raducu

> what these people are very good at is what AI is becoming very good at.

All knowledge work faces the same problem.

> Russia is in no hurry to capture the remaining 4-5%.

The grapes are sour.

Russia is absolutely in a hurry because the war is hurting it tremendously, it simply is unable to do so at the moment, but it hopes it can undermine Ukraine's external support, which is indeed going much better than its campaign on the ukrainian front.

> They've lost around 20% of their territory with no hope of getting it back.

I don't think that's the winning end game for Ukraine, but I also don't see there being a realistic winning end game.

One scenario would be that Russia f..ks of with that 20% taken and calls it a success (though strategically it is an enormous blunder), like they did in Finland and leaves Ukraine alone after that -- but is this realistic to happen?

Or Russia collapses internally and Ukraine does take back all the theritory.

Or the world steps in and secures a lasting peace for Ukraine.

Or Russia realizes Ukraine will outlast it because of external support and realizes how stupid and lose-lose this war is.

Or Ukraine collapses and Russia takes an even larger chunk and installs a puppet regime in Kiev.

But Russia seems unwilling to fuck off and call it quits because it's backed by China and has extended its influence in the USA and EU so it bets it can outlast Ukraine.

> What's the endgame here?

They can blanket band USING chinese models for all US/EU companies -- that's what they really care, not about chinese companies using chinese models.

Russia is the aggressor, absolutely.

However, let's talk realpolitik.

Ukraine innovates and Russia copies/addapts. Russia has ALWAYS been good at this game. Russia reminds me so much of the roman military in this sense, extremely good at copying and incorporating foreign military tactics/equipment.

Think of the initial long convoys ukraine destroyed, the bayraktars -- Russia addapted, the then the brilliant Eastern campaign where Ukraine managed to route the russians -- Russia adapted, that stopped working, then the indiscriminate massive artilery strikes Russia was doing -- ukraine started using better western artilery and himars, stopped the russians but the russians also addapted spread their logistics and so on; then the drones -- russia copied those even innovated with fiberoptic drones, then the Shaheds, ukraine adapted with fpv interceptors, then Russia started using jet powered interceptors, balistic missiles -- Ukraine initially stunned the world with Patriot interceptors but Russia kept firing until Ukraine ran out of interceptors, then the USV -- Ukraine had and still has great success with them, but look, Russia started using USVs also and Ukraine has no response to those.

It looks like the end will be decided by the country who just has more resources and willpower to continue.

I'm seriously scared for the next years -- the US is clearly turning hostile to Ukraine, Europe is turing far right with russian puppets everywhere.

> suggests it will just come back again in some other form.

They can pay the model provider, not amazon, but it would be less feasible because for now we have real competition among them.

History often rhimes but it doesn't always repeat itself.

I think there is genuinely a need for shopping agents and they could transform shopping in very good ways for the consumers.

Good thing for shoppers, bad for marketers, bad for people who relied on smarts, bad for margins, good for people who have capital and assets.

> the brine pond.

Then why not store the oil there directly instead of going all the trouble of pumping it in and out of the caverns? Is it so it doesn't rain on it?

Nobody said they were, but broken in terms of neurotypical circuitry, yes

On one hand I love your joke, on the other, this is HN not reddit and I usually downvote such responses, not sure what is the HN etiquette for such humor?

But the Turing test involves a conversation you can steer and I think that's where indeed it's a lot easier to smell LLMs.

> out-remembering people around us.

Doing an angular -> react refactor with Opus 5 at work.

Amazed at how good it is but also amazed at how dumb it is after I told it I want a very basic and specific code review:do the file names match the contents? is one functionality/concern split in multiple files? do the variable names match what they do? does a method read nicely?

Anyway, I was very surprised at how bad it can be at this.

I suspect it's because the LLM simply does not focus well enough on a single method.

My gut modelling is that the critical files are quite small, some 4k LOC and the LLM tries to read them all in one go and my requirement is surprisingly difficult but the LLM does not realize it, loads all the input into context and simply cannot find a good signal per token/group of tokens.

I guess I can do it because I literally cannot fit 4k LOC in my head and I actually read only tens of lines but I can reeealy dig deep in those lines.

The LLM can remember lots, it has 1M context window, was trained with needle in the haystack problems, but it can severely underestimate how much computation is required with TASTE/universal computation type tasks.

Yes, LLMs are incredible when it comes to delivering well defined outcomes, it's evident they were RL trained very well. But they are incredibly bad at understanding nuance in a text.

It's incredible how agentic coding benchmaxing seems to correlate with writing good poetry but at the same time finding good names for files and variables and wiring beautiful code does not.

Same with getting basic details of who said what in an "adversarial" conversation between 2 humans -- LLMs are incredibly bad at it, with all their "almost discovering new maths" skills

> I don't see a clear demarcation of the things that only "reasoning" can accomplish.

Producing beautiful, coherent, maintainable code. RL trained the models to produce correct code but boy oh boy is it ugly if it's 10k+ LOC

> Stand up a router.

it has to be some amazing router and while the models are open-weights, the knowhow to run them efficiently surely is not?

> So the price is rising and you have no choice but to keep paying more and more.

I presume you can't use deepseek?

> where are the 500 meter trees in the rainforests.

There could be structural reason as to why it's not possible, not pumping water *