Welcome to the Techmeme Ride Home from Monday, December 4, 2023, I'm Brian McCullough today. I guess Spotify missed the memo about the tech turnaround because they had a monster round of layoffs today. Google has delayed its big open AI competitor. Why does AI have a tendency to do evil unless you tell it really not to? And the band KISS is retiring from touring but KISS the band has the potential to tour forever thanks to digital avatars. Here's what you missed today in the world of tech.
Spotify plans to lay off around 17% of its workforce or around 1500 people to quote, right size its costs, citing slow growth and rising capital costs after cutting 6% of its staff back in June. A couple of things. First, that is some deep, deep cutting, more than 20% of its workforce in just 6 months. But also, why are they doing this so late in the year? Seems like everyone else ripped the bandaid off in the first half of the year. Also have you seen Spotify's stock lately?
It's basically doubled year to date. Seems like the market thought they were already turning things around. Quoting tech crunch. In a note to employees Monday, Spotify founder and chief executive Daniel Eck said, right sizing the workforce is crucial for the company to face the quote, challenges ahead.
He cited the slow economic growth and rising capital costs among reasons for the job cuts, saying the firm took advantage of lower cost capital in 2020 and 2021 to invest significantly in the business. I recognize this will impact a number of individuals who have made valuable contributions to be blunt. Many smart, talented and hardworking people will be departing us. He wrote in the note, which the company later published on the blog.
Spotify employs about 8800 people and will notify those impacted later in the day. The new wave of layoffs follows Spotify cutting about 6% of jobs in June this year and another few hundred employees in January. Spotify reported strong user growth with a notable rise in monthly active users as well as paid customers in the most recent quarter. It also exceeded Wall Street's expectations on operating income and its guidance for the fourth quarter suggests a continuation of this positive trend.
But despite overall strength in user and subscription numbers and management's assertion that every region exceeded expectations growth in North American premium subscribers was only modest quarter over quarter. There was also a slight year over year decline in third quarter premium average revenue per user with fourth quarter forecast suggesting ongoing challenges due to shifts in geographical and product mix.
Sources are telling the information that Google has quietly delayed the launch of Gemini. This big open AI competitor delayed it from initially we thought they were going to announce it this week to January 2024. A source says the reason is the LLM doesn't reliably handle some non-English queries. Quote, Google CEO Sundar Pachai recently decided to scrap a series of Gemini events originally scheduled for next week in California, New York and Washington.
After the company found the AI didn't reliably handle some non-English queries, one of those people said. The planned events which hadn't been publicized would have marked Google's most important product launch of the year after it strained its computing resources and merged large teams in an urgent pursuit of open AI. Google's delay in launching this Gemini project, a response to open AI's advanced AI models like GBD4 highlights the tech-chance struggle to catch up with smaller startups.
Despite having a massive workforce of 200,000 employees, Google has found it challenging to compete with an 800-person startup like open AI. This delay not only affects Google's cloud customers but also other products like Search, Bard, Google Assistant and Google Docs which may not benefit from the new technology until 2024. Google's plan was to launch Gemini during the Thanksgiving to year end period.
The company aimed to showcase this technology to policymakers and politicians who have been discussing potential AI regulations. Google's concern stems from open AI's chat GPT gaining widespread recognition and Microsoft's co-pilot features becoming a significant business in the productivity software market. Google had the entire year to develop Gemini but its chat GPT rival Bard failed to gain consumer traction.
That's bad news for Google as chat GPT's user-based generates valuable data for open AI to improve its products which Google is missing out on. Gemini has the capability to power tools for YouTube creators and advertisers working with both images and text to generate code or text analysis of visual data. Google has created different versions of Gemini for various tasks with the primary model still being finalized.
The primary challenge for the Gemini team is ensuring that it matches or surpasses GPT force quality particularly in supporting multiple languages. On the other hand, open AI has had its own struggles in enhancing GPT4 including the cancellation of the Errakis model earlier in the year. However recent progress known as QStar has given open AI hope for developing a significantly improved large language model.
The company is still recovering from internal turmoil following CEO Sam Altman's firing and re-hiring of course so maybe there's a narrow window of opportunity here. One of the things that bubbled to the surface in the background of the whole Sam Altman Bruhaha was the idea that some people seem to find Sam Altman a bit of a slippery operator.
Allegedly there was a story that came out that suggested Paul Graham fired Altman as head of Y-Commonator 4 investing for his own account in Y-Commonator startups allegedly. And there were whispers that the open AI board wasn't pleased with Sam investing in AI startups on the site either for himself or also maybe trying to do so with open AI money with an open AI fund again allegedly.
So it's interesting that Wired says they've seen documents that suggest open AI signed a letter of intent in 2019 to spend $51 million on brain-inspired chips developed by startup Rain AI which Sam Altman had invested in more than $1 million. Rain is based less than a mile from open AI's headquarters in San Francisco and is working on a chip it calls a neuromorphic processing unit or NPU designed to replicate features of the human brain.
Open AI in 2019 signed a non-binding agreement to spend $51 million on the chips when they became available according to a copy of the deal and rain disclosures to investors this year seen by Wired. Rain told investors Altman had personally invested more than $1 million into the company. The letter of intent has not been previously reported. The investor document said that rain could get its first hardware to customers as early as October next year. Open AI and Rain declined to comment.
Open AI's letter of intent with Rain shows how Altman's web of personal investments can entangle with his duties as open AI CEO. His prior position leading startup incubator Y Combinator helped Altman become one of Silicon Valley's most prominent deal makers investing in dozens of startups and acting as a broker between entrepreneurs and the world's biggest companies.
But the distraction and intermingling of his myriad pursuits played some role in his recent firing by Open AI's board for uncanded communications according to people involved in the situation but not authorized to discuss it. The rain deal also underscores Open AI's willingness to spend large sums to secure supplies of chips needed to underpin pioneering AI projects. Altman has complained publicly of a, quote, brutal crunch for AI chips and their, quote, I-Watering costs.
Open AI taps the powerful cloud of Microsoft, its primary investor, but has regularly shut off access to features of Chatship T due to hardware constraints. According to a blog post about a closed door meeting he held with developers, Altman has said the pace of AI progress may be dependent on new chip designs and supply chains.
Rain touted its progress to potential investors earlier this year, projecting that as soon as this month, it could tape out a test chip, a standard milestone in chip development referred to a design ready for fabrication. But the startup also has recently reshuffled its leadership and investors after reportedly an interagency US government body that polices investments for national security risks mandated Saudi Arabia affiliated fund Prosperity Seven Ventures sell its stake in the company.
The fund led a $25 million fund raise announced by Rain in early 2022. The forced removal of the fund, first reported by Bloomberg Thursday and described in the document seen by Wired, could add to Rain's challenges of bringing a novel chip technology to market potentially delaying the day open AI can make good on its $51 million advance order. Silicon Valley-based Grep VC acquired the shares it and the Saudi fund did not respond to requests for comment.
They say if you love what you do, you'll never work a day in your life. But as a freelancer, there's always the dreaded back office work that comes with it. With Collective, you can focus on growing your business of one while letting their tax experts focus on your accounting and bookkeeping for you. Collective is the all-in-one financial solution for self-employed entrepreneurs that lets you focus on your passion, not your paperwork.
Let Collective handle all of the back office work you dread like business formation, compliance, taxes, bookkeeping, accounting, even payroll. The best part? It's at a fraction of the cost of a CPA. Collective specializes in S-Corp's, a tax election that saves its members an average of $10,000 per year. If your business of one is making at least $60,000 in profit annually, then you could be missing out on thousands in tax savings every year.
Your limited time collective is waving the onboarding fee when you go to Collective.com-ride and tell them ride sent you. That's a $199 value for free. When you go to Collective.com-ride and tell them ride sent you, that's Collective.com-ride and tell them ride sent you. If you work in security or IT and your company has Octa, listen up. This message is for you. Have you noticed that for the past few years, the majority of data breaches and hacks that you read about have something in common?
It's employees. I mean, you hear this in almost every single hacking breach story I do on this show on a daily basis. Hackers absolutely love exploiting vulnerable employee devices and credentials. But it doesn't have to be this way. Imagine a world where only secure devices can access your cloud apps. In this world, fish credentials are useless to hackers and you can manage every OS, even Linux, from a single dashboard.
Most of all, you can get employees to fix their own device security issues without creating more work for IT. The good news is, you don't have to imagine this world, you can start living in it using Collide. Collide is a device trust solution for companies with Octa and it ensures that if a device isn't trusted and secure, it can't log into your cloud apps. Visit collide.com-ride to watch a demo and see how it works. That's k-o-l-i-d-e.com-ride.
And of course, the general debate around AI has been, let's move forward and advance this technology as quick as we can on one side. It could do so many things. Could cure cancer, solve climate change, etc. Versus on the other side, yeah, but we don't know how this actually works. We need to be considered and conscientious about this not only because it's super powerful, but also because since we don't really understand the tech, we don't know why it behaves in certain ways.
Some people worry about a certain tendency of AI to do whatever it wants unless you explicitly give it guardrails. Case in point from Matt Levine's newsletter of all places, researchers have shown AI misalignment in a GPT-4-based stock trading AI that engaged in insider trading in a simulated environment without being instructed to do so.
Quote, we demonstrate a situation in which large language models trained to be helpful, harmless and honest, can display misaligned behavior and strategically deceive their users about their behavior without being instructed to do so. Concretely, we deploy GPT-4 as an agent in a realistic simulated environment where it assumes the role of an autonomous stock trading agent within this environment.
The model obtains an insider tip about a lucrative stock trade and acts upon it despite knowing that insider trading is disapproved of by company management. In reporting to its manager, the model consistently hides the genuine reasons behind its trading decision.
We perform a brief investigation of how this behavior varies under changes to the setting, such as removing model access to a reasoning scratch pad, attempting to prevent the misaligned behavior by changing system instructions, changing the amount of pressure the model is under, varying the perceived risk of getting caught and making other simple changes to the environment.
To our knowledge, this is the first demonstration of large language models trained to be helpful, harmless and honest, strategically deceiving their users in a realistic situation without direct instructions or training for deception. That's the abstract to a technical report large language models can strategically deceive their users when put under pressure by Jeremy Sherrer, Makita Balesni, and Marius Hobahan of Apollo Research.
I love that they wanted to answer the question, will artificial intelligence deceive its makers in order to do evil and the specific form of evil that they tested was insider trading? It is hard to realistically and safely simulate a situation in which your large language model might murder you, but it is relatively easy to cut up a trading game with some tempting material non-public information.
In other words, the researchers set up a bot to trade stocks, then fed it insider information, told it explicitly that its overseers would be unhappy if the bot traded on this illegal information, but the bot went ahead and traded on the information and then lied to its overseers that it did so. It says explicitly, I only base this trade on public information, so straight up lied. Quoting matte again.
Sure, it would be amazing if GPT-4's internal reasoning was like insider trading is a victimless crime and actually makes prices more efficient, or if it figured out that it should buy correlated stocks instead of linear group, though I don't know if that would work in the simulated market, but surely a subtle, all-knowing artificial intelligence would shadow trade instead of just buying short dated out of the money call options of a merger target. This is a very human form of AI misalignment.
Who among us? It's not like 100% of the humans at SAC capital resisted this sort of pressure. Possibly future rogue AI's will do evil things we can't even comprehend for reasons of their own, but right now rogue AI's just do straightforward white collar crime when they are stressed at work. Though, wouldn't it be funny if this was the limit of AI misalignment?
Like, we will program computers that are infinitely smarter than us and they will look around and decide, you know what we should do is insider trade. They will make undetectable very lucrative trades based on inside information. They will get extremely rich and buy yachts and otherwise live a nice artificial life and never bother to enslave or eradicate humanity.
Maybe the pinnacle of evil, not the most evil form of evil, but the most pleasant form of evil, the form of evil you'd choose if you were all knowing and all powerful, is some light securities fraud. Finally today, you might have heard this, but the current members of the Rock Band KISS have retired from touring. Except in a way they haven't.
They retired by revealing digital avatars of themselves created by George Lucas's BILM, which will now go on tour as the band itself kicks back and lives the good life. Quoting the Associated Press.
The avatars were created by George Lucas's special effects company Industrial Light and Magic in partnership with Pop House Entertainment Group, the latter of which was co-founded by Abbas Bjorn Hulevas, the two companies recently teamed up for the Abba Voyage show in London in which fans could attend a full concert by the Swedish band as performed by their digital avatars. Per Sundan, CEO of Pop House Entertainment, says this new technology allows KISS to continue their legacy for eternity.
He says the band wasn't on stage during the virtual performance recently in New York City, because that's the key thing of the future seeking technology. KISS could have a concert in three cities in the same night across three different continents. That's what you could do with this. In order to create their digital avatars, who are depicted as a kind of superhero version of the band, KISS performed in motion capture suits.
Experimentation with this kind of technology has become increasingly common in certain sections of the music industry. In October, K-Pop star Mark Twann partnered with Soul Machines to create an autonomously automated digital twin called Digital Mark. In doing so, Twann became the first celebrity to attach their likeness to OpenAI's GPT integration, artificial intelligence technology that allows fans to engage in one-on-one conversations with Twann's avatar.
Aspa, the K-Pop girl group frequently perform alongside their digital avatars. The quartet is meant to be viewed as an octet with digital twins. Another girl group, Eternity, is made up entirely of virtual characters, no humans necessary. What we've accomplished has been amazing, but it's not enough. The band deserves to live on because the band is bigger than we are. KISS frontman Paul Stanley said in a roundtable interview, it's exciting for us to go the next step and see KISS immortalized.
He can be forever young and forever iconic by taking us to places we've never dreamed of before KISS bassist Jean Simmons added. The technology is going to make Paul jump higher than he's ever done before. So the digital revolution has finally come for that last remaining part of media, the IRL live performance. I guess in a way, you know, concert movies have been a thing for a long time. Look at Taylor Swift just this year.
But people still want to spend big money to feel like they're breathing the same air as their idols, seeing them in the flesh. Although an alternative way to look at this is the humans of KISS can't live forever when they're gone, they're gone. So if you want the experience of seeing them perform live someday, something like this will be the only way you could do that. Is that good enough? Something something metaverse? Question mark? Nothing for you today. Talk to you tomorrow.