I've got the strong feeling that AI model and agent requires different operating system (OS) paradigm that's data centric rather than file-system for more efficient, effective and trustworthy operations. This new OS should work seamlessly with data natively across different processors for examples CPU, GPU, TPU, NPU, accelarators, etc.
For working example, please check TabulaROSA (Tabular Operating System Architecture) proposed by the MIT team. Instead of normal OS system call, it utilizes data based operations with D4M that can work mathematically via associative array with structured or non-structered data [1],[2].
With the advent of new CPU acceleration with fully homomorphic encryption as demonstrated by Intel, the AI model and agent can even analyze the data without even decrypting them [3],[4].
[1] TabulaROSA: Tabular Operating System Architecture for Massively Parallel Heterogeneous Compute Engines
>As an example, North America is a huge area to make a book for!
I think AI/LLM with RAG is the ideal solution for this mushroom hunting/foraging.
With mobile phone fine-tuning multi-modal app capability and NTN satellite connection to the cloud, you're loaded for the bear, err mushrooms.
Check this video on Golden Chanterelle mushrooms hunting in Santa Barbara, Southern California where one lb can cost around USD20-30. The guys managed to gathered around 80lb for the trip or about USD1600 minimum if sold [1].
[1] Mushroom Hunting Catch and Cook (80lbs Found!):
Palantir is the designer of the lethal US missile targetting system that has ten years outdated data information [1],[2],[3].
For the love of God, who's the Palantir design architect that approved and relied on a single (outdated) database information system for mission critical missile operation?
[1]>In 2018, more than 4,000 Google employees signed a letter opposing the company’s contract to build artificial intelligence for the Pentagon’s targeting systems. Workers organised a walk out. Engineers quit. And Google ultimately abandoned the contract. Palantir Technologies, a data analytics company and defence contractor co-founded by Peter Thiel, took it over and spent the next six years building Maven into a targeting infrastructure that pulls together satellite imagery, signals intelligence and sensor data to identify targets and carry them through every step from first detection to the order to strike.
[2]> A chatbot did not kill those children. People failed to update a database, and other people built a system fast enough to make that failure lethal. By the start of the Iran war, Maven – the system that had enabled that speed – had sunk into the plumbing, it had become part of the military’s infrastructure, and the argument was all about Claude.
[3]>The building in Minab had been classified as a military facility in a Defense Intelligence Agency database that, according to CNN, had not been updated to reflect that the building had been separated from the adjacent Islamic Revolutionary Guard Corps compound and converted into a school, a change that satellite imagery shows had occurred by 2016 at the latest.
Recently there's HN discussions on the topic of local AI/LLM being utilized by researchers from IEEE Spectrum magazine, probably worth a look up [1], [2].
[1] Local AI is driving the biggest change in laptops in decades (260 comments):
>It is absolutely stupid to talk about this as edisons revenge. If Tesla had the modern high power transistors needed to get high voltage dc out of the ac produced from a spinning turbine he would be all for high voltage dc too.
This!
The soon people realized these facts the better. The pervasive high rise buildings did not happen before the invention of modern cranes.
Exactly twenty years ago I was doing a novel research on GaN characterization, and my supervisors made a lot money with consulations around the world, and succesfully founded govt funded start-up company around the technology. Together with SiC, these are the two game changing power devices with wideband semiconductor technology that only maturing recently.
Heck, even the Nobel price winning blue LED discovery was only made feasible by GaN. Watch the excellent video made by Veritasium for this back story [1].
[1] Why It Was Almost Impossible to Make the Blue LED:
Gallium is expensive to extract because it is extremely diluted in the environment.
It accompanies in very low quantities aluminum and zinc, so it is extracted only in the mines of aluminum or of zinc, as a byproduct.
However, the abundance of gallium is similar to that of lithium, while gallium is used in smaller amounts, so there is no risk to not have enough gallium in the near future.
On the other hand, all semiconductor devices with gallium also use some indium. Indium is used in even greater quantities in all LCD or OLED displays, to make transparent electrodes.
Indium is an extremely rare element in the entire universe, comparable with gold, so for indium there is a much greater risk that its reserves will become insufficient.
This could be mitigated by extracting such critical elements from the dumped electronic devices, but this is very expensive, because only small amounts of indium are used per device, so very large amounts of garbage would have to be processed in order to extract a sizable amount of it.
There's a component of modern culture that trains and expects people to be extremely pessimistic about long term human development. It results in situations above, where without any further information people just assume by default that were going to run out of a thing and are on some collision course with not just a disaster, but every single conceivable one.
(Gallium is a byproduct of aluminum production. We aren't going to run out.)
My understanding of most elements is if we want more it’s either pretty easy to make from something else we have a lot of, or we need to redo the Big Bang, the latter being, in my opinion, a bit of a disaster scenario.
Even synthesizing helium is prohibitively expensive. Unless you want whatever heavy decay products we have from nuclear waste, synthesizing elements at industrial scale probably isn’t happening.
Unless by “make from something” else you mean extract the element from existing chemical compounds found in Earth, in which case we’re still just using existing deposits on Earth.
On the other hand, it is possible to run out of a metal when all of it is either somewhere in some device or scattered among landfills (i.e. not concentrated in a place like a mine).
That is true, but gallium is present in the aluminum and zinc ores only in minute quantities.
We will not remain without gallium, but it is impossible to scale up the gallium production to a higher level than provided by the current productions of aluminum and zinc.
So there is a maximum level of gallium that can be used per year and it would not be possible to increase the production of blue and white LEDs and of power transistors above that level.
Fortunately, the amount of gallium used per device is very small, so it is not likely that we will hit that level soon. A much more serious problem is the associated consumption of indium, for which the resources are much less.
Practically speaking, sure. It's obviously not cost-effective to extract it. But it's there if someone can get it. I don't expect anyone to be extracting gold from ocean water, but there are other source of other elements that may not be cost-effective now but could be in the future or may simply become necessary despite the cost.
Cost scales with refinement effort, so it just results in more expensive TVs. That said, pretty sure we'll have drowned the planet in landfilled TVs long before this becomes a serious issue
From your earlier comment, your curiosity was more about what happens after we run out.
In your question you stated the running out as a given fact ("When" we run out, not "if").
If that was what you wanted to say I can't tell you, but that's definitely how it was received and thus you also got the harsh response. Since it reads a lot like doomsday thinking.
(Example: Does that mean when we run out of oxygen there are no more humans?
Yes, my curiosity was about when we run out, because I didn’t know if we would run out. That was the whole point of the question. Have some leniency, we’re not all experts about everything.
Sidenote: Whenever someone tells you that (vital) reserves of some ressource are going to run out soonish (implying drastic consequences), you should be extremely skeptical:
Such predictions have an abysmal historic track record, because we tend to find workarounds both on the supply side (=> previously undiscovered reserves) as well as flexibility on the demand side (using substitutes).
This applies historically for oil, lithium, rare earth metals and basically everything else.
edit: I'm not saying we're never gonna run out of anything-- I'm just saying to not expect sudden, cataclysmic shortages in general, but instead steadily rising prices and a somewhat smoothish transition to alternatives.
I always add "cheap" to the sentence. It seems they are always talking about the cheap version of anything. Going to run out of water? Or are we running out of the "cheap" version of water that does not have to be processed?
This is a valid point: quickly depleting reserves often indicate that pricing is not sustainable. Which is bad.
But non-sustainable pricing is very different from "cataclysmic collapse", and too many people expect the latter for too many things, which is just not realistic in my view (and historical precendent makes a strong case against that assumption, too).
A society where water prices gradually increases to "reverse-osmosis only" (instead of "pump-from-the-ground-everywhere") levels is very different from a society where water suddenly runs out.
> Such predictions have an abysmal historic track record, because we tend to find workarounds both on the supply side (=> previously undiscovered reserves) as well as flexibility on the demand side (using substitutes).
That's a classic example of the "preparedness paradox" [1]. When no one raises the alarm in time or it is being ignored, resources can go (effectively) exhausted before alternatives can be found, or countries either need to pay extraordinary amounts of money or go to war outright - this has happened in the past with guano [2], which was used for fertilizer and gunpowder production for well over a century until the Haber-Bosch ammonia process was developed at the start of the 20th century.
And we're actually seeing a repeat of that as well happening right now. Economists and scientists have sounded the alarm for decades that oil and gas are finite resources and that geopolitical tensions may impact everyone... no one gave too much of a fuck because one could always "drill baby drill", and now look where we are - Iran has blasted about 20% of Qatar's LNG capacity alone to pieces and blocked off the Strait of Hormuz, sending oil prices skyrocketing.
I don't see the Guano industry as a straight counter-example, it even illustrates my point:
If you had made predictions/scenarios in 1850 based on Guano deposits running out within a decade or two, you would have mispredicted completely, because a lot of the industry just transitioned to sodium nitrate (before synthetic fertilisers took over). Nowadays media landscape would've gladly made such doom-and-gloom predictions for global agriculture back then.
I completely agree that quickly depleting reserves often indicate non-sustainable pricing for ressources (which is obviously bad long term), but that is very different from sudden collapse.
I've seen articles from the 1880s claiming oil will run out by 1890. 140 years latter...
Yes we can run out of oil, but nobody really knows if or even when that will happen. Right now I'm guessing we won't run out because wind and solar is so much cheaper for most purposes everyone is shifting anyway - this will take decades to play out.
> Yes we can run out of oil, but nobody really knows if or even when that will happen.
We can run out of cheap and accessible oil very, very fast if the shitshow in MENA continues to escalate. Qatar already lost 20% of their LNG capacity in a single strike.
The US may have enough domestic oil production to sate its domestic demand, but the prices would still skyrocket even for them. Europe meanwhile, we're straight fucked here. Technically the oil hasn't run out, it's still in the oil fields of the journalist-butcher country and other sheikdoms, but that doesn't matter if it cannot be pumped out any more because the wells got blasted to pieces or if it cannot be transported thanks to Iranian mines, Europe is still running out of oil in practice.
What does "Europe running out of oil" mean to you? Gas at the pump for >10€/l, potentially with some rationing scheme? Do you honestly think that's gonna happen?
It is easy to get infected by the media narratives that are notoriously biased towards maximum drama, but I firmly believe that we are not gonna escalate into such a scenario.
There's always options; sorting priorities because of price, radical electrification of transport, or, at the extreme end, picking up coal hydration again (worked well enough to keep the Nazi war machine running for quite a while, with much worse access to crude).
For comparison: Copper prices did increase by 500% since 2000, but people barely even care, and that's how I would expect "shortages" to typically go.
"Reserves" are the name of something that exists only at a set price. Change the price, and the reserves change too.
The people that rush to tell you that reserves are running out tend to omit what price they are talking about. That way of expressing oneself is normally called "a lie".
>I built my own training GPT and passed in 13 days, which was a company record. The training manager knew I'd built the app but never showed an interest, which makes sense: it could replace about a quarter of his role.
I'd really love to read a dedicated article on this side project.
Apparently, Karpathy is into AI based education business with Eureka Labs [1].
The training was basically to sit and read books all day, and it could be so much faster. Surprisingly there is no reliable test / quiz prep tool available online, just some free ones that contain errors.
Given this company is basically the best, they should really build their own revision / quiz tool. The most valuable part of training was a webinar which I took notes from, and turned into revision cards alongside info from the books and revised the way I know works for me. They could do this bespoke for each trainee in seconds now.
All the comparisons are with scripting and untyped languages perhaps for faster development and more intuitive eco-system to increase developer productivity.
In the age of IntelliSense, auto-completion and AI assisted coding, does the choice of scripting and untyped language justifiable for increased in productivity at the expense of safety and reliability?
If you're building data system not just for exploratory, surely modern compiled and typed system languages like Rust and D language make more sense for safety and reliability for the end users?
Even more so with D language where you can even have scripting capability for exploratory and protyping stage with its built-in REPL facility [1],[2]. This is feasible due to its very fast compile time unlike Rust. It has more intuitive "Phytonic" syntax compared to other typed languages [3]. You can also program with GC on by default if you choose to. Apparently, you can have your cake and eat it too.
One general problem or challenge with statically strongly typed languages is, that one can quick get to a local optimum, but that local optimum might lack some flexibility, that is needed later on, only discovered after some usage and seeing many use cases. Then a big refactoring is ahead, possibly even of the core types of the project. If that is allowed and introducing such flexibility thought of, it often happens, that expressing it in types becomes quite complex, which, without a lot of care, will impact the user of the project. The user needs to adhere to the same types and there might then be quite some ceremony around making something of the correct type, to use it with the project.
It is safer, but it is not without its downsides. It demands a careful design to make something people will enjoy using.
If you're "building data system not just for exploratory" then you're probably not going to be using any of the presented options. However, in my experience Clojure has an ecosystem where there it is very easy to transition from exploring/playing with data at the REPL to a more robust "pro" setup that's designed to scale, handle failures, etc.
I understand the sentiments but I disagree with the approach, it's probably efficient for exploratory but not effective for everything else including prototyping and systems development.
For any engineering work, including software engineering you choose the best tool for the job. In D you can have the high performance tool capable of bit shifting, string processing, array manipulation (to name a few) and from scripts to highly concurrent low-latency applications (see presentation in the ref [3] above by Prof. Shah from Yale).
It's a shame that the proper typed programming language are being ignored just because of programmers' locally sub-optimal preferences and limited exposure. The productivity increased using typical scripting languages including Python is diminishing everyday with the proliferation of IntelliSense, auto-complete and AI assisted coding.
For production codes, the scripting language based systems if they ever made it to production (mostly do e.g AirBNB, Twitter, Shopify, Github, etc) will be a maintenance headache and user nightmare, if the supports are not great and not unicorn start-ups. The last thing you want is that your saved eclaim form that you spent many hours preparing totally dissapeared since the system cannot recall the saved version. Granted this can be because of many reasons, but most of the problematic production systems are mostly written in scripting languages including Python because these are the only language the programmers know and familiar with. Adding to the insults are the readily available so called "battery included" libraries are convenients but ironically written in other compiled but unsafe system language in C/C++.
I think you're going to trouble convincing people a compile-loop language is going to be on-par with a REPL/interactive setup. You can look at some extreme example like MATLAB. With all your tools you're never going to reach the same level of interactive productivity with D for the subset of problems it's address.
You can have all your tools dump out and rewrite the oodles of boiler plate your typed languages require - but at the end of the day you have to read all that junk... or not? and just vibecode and #yolo it? But then you're back to "safety and reliability" problems and you haven't won anything
Also "safety and reliability" are just non-goals in a lot of contexts. My shitty plotting script doesn't care about "safety". It's not sitting on the network. It's reliable enough for the subset of inputs I provide it. I don't need to handle every conceivable corner case. I have other things to do
> Adding to the insults are the available readily available libraries are convenients but ironically written in other compiled but unsafe system language in C/C++
No on cares if you leak memory in some corner case with some esoteric inputs. And noone is worried your BLAS bindings are going to leak your secrets. These are just not objectives
My point is that Dlang scales from beginner to expert, from scripting to highly concurrent low-latency applications. Why settle for sub-optimal scripting languages if you can have the real deal with much better performance and freely available open source?
In the automative world if you can afford it, you need daily drive car for the job and supermarket runs, weekend supercar for fun/showing off, and off-road 4x4 vehicles for overnight camping. But in the software world D can cater for mostly everything with free open-source compilers, minimum productivity overhead and much cheaper to host as well [1].
Funny you mentioned BLAS, since Dlang BLAS implementation has also surpassed the run-of-the-mill high performance BLAS library that these scripting languages can only dream of (Matlab calling the 3rd party Fortran codes no less) [2].
brotha... please.. you're making me laugh so hard, my parentheses are shaking and getting unbalanced. Non-lispy langs that "provide REPL facilities" and Lisp dialects share the word "REPL" the way a rowboat and an aircraft carrier share the word "vessel". A Lisp REPL is an architectural relationship between you and a running system. Dlang's REPL is a nice sketchpad - worse than Python's. It's in the same "sketchpad league" with C#'s, Ruby's, Kotlin's and Node's. Clojure REPL is closer to what Smalltalk had - you're always inside the machine.
Imagine the NATO reaction if on the very first day of Russo-Ukrainian war offensive is by Russia performing missiles bombing murdering 100 kids studying in Ukraine primary school.
Trump candid reaction to the Iranian school incident when asked by reporter was "I can live with that".
There were significant civilian casualties right from the start of the war in Ukraine, and several massacred villages.
Russian air defense shot down a civilian airliner mostly full of Dutch nationals and the response was just condemnation and tweaking the sanctions a bit.
My heartfelt sympathy to the MH17 victims and families, but the airplane was flying in the risky warzone. That does not discount the fact that it's an atrocious act by the Russian backed military.
However, to send missile to primary school killing hundreds of school girls on the very first day of the war, if intentionally is just pure evil.
It wasn't intentional; the building was used by the military years before. The US had really badly out-of-date intelligence and was negligent in updating it. There's absolutely no military benefit to bombing a girl's school.
Remember Hanlon's Razor, and remember how incompetent the Trump Administration has been in everything ever since he took office.
But 10~15 years outdated intelligence about an area considered a significant adversary that is penetrated by oodles of humint sources... hard to believe.
Israel has hit schools before knowing full well that it's a school (in war against Egypt). May well be policy.
Many of these kids would have been kids of IRGC. Likely that was the reason.
> There's absolutely no military benefit to bombing a girl's school.
Objection. For a hypothetical actor wanting to set the world economy on fire there might be a benefit of enraging the enemy to lower the risk of early deescalation.
The attack on the Iranian primary school (not makeshift sheltered building) is on the very first day of the war, not several weeks, months or years.
Not to downgrade the incident, but the Mariupol incident you mentioned probably happened in 2022 while the all out war started in 2014 [1],[2]. If you can refer and link to the particular incident it'll be helpful for verification.
While the Ruissan invasion was ongoing from 2014, the 2022 full scale invasion is different both in scope and volume. It is viewed as its own global event and has it own huge Wikipedia article. Iran had also been attacked by the USA (and Israel) previously.
The theater was marked with huge inscription of "CHILDREN" on tarmac, in the pilots' native Russian. They killed them regardless.
Either way do you think that if it happened on day one instead of 3 weeks in the reaction would be any different?
Yes on day one people barely know what's happening, life goes on as usual.
After several weeks of bombardment and siege like was happened in Mariupol, children were already stop attending schools, moved to other schools, go to bunkers, live in makeshift shelters or migrate to different cities [1].
"The Geneva Conventions state that the parties to a conflict must do their best to protect civilians, which may include moving civilians and civilian objects under their control if they are close to military objectives." [1]
[1] Fact check: What do we know about the airstrike on a school in Iran?
Fun fact, the conquer of Constantinople was prophesized in the sahih hadith (authentic saying of the Prophet) [1].
Another fun fact, it took about 800 years, or four Islamic caliphates before the event happened in the Ottoman Caliphate era, after Rashidun, Umayyad and Abbasid Caliphates [2].
And another fun fact, one of the prophet trusted sahabat namely Abu Ayyub al-Ansari was involved in the very first campaign of the First Arab Siege of Constantinople during the Rashidun Caliphate [3]. He died during the event in his old age of over 90 and become the patron saint for Istanbul, Turkey. His tomb/mosque was the official inauguration place for almost all of the Ottoman Caliphs [4]. He also famously hosted Muhammad in his house for several months during the migration or Hijrah to Yathrib (modern Madinah), where Hijrah date is the start of Islamic calender.
[1] Muhammad Al Fatih – The Sultan who did the impossible:
>Am I missing something? Why is everyone talking about sandboxes when it comes to OpenClaw
>And now, what, having inference done by Nvidia directly makes it better? Does their hardware prevent an AI from deleting all my emails?
Because other people including Nvidia are mainly focusing on different aspect of data security namely data confidentiality while your main concern are data trustworthy.
Don't conflate between these two otherwise it's difficult to appreciate their respective proposed solutions for example NemoClaw.
For working example, please check TabulaROSA (Tabular Operating System Architecture) proposed by the MIT team. Instead of normal OS system call, it utilizes data based operations with D4M that can work mathematically via associative array with structured or non-structered data [1],[2].
With the advent of new CPU acceleration with fully homomorphic encryption as demonstrated by Intel, the AI model and agent can even analyze the data without even decrypting them [3],[4].
[1] TabulaROSA: Tabular Operating System Architecture for Massively Parallel Heterogeneous Compute Engines
https://dspace.mit.edu/handle/1721.1/126114
[2] D4M: Dynamic Distributed Dimensional Data Model:
https://d4m.mit.edu/
[3] Intel Demos Chip to Compute with Encrypted Data (121 comments):
https://news.ycombinator.com/item?id=47322815
[4] Intel Demos Chip to Compute With Encrypted Data: Fully homomorphic encryption chip speeds operations 5,000-fold:
https://spectrum.ieee.org/fhe-intel
reply