Michiel Frackers: Elon Musk launches Grok, the first product from his new company xAI

Imagine you have an extremely successful startup in 2023, like Sam Altman of OpenAI, and develop products like ChatGPT that many people around the world are excited about and Microsoft and investor Marc Andreessen are pumping billions into your company.

Oddly enough, you don't sleep well, knowing that any moment Elon Musk could introduce a competitor of which two things are certain; first that he can cough up or arrange enough money to invest billions and second that he has the talent and perseverance to tinker with it until it becomes something good. And oh yes, you also know that that competitor's name will be something with an X.

Because in addition to Tesla (with the model series S 3 X Y, read those letters as a word), Space X and X (formerly Twitter), among others, Musk still has time left to launch Grok this weekend, the first product of his new AI company xAI. There was a time when Musk was an investor in Sam Altman's OpenAI, but today it appears there is no love lost between the gentlemen.

The universe has a character?

xAI's mission is to "understand the true nature of the universe," and that kind of lofty talk creates obligations.

Unfortunately, the first version of Grok is still only accessible to users in the US. There is a waiting list, only it is still closed to me; perhaps registration for visitors from the US is possible.

It's hard to judge a product based on a 1-page website, but what stands out among all the rhetoric is the focus on efficiency. It seems like xAI wants to try to generate maximum output with minimal "compute," minimizing the need to invest in expensive and barely available chips from Nvidia. 

xAI does not conceal, quite in Musk's style, who it sees as its main competitors: 

"On these benchmarks, Grok-1 displayed strong results, surpassing all other models in its compute class, including ChatGPT-3.5 and Inflection-1. It is only surpassed by models that were trained with a significantly larger amount of training data and compute resources like GPT-4. This showcases the rapid progress we are making at xAI in training LLMs with exceptional efficiency."

A little further on, Claude, from Anthropic, is also briefly roasted and with that it is clear: it is game on for xAI against OpenAI (maker of ChatGPT),  Inflection (maker of chatbot Pi) and Anthropic (maker of Claude).

Meet Grok, the AI-tool that screams 'Look at me people, I'm something else!'

Musk will enjoy the spotlight on Grok, as he's been struggling lately with a lot of hassle around X, which seems to be losing as many billions in revenue as letters in the company name a year after he took over Twitter and privately things have been rather tumultuous, to say the least. This is not entirely unexpected, as the man apparently enjoys creating companies as much as offspring (now ten or eleven Muskies, the exact number varies by website).

"The result of the summit seems a promise to hold more summits."

The timing of the launch of xAi's first product is no coincidence. Musk attended the first UK AI Safety Summit last week. Prime Minister Sunak scored in his homeland as he succeeded in getting China and the U.S. to come to Britain, but Vice President Kamala Harris avoided the Chinese minister, and the final outcome of the big jamboree is summed up by New Scientist in the most British way possible: "the result seems to be a promise to hold more summits."

Tomorrow, Monday, November 6, OpenAI is holding its first-ever DevDay, a moment to inspire and highlight developers. For Musk, reason enough to want to ruin this celebration of OpenAI as thoroughly as possible with the launch of Grok this weekend. Unfortunately, at the time of writing this piece, it is not yet clear who is the "select few" who will get access to Grok, but apparently priority is being given to paying subscribers of X - that other X, that is, not xAI.  

Key Executive Order on AI

Unlike Prime Minister Sunak's mediocre Davos imitation, President Biden' s presidential order on AI presented Monday was a much more concrete starting point for serious policy. I think it's a good first step, but wonder why Biden does little to protect citizens' privacy.

Who controls how AI-systems are fed data, who has access to that data and how can disinformation be removed? The U.S. clearly lacks legislation in this case that compares to GDPR in the European Union.

Biden demands that so-called "red teams" be used by AI developers, teams of experts who mimic the attack or exploitation capabilities of a potential adversary and attempt to attack a company or system. The results of these attacks should be shared with the government. The president also proposes the introduction of an AI watermark so that users can instantly recognize AI-generated material.

It is striking how the various media analyze from their own perspective. I find the most thorough analysis to be that of Anjana Susaria, professor of information systems at Michigan State University, and Scientific American's broader perspective is also very good. In contrast, Crunchbase obviously looks at it primarily from an investor's perspective, while Brookings lets a parade of their smartest people loose but offers no overall conclusion.

I also found it remarkable that President Obama played a role in the creation of Biden's policy, whereas presidents usually have little concern for the opinions of their predecessors.

Open source AI models are promising

That developments in AI are happening many times faster than policymakers worldwide can keep up with, is evidenced by the staggering numbers in this article on the success of Llama, Meta's somewhat open-source version of AI.

Another open source variant, that of France's Mistral, has achieved appealing results, and Mistral is currently looking for money. For more than a quarter of a billion, actually. According to The Information, the amount the Parisians are looking to raise is $300 million, and according to Business Insider, they won't spit on $400 million at Mistral either, should anyone want to invest in it at a valuation of, seriously, $2 billion. For a company that is six months old. Then again, Paris is also an expensive city.

AI cares about us

The latest AI news for this week: scientific research shows that the results produced by LLMs like ChatGPT improve significantly if you indicate fear, or feel pressured, when entering the prompt. Answers become more truthful and responsible. This would imply that AI cares about our feelings ... for how long?  

The most complicated thing about Sam Bankman-Fried was his hair

Talking about truthful and responsible answers: Sam Bankman-Fried of crypto exchange FTX gave a pathetic performance in court, turning out to be a simple thief who managed to fool the world's top investors into thinking he was brilliant. There was much focus on his personal circumstances, as his parents got a lot of attention, especially from the mainstream media. Reactions to the verdict seem to indicate that more regulation of the crypto sector from Washington is on the way.

Bankman-Fried and his colleagues stole billions from FTX customer funds, which they used to invest in startups, for political donations and for loans to themselves. The fact that they ran crypto exchange is irrelevant. What was once again proven, however, is the time-honored cryptocurrency cliché: not your keys, not your coins.

In other words, if others can access your assets, they are already no longer yours. Putting crypto on a central exchange is the exact opposite of what blockchain can be; a decentralized network, without a center like a crypto exchange.

Kraken launches $200 million crypto investment fund

Despite all the headwinds for the sector and negative sentiments about crypto exchanges like FTX, the US crypto exchange Kraken has raised a serious investment fund. Crypto is no longer the favored sector among venture capitalists; that is, of course, AI. Investment in crypto this year is four times lower than in 2022. The timing of the fund could actually turn out to be great, because the crypto markets are rebounding strongly and appear to hold.

99-year-old Charlie Munger hates vc's

Warren Buffett's right-hand man hopes to turn 100 years old on January 1 and is no fan of venture capitalists: "you don't want to make money by screwing your investors and that's what a lot of venture capitalists do." So much for my hope that you get milder as you get older.

Munger says venture capital "can be a very legitimate business, if you do it right," and if you put the "right people" in positions of power. But that is not currently the case, he says. "The people making the most money from venture capital are a lot like investment bankers. They say what new, popular area they're going to invest in," Munger said. However, "they're not great investors - they're not great at anything."

Munger has a point, as the average return on venture capital over the last 20 years was 11.8%, compared to 12% for the Nasdaq Composite. Munger and Buffett themselves are having a nice weekend, as they announced a few hours ago that Berkshire Hathaway posted 40% more operating profit in the third quarter than a year ago and is sitting on $157 billion in cash. The old friends, and I say that respectfully, are profiting from buying short-term, high-yield government bonds.

SPOTLIGHT 9: JUBILANT WEEK ON WALL STREET

Wall Street closes its best week of the year with even more gains.

I think this is unprecedented in the 31 weeks I've been producing this newsletter: the entire Spotlight 9 is up. Wars, bombings, global tensions, investors apparently don't care as long as they think the markets will go up.

This is one of those weeks when everyone with a full savings account wonders if saving was smart with the horrendously low interest rate, especially corrected for inflation. So Buffett and Munger prefer to put it into short-term government bonds, perhaps we should all follow them. AP sums up all the jubilation in a short, lucid article.

IN OTHER NEWS

• Mark Zuckerberg, while living out his business and life motto "Move fast and break things," tore his ACL during a round of MMA sparring. He shared pictures on Instagram, but I'm still looking for his personal statement on the role of Meta social media networks in the spread of terror. But then he hides behind a generic corporate statement.
• Meanwhile, Threads, the failed Twitter copy of Meta, tries to lure fifty-somethings with pictures of mature women. I know, because it happened to me. The need to succeed with Threads must be high for Zuckerberg. Threads could have known that you lure me faster with videos of huskies.
• Just for giggles, check this out: British Prime Minister Sunak gave a speech on technology with audio quality worse than a pre-American Civil War sound recording. Funny enough, the video is still up on the official YouTube account of Downing Street 10. My guess is the intern who uploaded it had his headset on and was listening to something else than Sunak's mangled speech.
• Ex-Snap employee uses Snap Maps to contact people in Gaza. Not all technology was laughable or bad this week. It is heartwarming that despite political and religious divisions, this technology helps people to find and understand each other.

Michiel Frackers is the Chairman of Bluenote and Chairman of Blue City Solutions

www.bluenote.world
www.bluecity.solutions