1. Why does AI need that folder structure? Why not a flat list of files and let the AI agent explore with BM25 / grep, etc.
2. pre-compute compression vs compute at query time.
Kaparthy (and you) are recommending pre-compressing and sorting based on hard coded human abstraction opinions that may match how the data might be queried into human-friendly buckets and language.
Why not just let the AI calculate this at run time? Many of these use cases have very few files and for a low traffic knowledge store, it probably costs less tokens if you only tokenize the files you need.
> Why does AI need that folder structure? Why not a flat list of files and let the AI agent explore with BM25 / grep, etc.
Progressive disclosure, same reason you don't get assaulted with all the information a website has to offer at once, or given a sql console and told to figure it out, and instead see a portion of the information in a way that is supposed to naturally lead you to finding the next and next bits of information you're looking for.
> use cases
This is essentially just where you're moving the hierarchy/compression, but at least for me these are not very disjoint and separable. I think what I actually want are adaptable LoRa that loosely correspond to these use cases but where a dense discriminator or other system is able to adapt and stay in sync with these too. Also, tool-calling + sql/vector embeddings so that you can actually get good filesystem search without it feeling like work, and let the model filter out the junk.
> let the AI calculate this at run time?
You still do want to let it do agentic RAG but I think more tools are better. We're using sqlite-vec, generating multimodal and single-mode embeddings, and trying to make everything typed into a walkable graph of entity types, because that makes it much easier to efficiently walk/retrieve the "semantic space" in a way that generalizes. A small local model needs at least enough structure to know these are the X ways available to look for something and they are organized in Y ways, oriented towards Z and A things.
Especially on-device, telling them to "just figure it out" is like dropping a toddler or autonomous vehicle into a dark room and telling them to build you a search engine lol. They need some help and also quite literally to be taught what a search engine means for these purposes. Also, if you just let them explore or write things without any kind of grounding in what you need/any kind of positive signals, they're just going to be making a mess on your computer.
Maybe it depends on the use case, but my opinion is, if you do need to apply compression, it should be done via a tool call real time instead of in a pipeline.
For example, if you’re trying to summarize the status of a project, instead of feeding an agent (in real time or via summarization pipeline), it’s better to write a script that summarizes the status of all of the jira tickets, instead of asking the agent to read all of the tickets to create a summary
Another small data point, I think people would prefer to ask questions of an AI model instead of reading the generated summaries.
> Why does AI need that folder structure? Why not a flat list of files and let the AI agent explore with BM25 / grep, etc.
It doesn't. The human creating the files needs it, to make it easier to traverse in future as the file count grows. At 52k files, that's a horrendous list to scroll through to find the thing you're looking for. Meanwhile, an AI can just `find . -type f -exec whatever {} \;` and be able to process it however it needs. Human doesn't need to change the way they work to appease the magic rock in the box under the desk.
why? The human would just talk to the AI agent. Why would they need to scroll through that many files?
I made a similar system with 232k files (1 file might be a slack message, gitlab comment, etc). it does a decent job at answering questions with only keyword search, but I think i can have better results with RAG+BM25.
Just because AI exists doesn't mean we can neglect basic design principles.
If we throw everything out the window, why don't we just name every file as a hash of its content? Why bother with ASCII names at all?
Fundamentally, it's the human that needs to maintain the system and fix it when it breaks, and that becomes significantly easier if it's designed in a way a human would interact with it. Take the AI away, and you still have a perfectly reasonable data store that a human can continue using.
I used GLM5 quite a bit, and I'd say it was maybe on par with Sonnet for most simple to medium tasks. Definitely not Opus though. Didn't test super long context tasks, and that's where I would expect it to break down. A recent study on software maintainability still showed Sonnet and Opus were peerless on that metric, although GLM series of models has been making impressive gains.
If everyone turned to medical professionals the system would fall apart. Many try but fail to find a space in the current system. Internet forums / llms offer another path that many have used to move forward.
Its like telling people don't talk to friends and family only go to a professional. This is how we end up in a worse mental crisis.
There is a huge difference between friends and family, whom you have known for a long time and hopefully have a vested interest in your success than an anonymous internet stranger.
> People really should turn to medical professionals and not internet strangers for help.
For your own safety, you are to only discuss your health with NHS certified healthcare providers and no one else. Doing so with others can lead to unprovoked, unsanctioned and dangerous anecdotes, advice and memes. Worse, you might find community and make friends with people who have similar life experiences, which can distract from your state sanctioned treatment plan. This extends to your friends and spouse, they might mean well, but they are not medical professionals.
Your health is your private matter, let's keep it that way!
Your sarcasm sounds like we'd be living in some extreme 1984-like dystopia, while the reality is: there is more bullshit one can read online than a bullshit one can get from a poor doctor/therapist.
From my experience, the best one can do is to get a good and affordable therapist by a word of mouth.. and sometimes one can get lucky as such person is doing also service for a state, for free. Main point is to actively start searching
Main point with depressed people is, they often do lack initiative and enthusiasm to go find that unicorn therapists. When you are struggeling already in general and then you have to struggle even more to maybe find help (and then you don't know whether you can afford it) .. no wonder people turn to online help and LLM'S.
So yes, you are right in that activly start searching is the better way. But that insight is often lost on the target audience.
Well in my European country therapists hourly rates are very flat and practically the same as hourly rates for physiotherapists, massage practitioners and basic doctors, plus an LLM can really quickly tell that, so it is much faster to get to know if one can afford it than to look for a general solution of the whole problem.
1/ When authors use AI for editing, it reduces their credibility.
2/ As much as I don't like the current administration (and Israel leadership), there is absolutely no way the assumptions this article makes about them are false.
There is no way the US/Israel didn't calculate that:
- the straight would be closed
- a new leader may represent similar idiologies of the past leader.
Everything that has happened so far (in regards to Iran attacking neighbors) has been extremely predictable. There is just no way these weren't calculated in.
>there is just no way these weren't calculated in.
the American government is publishing war footage intercut with Call of Duty scenes. The American secretary of defense is a former television personality with more tattoos than people in a trailer park. He said rules of engagement are stupid because they stop you from "winning" while the US bombed a girl's school.
They literally fired the people who calculate things and wage war based on memes, vibes and chatgpt recommendations
> There is no way the US/Israel didn't calculate that: ... the straight would be closed
It has always had this potential, as it has happened before: see https://en.wikipedia.org/wiki/Operation_Earnest_Will (1987). But based on this history I would assume that many in the admin did not find the threat as credible as it was then. We dont seem to have a good grasp on how things have gone in the black sea. We clearly did not anticipate the level of drone attacks that have been put out by Iran.
Nothing says "we did not have a plan" when easing Russian sanctions while you ask Ukraine for help with defenses.
> a new leader may represent similar idiologies of the past leader.
I could see making a bet that with the current water crisis there the this would tip them into an "Arab spring" moment. For any one aware of the history there, it was a poor one at best.
If the US decided that stopping oil production in Iran was important (restricting global oil supply), what other options does the US have ease the impact on oil prices other than Russian sanction easement?
Yeah, it looks bad, but there just isn't really any other ways for the US to magically pump more oil out of the ground instantaneously to compensate for the war.
That's exactly what we are doing, releasing the Strategic Petroleum Reserve.
The fact that it's both a record release and still not sufficient suggests they underestimated how bad it would be. But the US did prepare for this eventuality. And now we're throwing away our ability to be prepared for any upcoming crisis.
Maybe the US military commanders, generals and Pentagon knew this but the civilian leadership at the top chose to completely ignore it and can't really articulate a plan or what the plan ever was.
This conflict was a long time coming: Trump claimed Biden or Obama will start a war in Iran and that is why they are weak presidents. Trump sees himself as a peacemaker (flying in to negotiate deals with TH and KH, negotiating Ukraine war, etc).
I think there is more going on to cause Trump drastically change his self-image.
I don't think this is a Trump administration driven decision.
All reports are saying the US generals were against this. And a UD senator (Graham I think) just admitted he lobbied trump for the war, comparing him to Roosevelt, and coached Netanyahou on how to lobby trump. Just look at the article:
> There is no way the US/Israel didn't calculate that:
I don't really believe the buffoons in US leadership calculate much. It's all vibes.
I firmly believe it will become a case study in how many ways a comically incompetent government can damage a country.
As for Israel... I think their calculation is simple. They don't really care about how much damage they cause to the world economy, as long as they get to kill Muslims in general and Iranians in particular. They want death.
Israel will aggressively destroy anyone who attacks or intends to attack them.
They have peaceful relations with Muslim nations, Jordan and Egypt especially.
I acknowledge Israel's current two decade strategy with the Palestinians is not kind, but they aren't cartoon villains that just want to kill Muslims.
If its all vibes, then how does trump hungry for a world peace prize vibe with the war? Or the many clips of Trump trashing Obama and Biden for potentially starting a war in Iran?
Trump calling something bad but doing the exact same thing he talked against? No, I can't believe it. What a surprise. This definitely never happened before. At least before December 2025.
The Chairman of the Joint Chiefs of Staff is reported to have recommended against further air strikes on Iran[1].
----------
"Chairman of the Joint Chiefs of Staff, Air Force General Dan Caine, has warned that strikes against Iran could be risky, potentially drawing the US into a prolonged conflict, US media report.
Caine has reportedly cautioned that a military action could have repercussions across the region, potentially including retaliatory strikes by Iranian proxies or a larger conflict that would require more US forces.
In a lengthy post on Truth Social, Trump described the reports as "fake news".
I agree that many people inside and outside the US gov didn't want this war for various reasons, but of the people that wanted this war, they must have calculated these very obvious risks.
The article touches on this topic, but my guess is Iran isn't part of the USD/petrol trading. If the US can convince the new leadership in Iran to start trading in USD, then that would be very good for the USA (and bad for CN, RU, and IN).
If the POTUS starts a war against the advice of the The Chairman of the Joint Chiefs, he should have a plan beyond "wag the dog" to distract from the Epstein files. I'm not convinced Trump actually has a plan unfortunately.
I think you give too much credit to the US and zionists. They probably convinced Trump that it would be another Venezuela, and because of their hubris they decided to go for it anyway. Remember how at the beginning it was supposed to only last for 2-3 days? Then 1 week, then 2 weeks, then 4 weeks, then until September. They clearly didn't see this far.
I'm very surprised anyone would think Iran would be a Venezuela.
Venezuela's leadership was barely legitimate (with voter fraud / dictactorship 3 years ago) whereas the supreme leader in Iran has had power for 36 years.
> There is no way the US/Israel didn't calculate that - the straight would be closed, a new leader may represent similar idiologies of the past leader.
A few things to remember here. First, Israel and US have divergent strategic goals. (Well, that presumes the US has strategic goals, which appears to be false given the struggle the administration has had over the past week to explain why the fuck we're at war with Iran.) Israel's apparent goal is the complete destruction of the Iranian state, and Netanyahu certainly seems to believe that Israel will suffer no consequences as a result.
The second is that Trump has never faced any consequences for his actions. If anything goes wrong, he just lies and says that it's all right, changes the topic and since no one talks about anymore, hey, it's been fixed. It also seems as if he believes that nobody else truly has agency, so the idea that the enemy gets a vote in war may truly be foreign to him.
Note also the quality of people that Trump has surrounded himself with in this term. The head of the military is someone who washed out of the military officer corps (and also essentially failed in every managerial career he's had since them). They openly denigrate the importance of things like logistics in military, in favor of big, manly things like the awesome power of their missile salvos. I believe Hegseth legitimately doesn't give a crap about the boring things like naval escort missions because that's not manly, and instead cares more about how much big kaboom has been delivered to Iran, and so far the evidence of how the operation has gone to doubt completely vindicates that belief.
Fourth, even almost two weeks into the strait being closed, the US military has done nothing to reopen it. The strait is not closed because of the existence of mines, or because Iran is targeting ships; it is closed because shippers are absolutely terrified to send their ships through it. Reopening it thus requires giving those people confidence to send their ships through it, and that confidence of course requires clear, public statements. That is not happening. Instead, we get Trump giving off a different explanation of how to reopen it everytime he's asked, followed up by the US Navy denying whatever Trump said (e.g., the US Navy is unwilling to provide any naval escort). There is insufficient materiel in the theater right now to reopen the strait, and nothing is being shipped to the strait that can reopen it. From all apparent evidence, the current plan for reopening the strait is praying that it reopens tomorrow, although I have doubts that there is enough self-awareness or religiosity to actually do any praying here.
The risk of Iran closing the Strait of Hormuz is so obvious, the catastrophe of such an action is so well-known, that you would have to be a colossal idiot to go into a situation where Iran might plausibly close the strait without a plan to reopen it swiftly. And yet all available evidence leans in that direction. So now many, many people are forced to countenance the sobering idea that the US government is led by an idiot who will destroy the economy without realizing that's what he's doing. It's time for us to wake up to the fact that there are no adults in the US government anymore and do something about that.
yeah, this headline doesn't pass the sniff test to me too:
The traffic cameras can't tell you who is in the vehicle. Maybe they know which plate he typically rides in? But the much simpler explanation is he had a leak.
If the cameras can see the driver, they can also see the front passenger. And who gets in or out of the stopped car. Istael/US probably analyzed months of footage to establish patterns and uncover their habbits and schedule.
Most VIP situations, the VIP person sits in the back. Luxury car brands like bentley or limos are specifically designed for a comfortable back seat. The backseat typically has much darker windows
Yes but they also make a show when they get in or out of the vehicle. Several bodyguards flanking them, opening doors for them, the driver getting out of the vehicle, red carpets etc. Khamenei was also wearing distinctive clothing to make it even more obvious.
It's also possible that they simply went ahead and hit the vehicle, knowing there was a good chance that their target would be inside. It's not like they've been all that picky about collateral damage.
reply