Enough on AI. Lets fly on a rocket!
Plus, well you know, boring old AI stories about training, money, impacts on our brains, Menu Mistakes and a closing thought
I never anticipated spending so much time on AI when I started this newsletter last fall. But the hubris and hype were too great to pass up. For those interested I have an AI article published through my new newsletter :
MAIN COURSE - Let’s distract ourselves by pretending we live in a era of rocketry
I have written a few times about Elon Musk and SpaceX - here and here - so I thought it would be time to look at other major rocket options, as NASA is looking for alternatives to SpaceX because of technical issues Elly and Donnie fighting. It is actually quite sad that the vaunted NASA of the 1960s is scrabbling around with Silly Con Valley robber barons in order to, you know, do stuff in space.
First up for consideration is Blue Origin of nifty name and oddball owner. Currently the bulked up Bezos - steroids anyone? - is taking over Venice to marry again, but his rocket firm hasn’t actually been covering themselves in glory. They can’t decide whether they are developing thrill rides to the atmosphere’s edge for ultra wealthy celebrities or serious rocketry to help NASA get to - wait for it - Mars. (The TechBros really took that pop pysch book Men are from Mars seriously)
Jeff’s rocket vision is long on narrative and short on actual real milestones, like launches. The first one in January actually got into someone’s definition of space but its booster rocket experienced a rapid unplanned disassembly (explosion in plain English). The second one planned for the spring was rescheduled until August. On top of this poor track record, are Amazon’s legal commitment for their Starlink competitor to launch over 1600 satellites by end of July 2026. Looks again. Yes 1600 rocket launches by 2026. Should be easy peasy if your name is Project Kuiper - named after the very distant Kuiper Belt - and you are only getting Katy Perry just to the top of earth.
Now there is another choice coming from left field, across the Pacific. Screaming in like a bullet is Honda, giving us shivers up the spine by reminding us of the go-go early 1990s when Japan Inc was where it was at. Turns out Honda - famed maker of motorcycles, cars, and riding lawn mowers - has turned their hand to rockets. They started aerospace research in 2019 and their first successful flight happened last week. Now we are talking. A legitimate manufacturer getting into rockets should give us all more confidence that Mars is within our grasp.
Their very successful launch was with a 20 ft rocket. WAIT A MINUTE. Doesn’t that sound kind of small to you? Anyway, this teeny rocket went up to the grand height of - hold your breath now - just under 900 feet. The height of 3 whole football fields in the air. I think I’ve seen teenage boys’s backyard contraptions go higher.
That’s what all these rocketry shenanigans seem to me: Boys Own Stories. Mars here we come!
QUICKBYTES: The cost of training and amazing chip engineering gives us what?
I have a couple of companion pieces here dealing with the prodigious amounts of money invested in AI.. The mostly US dollars spent on expensive chips, data centres and training for the AI models is the largest investment in any technology. Ever.
The models are so greedy for information that they are reading each others’s outputs. A new study shows that they are spending so much time scraping the websites of libraries, archives and museums that real human beings are unable to access these true warehouses of knowledge. These websites are often brought to their knees and have to go offline.
There are some that say this is just a price we have to pay to get us to AGI (Artificial General Intelligence, the meaningless standard that keeps being redefined).
So how do we square this insatiability of data with what the answer engines, as some AI tools like to be called, provide. I might be stepping on their precious toes a bit by looking at a study that shows what these engines are actually responding with. Given their multiple PhDs in every piece of known data, the list of what the answer engines are mentioning in their answers is……decidedly humdrum.
So we had to spend tens and tens of billions of dollars to get answers sourced from Youtube, Wikipedia, Healthline, Reuters and the real “obscure” one on the list - Reddit. Was it all worth it to just find more quickly if your favourite old actor is still alive, what’s in a Snickers bar, or details for a new shrimp cocktail recipe?
SPECIAL DISH: Even state propagandists are cheapskates and use free AI
I have had a lot of interest in the supposed truckloads of money being made in AI. I’m sorry to report - as is the case in the product cycle of almost every new technological innovation - that we are in hype stage of the cycle where NO money is ever made. I know it is hard to detect but there is quite a bit of AI hype happening.
There also is also NO monetization strategy in sight. This is on top of the enormous investments that have been made and are still being made. The inference process for operating AI tools on an ongoing basis are also very chip greedy. That means more costs.
Here is a quick back of the napkin example: OpenAI claims to have up to $10 billion in revenues but their costs are $25 billion. A tidy loss of $15 billion. Only 2 other companies had losses anywhere close to that - Warners with a mega write down - TV is losing to streaming as you might have heard - and the US Postal Service. Both were still $5 million less short, if you know what I mean. In addition, I’m also not prepared to put stake in those $10 billion in revenues until we have a couple of years of this revenue actually renewing.
Here is a key reason I think that things will change dramatically over the next 5 years - which is 500 in AI years: Everyone uses the free version. The more adoption of free we have, up go the costs. Plus increased usage of the free models drives costs up further. Substantially. With no more revenues.
Let me give you an unusual Exhibit A. Top of mind for many people is the Iran - Israel battle (which took a major fork in the road with the US bombing). The leaders of both countries, Iran and Israel, have been putting out lots of propaganda, much of a visual nature (as above). You know the type of stuff, videos of massive bombings causing damage to the other side. Shared everywhere. I’m sure many of you will be surprised to know that much of this video footage is fake, generated by everyone’s friend AI. But here is the hard to laugh at humour: they didn’t even use the paid version of the AI video generators. The VEO watermark for the free version is apparent on some of the footage.
MENU MISTAKES (ripped from Apple News)
I hope OpenAI becomes a for profit company as this sounds like my type of investment.
Brought to you by the BNPL (Buy Now Pay Later) experts Klarna who might actually be run by nothing but digital agents.
EXTRA HELPINGS: Another week. Another poor study for AI
As the studies come in, the issues and problems with AI and our usage of it keep ratcheting up. Here is a recent small study that compared 3 groups of users writing essays: having ChatGPT access, Google search access, or just the native human brain. Participants all wore a headset recording brain signals.
Some findings adding grist to the mill as we puzzle through our AI futures:
The essays written by students using Chat GPT were quite homogenous. The brain only essays showed a wider range of language and were heterogeneous in terms of quality as judged by teachers reading the essays
The AI students could quote little of their essays. Brain only students could quote about 90% of them
The EEG analysis showed that the three groups were using different neural connectivity and cognitive strategies.
Deeper learning came from the brain only group in terms of memory, accuracy and ownership
A LITTLE SPICE
Paradoxically in this age of online disinformation facts are easier to come by then ever
Bellingcat
Thanks once again to all my readers. I appreciate the comments, likes, messages and restacks from many of you. Next week will be a half year review of my 2025 Technology Predictions.
I really don't understand the love affair with rockets into space. Personally, I would be more than happy to take all the billionaires and send them to Mars and leave them there.
I hear Venetians are thrilled about Jeff's upcoming nuptials. 😡 I was leading a workshop recently for teachers and they were talking about how many students used AI to do their English homework (they were French). They don't even interface with the texts at all, so they are not learning anything! We talked about ways to engage with the content off-line and let them use AI for some elements of tasks to enhance/make it fun. Prohibiting doesn't work.