Very well done. I lived in Japan for years, love Japan deeply, and this essay rang true in many ways.
Two thoughts:
- Japanese management style and processes are probably fruitful ground for understanding how teams of agents should work. H-firms require inspirational leadership, and agents don't need that.
- There is an interesting opportunity to turn Japanese process knowledge into a trainable environment, which of course should be done in such a way to benefit Japan and the Japanese people ("The type of deep process knowledge that has accreted within companies like Kyocera and Toto is almost impossible to replicate")
Historical scandals are finally coming to light now that the AI issue has raised awareness:
- Ernest Hemingway trained his own neurons on Tolstoy, Twain, and Turgenev without ever paying them royalties!
- William Faulkner trained his neurons on Joyce and de Balzac
- George Orwell trained his neurons on Swift, Dickens, and Jack London
- Virginia Woolf trained her neurons on Proust and Chekhov
Now that these historical wrongs have been exposed, it is obvious that some reparations are in order, likely from anyone who has benefited directly or indirectly from these takings!
If AI is overblown and permanently flawed, there is nothing to worry about.
If AI becomes as powerful as some fear/hope, productivity will be so high that we will need to do very little work for a superior standard of living. Costs for housing, healthcare, education will collapse, and there is nothing to worry about.
This article somehow tries to straddle both positions, that AI is fundamentally flawed and can never really accomplish useful work yet we should be angry and fearful.
There's this other case, which I think many people believe, that AI becomes as powerful as we fear/hope, and productivity is so high that only a few people need to work, but the superior standard of living is enjoyed by only those few people. We don't make the transition to post-capitalism and our economic dystopia is exacerbated and further entrenched.
Deleting data like that is a crime investigated by the FBI. In a very sad story, a brilliant former coworker made a mistake of deleting data after leaving employment and ended up in prison. Brilliant guy, momentary mistake. Overzealous employer.
Once they realized that DoW had locked themselves out of Mythos because of their beef with Anthropic, Trump invited Anthropic to the White House, and in that meeting they convinced Trump that Mythos is a big deal, and that China is distilling their models.
Excited to have a powerful tool, now they are saying it should be used by Government agencies first, and therefore, regulation.
Key takeaway: when defense types hear something is DANGEROUS, they want more of it. That's the outcome of discussing x-risk with the federal government. "Existential risk? That sounds GREAT! How can we get more of that, make it more dangerous?"
Yeah, it’s the same reason that Alex Karp goes on those unhinged apocalyptic rants about Palantir. It’s not for public consumption, it’s for defense insiders. The old logic prevails: a world destroying system is bound to exist, so WE must control it. Spare no expense.
> The old logic prevails: a world destroying system is bound to exist, so WE must control it. Spare no expense.
Except it is both true AND it works. Keeping your foot down on who can produce weapons-grade fissile materials is working out pretty damn well so far.
And the Russo-Ukranian War is proving any idiot with a few rubles can cobble together incredibly efficient combat drones. We need to be probing the limits of that yesterday.
It can both feel bad and be the right thing to do because the alternatives are worse.
Budgets are driven by congress from input by industry, senior military leaders, and current operations. Perceived risk is a completely unrelated political reality for uninformed voters.
> Budgets are driven by congress from input by industry, senior military leaders, and current operations.
So, yes? I can certainly see why defense contractors would prefer stability i.e. a cold war over a real war. But in a less dangerous world there's less need for defense contractors. e.g. '93 last supper
> Perceived risk is a completely unrelated political reality for uninformed voters.
Doesn't seem relevant to this thread unless we interpreted dangerous differently.
I'm working on some serious data analysis + realtime async code, and I use 200-400 million tokens a day with Claude Code alone (via ccusage). The complexity of the code seems to have a big impact on the number of tokens used. On simpler projects I use many fewer tokens.
My programming endurance is much greater now (2-3x focused hours per day), my productivity per hour is multiples higher, and I code seven days a week now because it's really exciting.
All told, I would pay for these tools as much as I would pay for full-time human programmer(s).
It's easy to see how this will play out. The entrepreneurs will get nothing. Most likely everyone else that has been paid (investors, etc) will keep what they received. Whether Meta or the CCP ends up with the proceeds of the entrepreneurs, that's anyone's guess.
How could we determine which device on mobile network is a faceless cellphone and which is a proper device needing real sweet Internet connection? And won't that make things more complicated than just v6 deployment?
Can argue that NAT, which interrupt layers ment for end device do basically the same as popular user hostinle unchangable mobile OSes, but I don't think latter is good either.
If you're assigning addresses, you can make the addresses in a ULA as short as you want. You're supposed to use a random 40 bit network id but if you can accept that you may need to renumber at some point there is no reason you can't use fd12:b:a:d::beef or whatever.
Because AI is attacking, plagiarizing, competing with, and destroying the most common industry of people here on HN, so suddenly it mattered more to people who were previously unaffected.
Some people have been concerned with this kind of politics all along. Some people are realizing they should be now, because of AI. And that's okay; both groups can still work together.
Two thoughts:
- Japanese management style and processes are probably fruitful ground for understanding how teams of agents should work. H-firms require inspirational leadership, and agents don't need that.
- There is an interesting opportunity to turn Japanese process knowledge into a trainable environment, which of course should be done in such a way to benefit Japan and the Japanese people ("The type of deep process knowledge that has accreted within companies like Kyocera and Toto is almost impossible to replicate")
reply