My Bets and Counter Bets

The next 10 years

While I’m no oracle or gambler it can be fun to note some predictions or personal intuitions from time to time. It doesn’t mean that I’m dead set on these or that something can’t come in 6 months and completely change the assertions. But i’m reasonably confident and will be basing some of my long term decisions on these. Besides, it will be fun to come back 10 years later and see what I missed and where I was spot on. I have some cynicism and for a while believed the world is on the brink of collapse and it does feel that way. But after being that black-pilled for a while i cannot help but find the embers of the indomitable human spirit within. The more dark clouds overcast the sky, the more cold wind blows the brighter they shine bringing hope in face of despair. It’s easy to give up. It’s easy to justify doing nothing if everything’s going to shit anyways. Yet we persist and find hope.

Bets

• Face to face communication becomes preferred
• Crisis is narrowly averted
• Inference beats intelligence
• Satellite internet becomes the first choice for non 5G areas
• Robotics more mainstream but specialized

Counter bets

• No full self driving
• No white collar crisis
• No state cyber resilience
• No revolution
• No getting off scot free for the past insanity

Face to face

This is for both official and unofficial settings. Social media will eat itself. Algorithms tracking attention, feeding algorithms pumping out “content”. Until nothing is genuine. Until everything is just the same gray mush. Even before the gen AI this was the case. Think Skinner-boxes and echo chambers that are the social media platforms. Cost of producing professional-ish looking media will drop even more and the bar will be raised. If everyone can quickly generate a presentable looking visual, clip of video or music jingle, how do you separate the OK from the good and the exceptional (tangent: latter one I still feel will be doable. I believe that same as in coding artists who are there not only for the money enjoy and need to be in control and thus will add those final touches and little things done in taste that make the piece truthfully their own)? Thus, being there and being present and going the extra mile will mean everything. And sure it might be most efficient to outsource to the lowest bidder, but where anything great in life done by the lowest bidder?

Crisis averted

I don’t have much to reason here apart from believing in the good of the people. In the end it will be alright. Yes there are major wars going on, yes the political situation is tense, yes there is incompetency and malice everywhere you look. Still, I’m an optimist for some reason. If everything’s lost, what’s the point in doing anything and if there’s hope, then great things did turn out fine. In the meantime why not build, create, fight? Don’t let the bleakness stop you. It’s the hope and belief that got our kind trough tough times before. That’s the only solution we got, we fight and struggle and keep moving forward in spite of all odds. Again, no real logical reasoning here, just sharing the feeling I have.

Inference > intelligence

Of course the hot new topic for current time and looks for some time to come is GenAI and LLMs. I will make only a narrow claim here that for a while now all the gains have relied on different breakthroughs. First transformers, then reasoning, now we have harnesses. These are not guaranteed and there’s no evidence we will keep finding new ways to push the models. There’s also plenty of evidence that the current LLM labs are built on shoddy if not downright fraudulent financials. The market is full of hype. But as with computers the real revolution came when everyone could afford one, same with internet. I bet that cheap fast inference will get the most gains. My intuition is that exploring five 8/10 variants beats having one that’s trying to be a perfect one. And if we can get near instant feedback, let’s say I describe a feature and get 5 different implementations of it in a few seconds I’m way more happy and can quickly see which peaks my interest and explore that, ask follow up questions etc. This also goes to reducing the costs, so that on device inference is also near instant. Like VIM keyboard navigation but for thoughts. Imagine being able to locally embed all your files in a model and quickly ask it questions, converse with it get suggestions. I don’t know much about actual electronics, but my gut feeling is that frozen models kinda like FPGA but for model weights (already been developed). Maybe analog computing? I don’t know if transformers can be done with analog voltages, but feels that neural networks where it’s probabilities and matrices of weights and parallelism and it’s -1.0 to 1.0 instead of 1 and 0 fit here very well (those who dabble in silicon will likely scream at me, but again just my wild speculation as a complete noob in the field).

Satellite internet becomes first choice

This rests on the assumption that getting cargo to space becomes cheaper and small electronics stay cheap too. I’m not in the industry but it makes sense that laying fiber / building radio towers is not as cost effective as having a few satellites overhead. Imagine if with GPS we instead relied on radio beacons every ten kilometers to triangulate position (yes, I’m aware that using radio towers is how cellphones often triangulate the position). With new cellphones coming with satellite connectivity it stands to reason that as technology gets over the build out bump it will become more economically viable. The current systems (ie: starlink) already gather data. Do people need fast internet if a cheaper low speed one is available? I’d argue most normal friends would pay extra 20 if that means their slop movies don’t buffer as much. And others would benefit with this too.

Robotics more mainstream but specialized

In the same way LLMs are are very “fuzzy” text search, robotics is of-course gaining much attraction at the moment. Mixture of experts architecture etc. I’d bet that samish trends will come to robotics, of course with much longer timescales since integration, production of parts has way more bottle necks. But I’d imagine a common platform of let’s say robotic arms with modular tools (hands, suction cups, clippers etc.) with general / fine tunable specific models. I’ve bought a robot vacuum cleaner two years ago and never regretted it. It does it’s job and does it well enough. Would a humanoid robot get into the crevices and find all the little dust specs better? Maybe. But if it costs 100 times as much, works 5 times slower and risks kicking a vase over from the table It’s worse compared to a simple “dumb” robot vacuum. I don’t want a general “personal slave”. For me a device that, sorts clothes based on color, puts them into a washing machine, takes them out, puts to dryer and then finally neatly folds them is more than enough. Dishwasher is more than enough. There’s little more to do around the house (Maybe I’m underestimating since no children yet). And in industrial settings robots have already taken over. If there’s a human on the assembly line it’s either: cheaper or requires more dexterity. So I only see market fit for specialized machines around the house that help with repetitive well defined tasks. Even in my example there are caveats: does it check the pockets so you don’t wash banknotes? Does it zip up the zippers so that the clothes are not damaged. How does it deal with things getting inside bed sheets after washing?

Counter bets

No FSD

From my brief visit to LA it was pretty surreal to watch those waymos driving around. LA and US in general seems to be pretty favorable for self driving. Coming back in EU it feels way more in-depth road sign wise and road quality wise. It snows here, it heavily rains, roads are not as good. I’m not even going over third world countries with poor infrastructure. Waymos have sensors, they have the city mapped out, they drive in sunshine. Try that on a typical snow covered road with more specific signs and GPS maps that are not up to date. Even without self driving there are cases of “death by GPS”. Will a car be able to say “OK, the road is getting worse and I’m still 10km from destination, I should turn around here on the grass, wait, maybe not I can see water there, I might get stuck, let’s check what’s around that corner and if there’s no clean spot to turn around i will reverse”. Yeah. Tough luck. My concern isn’t even the accidents. Good safety systems with reaction times better than human already save lives. My concern is cars getting stuck in “common sense” situations and being annoyingly slow and unreliable to make decisions. I’m all for more safety and assist features (as long as they are not “your car won’t allow you to speed up because there was an ad with a picture of speed limit sign nearby” or “Government access to stop this car because there’s someone they don’t like driving”)

No white collar crisis

This is a big one for me and I’m the most confident about it out of all my predictions. There’s a lot to say, but I’ll stick to a few points here. Real job is very different from a well defined task. Developers underestimate the amount of tribal knowledge by a far margin. Humans have an insane “context window” and they know the “value” of information vs noise. Great take I’ve heard is that LLMs replace tasks, not jobs. I agree with it. Fuzzy search, summarization (same thing), writing small QOL tools. Automation based on clear evaluation and rapid feedback loops. These are where the technology shines. For judgment, intuition, experience you will need humans. LLMs miss obvious issues, they are confidently wrong and pursue paths that to any sane professional are obvious dead ends. Sure they can “stumble” upon a solution if there’s a clear end goal requirement, but that is fuzzy search in a loop. They are trained on the internet, but again any person doing a job for a few weeks will gather knowledge and get “trained”. Context window expands trough our whole life. LLMs have hard limits on context. Next best solution is retraining. Also there’s very little data on the internet that “compares” fuzzy data. Take two meals, two used parts two listings. Apart from price you don’t get much training in terms of which one is better. And it deused phonepends on the context too. But regular jobs have a lot of value judgments and at least from what I’ve seen the models are not there by a long margin. Go ahead try it out. Take two pictures of something you know the value of and compare them using the tools. My anecdotal evidence is that they are widely inconsistent, miss obvious details and have such different rating between sessions that they are of no use. Also AI bros widely underestimate the amount of small decisions and knowledge average worker has. They fail to take into account “ecology”. Like those small food delivery robots. Great on paper, laughable in practice. And not for a lack of trying.

No state cyber resilience

Contract workers with lowest bids for governments have caused immeasurable damage and weakness in cyber resilience. Legislation moves at an ants pace compared to technological advancements and it’s hard to attract talent or CARING people to the black hole that is the bidding for contract work. Unless it’s companies that already specialize in tech, the whole sector is so full of corruption and opportunistic players that outsource and undercut everything that to think the systems are any good is laughable. Also the software running there is ancient. I have a really bleak outlook on this. It’s a problem. It needs fixing. But regular people sadly learned to accept that software doesn’t work anymore and that government services are clunky. They don’t have to be. But the incentives are not there. They have little skin in the game if the code is shoddy and poorly architectured. As long as it somewhat works, it’s good enough. And there’s no use keeping it up to date or refactoring. And the people in power change so often that no one is ever responsible.

No revolution

At least where I live. Old people rule and youth is ever shrinking. It’s like a long nightmare that never ends. It always feels like the breaking point is nearer and nearer but it never comes. And as we get older, have less energy and more and more to loose and get more comfortable in our mediocre lives nothing is soon to change. Reluctantly I must say that only slow change with maybe some areas of life going in a good direction because of youth building their own spaces and solutions will come. Other than that we will live on legacy pile and spread our roots there. Maybe the main problem is that no one has a better outlook. A better plan. Will to risk it. Watch hyper-normalization by Adam Curtis for more context on that feeling about state of world.

No getting of scot free for past insanity

This is directed mostly to LLM players and grifters. Wide ranging statements, the insane irresponsibility around software practices and financial manipulations will come to an end. You can only bend reality so far before it snaps back and hits you in the face. And the more you bend it the more cruel and debilitating that snap will be. Market is insane, promises are insane, costs are insane, debt is insane. Maybe it’s due to the capital being in old peoples hands where they see technology as magic. And once a domino falls there will be no stopping it. A default on a single data center debt, disappointing IPO results, growth slow down for a major player. There’s no telling when, but I for one just want the insanity to be over and to get back to regular technology work which is rewarding because of being genuinely useful and not because there’s money to be made on speculation and promises with no data to back it up.

My conclusion

There was a sort of meme / divisive question some time ago.

In front of you appears red and blue button. If more than 50% of the people presses the red button, everybody who pressed the blue button dies. If more than 50% of the people presses the blue button, nobody dies. What button will you press?

Brain says red. Of course it’s the only logical and defendable choice. Yet I can’t help myself. Idiotic sacrifice? Yes. But isn’t it what makes us human? The most famous story in the world is about a guy who sacrifices himself for the world and dies on the cross. All the greatest stories of mankind are.

Even in a strange game when the only winning move is not to play, the true human spirit will find a way. We will break the chess board, make a queen out of a twig we found in the dirt with bare teeth if needed. In the end it’s going to be alright. It always is. Maybe not in the way you hoped for, but it always is.