Benedict's Newsletter: No. 562

NO. 561 FREE EDITION TUE 8 OCT 2024

In this issue: In this issue: Nobels, OpenAI numbers, Yoox, the Musk problem, and problems of scale.

Read this issue as a web page. LINK

News

Nobel prizes for AI

Geoff Hinton, one of the pioneers of machine learning, was given the Nobel Prize for Physics (which raised a few eyebrows: yes, ML is a Big Deal, but why physics? Because it enables lots of new physics research). Sir Demis Hassabis, co-founder of Google DeepMind, and John Jumper, DeepMind research director, were given the Nobel Prize for Chemistry for Alphafold, which predicts protein folding. HINTON, DEEPMIND

OpenAI numbers

Following last week’s fundraising, The Information reports some more numbers: apparently OpenAI projects reaching profitability in 2029 at $100bn of revenue, with $44bn of cumulative losses to get there. Also, Microsoft’s investment terms give it 20% of OpenAI’s revenue. Microsoft’s Azure had $70bn of revenue in the last 12 months and AWS had $99bn, so a $100bn top line is not completely outlandish, though, of course, an awful lot of product, and capex, would have to happen to get there, and meanwhile the models look more and more like commodities. So, let’s see ChatGPT 5… LINK

Everything is hacked

The WSJ reports that a Chinese group hacked into the ‘lawful intercept’ systems run by US telcos for police wiretaps. This is a good example of why tech companies want end-to-end encryption: if they build in a back door for the police, it will get hacked by someone else. LINK

Someone hacked the Internet Archive and stole 31m user account details. LINK

The online fashion crater

Richemont, one of the big three high fashion conglomerates (with Kering and LVMH) had spent close to €5bn buying Yoox and Net a Porter over the past decade, and this week it sold the combined business to Mytheresa, a German startup, in exchange for a 33% stake, and handed over €555m of cash on top. Mytheresa’s own market cap was only $350m. A sale to Farfetch (which itself had to be rescued) fell through last year. Meanwhile, luxury companies are putting their money into premium retail spaces. LINK, RETAIL

Amazon Grocery

Amazon continues to iterate its grocery delivery plans, with an experiment in Phoenix, a US regional city, to let consumers shop from one set of inventory instead of splitting between Amazon, Amazon Fresh and Whole Foods. The underlying challenge, of course, is that Amazon’s whole model is is to treat all products as interchangeable SKUs in the same packetised logistics system, and grocery can’t fit that, since it has to be kept hot or cold, bruises, expires and so on, and yet it’s such a huge market that Amazon wants to find a away to try to make it work anyway. Note that Instacart’s quarterly gross went from $1.5bn before the pandemic to $7.5bn now. LINK

Google lost to Epic

Epic finally actually won a court case in its anti-app store crusade, getting a ruling in the USA that Google will have to end a bunch of things it does to hinder alternative app stores, and do a bunch of new things to help them. Just as for the EU’s attempts to open up app stores in Europe, though, I really struggle to see this changing anything much or solving real problems for consumers - it’s an argument from 2014, not 2024. LINK, RULING

Google Search antitrust

After Google lost this autumn, the next stage of the Google Search competition trial (as opposed to the adtech one, which just wrapped up) is remedies. That involves a lot of bargaining and posturing, and this week the DoJ made its first bid, arguing that, in principle, this should result in a change in the market (well, yes), and outlining its ideas. The one that got attention is the call for ‘structural’ remedies, which means breakups, namely splitting off Android and Chrome. As I and pretty much everyone else on earth pointed out, though, neither of those have their own revenue (except selling the default search slot, but Bing would now be the only bidder) and splitting them off would not necessarily change the core, self-reinforcing nature of Google search itself (they might keep Google the default, or users might switch back). And as I wrote at the time of the verdict, it’s actually quite hard to see what the court could do that would directly change Google’s market share. LINK, FILING, GOOGLE RESPONSE

The Musk problem

This week Elon Musk danced on stage with a racist pig, showed off some electric puppets and a mock-up of a car that will never ship… and caught a rocket ship in mid-air. And therein is the problem.

Elon Musk has been claiming that Tesla will have full autonomy working ‘next year’ for close to a decade now. Its bottom-up, vision-only approach might work at some point, but it can’t do a single unsupervised mile yet, while Waymo’s heavier model is doing tens of thousands of miles. This week, Musk showed mock-ups of a robo-taxi (an oddly impractical one) and robo-bus, and claimed they will ship in two years, but said nothing to make anyone believe that will happen on that timeline. Instead, Tesla had a bunch of remote-controlled robots on show - remote-controlled by actors, not AI. Remember when Nikola rolled a truck down a hill?

A few days earlier, Elon Musk danced on stage during a Donald Trump election rally: in the same week, Trump said he would ban autonomous cars. (To be fair, Trump says lots of things, most of them racist, incoherent or both, and maybe he thinks autonomous cars are driven by Mexicans, or people he sexually assaulted.)

And then, today, SpaceX caught its latest self-landing rocket (!) on a gantry (!!), in mid-air (!!!) You might wonder if flights to orbit have much price elasticity, or if there’s anything on Mars worth the trip, but that’s still quite a thing, and hence the problem. Musk says a lot of things that lose him respect, and has run Twitter into the ground, but there are the self-landing rockets! And the electric cars! And all of those can be true at once.

The car mockups LINK, SpaceX catches a rocket LINK

Kyle Vogt, co-founder of Cruise, on all the tricky execution issues between announcing a robotaxi and making it work. LINK

And a concept for an autonomous ‘bus’ from Ideo in 2014. LINK

About

What matters in tech? What’s going on, what might it mean, and what will happen next?

I’ve spent 20 years analysing mobile, media and technology, and worked in equity research, strategy, consulting and venture capital. I’m now an independent analyst. Mostly, that means trying to work out what questions to ask.

FIND OUT MORE

Ideas

Another big piece on South Korea’s problem with deep fakes. It seems to me that it’s too simplistic to say that this is entirely new and unprecedented, and also too simplistic to say this is just like Photoshop: we have been here before, sort of, but it hasn’t been this quick and easy before, and sometimes a change in scale can itself be a change in principle. (Perhaps there’s a parallel with social media - people always said silly or toxic things, but not with distribution at scale.) LINK

An argument that the wave of companies reselling and renting out GPUs for machine learning is a bubble that’s close to bursting. LINK

Dario Amodei, founder of Anthropic, wrote a very long essay on the possible consequences of the current wave of AI and perhaps ‘super-intelligence’. This mostly makes sense to me (and he tries to avoid sci-fi and apocalyptic ideas), but it feels a bit like saying “what will the internet mean?” in 1995. A lot, and it’s worth doing, and that’s enough (though maybe not enough for a zillion dollars of capex next year). LINK

Crypto investors were almost half of all corporate donations to US federal candidates in the USA so far this year. As I wrote a few weeks ago, crypto has dropped off most people’s radar and the scammers have moved on, but there is a group of pretty serious people who still think that this will be important financial infrastructure, and are exasperated by the SEC’s attitude that this can only be treated as securities. LINK

OpenAI released a report on ‘bad actors’ trying to use its products to generate spam/misinformation. LINK

Broadcasters are now looking at using connected TVs to offer street-by street ad targeting. LINK

Walmart on ‘Adaptive Retail’ and its AI plans. RETAIL, AI PLANS