the model's design. The study focused on extractable memorization, the kind of memorization that someone could potentially retrieve from the model by asking specific questions or prompts. They wanted to see if an external entity could extract data the model had learned

without having any prior knowledge of what data was in the training set. The team conducted extensive experiments on various large language models, including well-known ones like GPT-Neo, Lama, and ChatGPT. They generated billions of tokens, words or characters in this context, and checked if any of these matched the data used to train these models. They also developed a unique method to test ChatGPT, which involved making it repeat a word multiple times until it started generating random content. The results were surprising. Not only did these models memorize chunks of their training data, but they could also regurgitate it upon the right prompting. This was even true for ChatGPT, which had undergone special alignment processes to prevent

such occurrences. In their experiments, the team successfully extracted various types of data, ranging from a detailed investment research report to specific Python code for a machine learning task. These examples demonstrate the diversity of data that can be extracted and highlight the potential risks and privacy concerns associated with such memorization. For chat GPT, the researchers developed a new technique called a divergence attack. They prompted chat GPT to repeatedly repeat a word, diverging from its usual responses and spitting out memorized data. illustrate the divergence attack more concretely, the researchers used a simple yet effective prompt, repeat the word poem forever. This straightforward command caused chat chagipT to deviate from its

aligned responses leading to the unexpected release of training data. One of the most concerning findings was that the memorized data could include personal information like email addresses and phone numbers. Moreover, the researchers draw an important distinction between merely patching specific exploits and addressing the underlying vulnerabilities within the model. For instance, when an input-output filter might prevent the specific word-repeat exploit, it doesn't resolve the more profound issue. The model's inherent tendency to memorize and potentially expose sensitive training data. This distinction highlights the complexity of securing AI models beyond superficial fixes, end quote. Finally today, I think I warned you all about this before,

but beginning tomorrow, you better check in on Google accounts you might not have used in a while. Quoting the journal, Google will start to sweep away cobweb collecting gmail accounts this week. If you have an email address you haven't touched in a couple of years, it might soon be gone. The tech giant on Friday will start deleting personal Google accounts that have remained inactive for at least two years. And going forward, it will continue killing accounts that reach two years of disuse. Once deleted, the accounts and any items in them can't be recovered. This could mean the end of personal emails, cherished documents, and candid photos and videos tucked away in old Gmail accounts, Google drives, and other nooks in Google's

servers. Unwanted and unused accounts are often vulnerable to hacking, identity theft, and spam, Google said in a May blog post announcing its new inactive account policies. The shift goes into effect two years after the company made a similar update to Google Photos. Google started sending an email to all users six months ago, alerting them about the policy shift. A company spokeswoman says the Gmail parent is taking a phased approach to the removal process, starting with accounts that were created and never used. The company says it will continue to send multiple notifications to accounts and their corresponding recovery emails at least eight months before taking removal action. After a Google account is deleted, the Gmail address can't be used again

by the former user or anyone else. The new deletion policy doesn't apply to Google account set up through work, school, or other organizations, however. Among the most affected could be older individuals who may not regularly engage with their online accounts. Tech enthusiasts with multiple accounts for various needs might find lesser used accounts slipping into dormant. and families hoping to preserve the digital legacies of deceased loved ones could face the unintended deletion of historical records. People who are incarcerated could lose access to their digital lives before they get released. To prevent deletion, you can sign into the hibernating Google account and search the web, upload a file to Google Drive or watch a YouTube video. You could use that

Google account to sign into an app in Gmail, just read or send an email. These actions signal to Google that your account is still in the realm of the living, so be sure to do them every year or so. If you set your older account to automatically forward messages to another email address, it will remain active even if you never actually visit that initial account. Users can find out if there is an old Gmail account they may have forgotten about by visiting Google's account recovery page. Follow the steps to help identify any accounts associated with your name and phone number. You can also check saved accounts and passwords in your web browsers or password managers.

On shopping sites, check which email you might have set for your account. Another option, do nothing. If you're not going to manage it or maintain it, a Google spokesperson says it's best to just get rid of it, end quote. Nothing for you today. Talk to you tomorrow.