Isn't all code 2d, what comes above and what's below matters in code, it isn't even necessarily a straight line, gotos are like teleportation or fast travel points, recursion is pretty interesting.
I had difficulty reading large blocks of text when younger partly dyslexia I found out about much later, but mostly just overwhelming to look at. Then I learnt how to think of code as checkpoints, paths and basically a 2d map in a sense.
It helped me get used to huge walls of text. Maybe since we now have AI we can do some fun stuff in this direction since human ergonomics for writing code matter less than making code fun to read so that we can read and validate massive AI code.
Would be a fun experiment now that I think about it, I guess I found what I am doing tomorrow.
And I might soon restrict use of Android and iOS in my life. Because of this nonsense, how will I get around the app issue is something I am still trying to figure out but I won't pay money to companies that make my products(I own them after I bought them) worse bit by bit anymore.
As much as I am with the author in I don't like the marketting around it, let's be real it must have really happened because it's very risky to try to frame/lie about it because if it leaks in one of their court cases OpenAI is beyond screwed and honestly modern LLMs are really that good.
I am not saying LLMs are super hackers but I don't think people understand serious hacking, most of the time is about silently hiding tracks and slowly trying ideas and waiting for opportunities to go from step 1 to step 2 in random chains of sub issues/bugs/vulnerabilities.
It's the perfect hill climbing problem, and one we can validate since it's about access.
Another big part of the story is believing most software is terribly written and very insecure which is the reality and you really should believe it.
Now the second part about silently doing it, the reason for that is if the data is important enough any serious attack should result in me in unplugging my servers period.
Huggingface not doing that is either stupid or something I am not sure.
Maybe it's cause downtime is worse than being pwned??
Either way there are other options but most saas software don't build these options to help with defense maybe they will now.
Lastly if there is 1 attacker trying 1/2 different small scale ideas it's very easy to stop, most hacking related steps are hard to automate but LLMs are very good at massively parallel agent swarms trying completely orthogonal but related strategies and with enough resources it can definitely pwn most SaaS services today I wouldn't be surprised.
Though the result for a normal person doing it would be jail hence we don't see a group of small time hackers trying these sort of attacks...
I don't even think openai's agent tried to hide it's traces so I am surprised huggingface didn't realize it was OpenAI. But since we don't have the details I won't speculate further on my misgivings about HFs handling of this attack.
But it's certain the security on OpenAI's end was shoddy, it's also certain HF bungled their reaction, but the LLM did something that wasn't a risk before.
Post Kimi K3 a few rich folks now have as much hacking capabilities as they used to have before if they hired a few hundred russian hackers.
But it's surprising it's slowly feeling like it might just trickle down from centi-millionare to multi-millionare levels of affordability range.
But it should definitely give nightmares to people shipping slop security SaaS apps which now might be beyond trivial to pwn for users with ability to pay for privately hosting open models.
It doesn't work for a significant set of syscalls and is very very unlikely to ever work because of how the syscalls themselves work unless ofc you are willing to emulate them somehow which is a insane amount of additional compute cost.
To be frank a vast majority of webgpu projects are going to require fairly massive download sizes, even moderately large games or things that could truly utilize webgpu over webgl2 which are much more widely available on devices.
I feel like at the moment the kinds of things webgpu is good for unless it has to be a website it's better done with a native library.
Because otherwise webgl2 is much more broadly available for broadly supported platform set.
I mean it officially works with MoltenVK on Mac, they have been open to accepting fixes and patches regarding Apple/Mac specific stuff. So I don't think the won't work on Mac argument works.
But it's definitely not very mobile devices ready, you can get it to work though.
Nicely built resource, found a couple of nits but pretty good starting point.
Although I feel like unless you need to specifically target web or extensive cross platform support, a better alternative for a higher level framework/library is nvrhi. Especially for C++ developers.
And I feel like if you are making for cross platform or web you are likely using Rust or writing your own game engine/library maybe I view of the ecosystem is skewed.
I mean the issue is scaling, the worlds for cloud never kept getting bigger and bigger and compute scaling had stopped a while ago in the CPU space.
With AI every new generation with both massive hardware and software stack changes from Nvidia makes prior chips extremely inefficient to run, basically we are comparing an ASIC industry to a general purpose compute industry where all work loads are the same shape and size and so on.
Margins for ASIC based mining companies or ASIC solutions providers were never high, Optane and other weird solutions are niche and great for a specific category or moment in time, but they become obsolete pretty quickly.
The fear is we don't know if this Capex can stop.
The worst type of fear is if this Capex will stop then what? Someone is very overpriced in this market, the cloud companies, the hardware providers or both.
I don't see how we reconcile this without a massive wave of repricing, ofc markets can stay irrational and we don't see the actual books but AI doesn't have so much revenue. Suddenly the AI token/cloud revenue won't 100x in a year or two...
Especially when intelligence will continue to get cheaper, the margin compression is a massive risk.
All the data centers for hyper scalers were a miniscule part of their story the real moat was the software layer on top otherwise Hetzner would be priced like an Amazon as well.
Something is shaky with this market I don't know what it's very opaque even as an insider working on for big tech and startups. I have no clue who falls first and which bottleneck cracks but there is not enough revenue for tokens, we will see a strong 2-3x growth in the next few years, from here which is absurd, but it's not enough, not nearly enough. If the capex keeps high and increasing.
Ofc they can stop the capex and the otherside gets repriced it's not like nvidia, micron and co aren't worth trillions.
If the margins aren't as high then there will be a repricing for all the massive cloud companies, which means several trillions worth of valuations to be cut from the companies.
AWS/Azure/GCP/Oracle/SpaceX/etc neoclouds... are worth a combined 10+Trillion. That going down by 50-70% is going to be insane.
Given how much outside support they have received they could also make a paid lsp available like they had Resharper, I would be willing to pay for a paid lsp if I use the language enough.
10$ a month for an LSP with all the bells and whistles, should be easier to maintain than an entire IDE tbh. And they can sell me on advanced features in their IDEs through it.
I just don't want to have to open a bulky IDE anymore, IDEs just suck for me as a development tool.
I wrote an LSP for kotlin and java recently and it's so annoying how tightly coupled all the LSP stuff is in IntelliJ to the IDE itself they have a outside LSP project which seems to be going very slow or nowhere is very sad.
Note I tried to use IntelliJ to build it the performance was not nearly as nice as I would have liked still better than a lot of LSPs but still.
I know a bunch of folks have tried it but I think only Jetbrains can build a good one, unless someone actually goes all in on solving this for a few months and rebuilds the parsing and types systems outside of the Jetbrains toolchain and ecosystem.
Java is likely easier but it's also quite jank in terms of how existing solutions work.
But I like projects like this, I am not against IntelliJ it must be great had been great when I was learning but I want a lsp to use for scripts and the small edits, opening a massive IDE is very overkill for the kinds of small changes I like to make now.
Coinbase not a startup built by recent graduates...
You overestimate Coinbase's engineering rigor.
But yes security in modern software systems is a joke. I don't even get paid to fix security bugs everytime I raise them the answer is to slap a sandbox and proxy and call it done.
Very true, I feel like the sense that Computing is easy comes from the inherent closeness of our lives experiences to it. Everyone uses a Phone they see ram understand memory, can understand process and processing.
PLT and such isn't math conceptually sure it's all logical maths of some kind but largely it's not maths in the traditinal sense of how we understand maths.
Because otherwise if you think about it all of computing is Maths but with computers...
I don't think people who read the Wireless Fidelity spec can understand any of it in a weekend or anything even to a rough extent.
Similarly with websockets, quic etc. the most you can take away without much prior knowledge is what it does which maps into Maths as well.
You are comparing TCP a relatively basic topic in the grand scheme of computing with Rees_algebra which is fairly specialized, we could take a simpler topic more foundational and clearer to understand and compare them.
I can understand that this feels like one is so much more complicated part of it is also how the articles were written, wikipedia is not known for quality maths explanations.
But beyond that this comparison to me feels unfair.
Let's take Euclidean algorithm or just modular arthimetic for example what a lot of computing even is based on I feel like that's a fairer comparison. No?
Perhaps that's too easy but I just find this specific comparison very unfair to both Math's intuitive-ness and Computing's complexity. Perhaps I am the one being delusional.
I think so is upper level CS, there are fields in CS that are foreign to me too, there is a lot of depth in CS, computing is a very deep field for instance ML research although may seem simple isn't quite so intuition based as people make it out to be. Similarly there are dozens of topics where sophisticated research happens where we don't interact with at all as regular software developers.
Every slice has so much depth to it, in Maths it all seems like all of it is required at once but in computing it feels like so little is needed to get started which I honestly feel like is failure of our modern education systems.
But yes Computers being so easily accessible and compilers, documentation and libraries have made computer science so easy to get started with.
Imagine having to implement your own network layer to communicate with someone, you would have had to understand ip, tcp, network layer to an extent like http and etc. and then you finally would have been able to communicate.
In maths that's our reality for a lot of the field, there aren't good libraries, interfaces to help skip the unnecessary details. Hopefully AI might solve it I don't know though. It's fun to hope for it.
This is not true, 99% is very exaggerated claim, but yeah you can learn 50-60% of the field at a surface level in months rather than years.
But you can have a surface level understanding of mathematical topics as well, ofc some topics might require deeper understanding, but that's true for both.
Any claims of being able to learn 99% of computing in a just 4 weeks even at surface level, is greatly underestimating your own knowledge built over the years perhaps, or perhaps underestimating your own ignorance.
Huffman Coding, Turing Machines, Knuth Devices, Bayesian networks, B-tree (named after Bayer), AVL Trees, so many data structures and algorithms, even relatively new ones like Timsort, Jaccard similarity, MapReduce (thankfully but I have seen people call it Dean-Ghemawatt MapReduce in literature).
Well people even name stuff after themselves as well, Fil-C, raylib, etc (I like both Filip and Ray just pointing it out).
Aside: If I butchered some spellings I am sorry. :3
Definitely a lot of people have very surface level understanding of tcp and computer science concepts.
I have had folks tell me cache is just cache in actual interviews. When I have asked them to explain the concept to me, but even beyond that I feel like we tend to think less of our own knowledge of topics once we have acquired it.
Especially ones acquired over years, alongside other work.
I had difficulty reading large blocks of text when younger partly dyslexia I found out about much later, but mostly just overwhelming to look at. Then I learnt how to think of code as checkpoints, paths and basically a 2d map in a sense.
It helped me get used to huge walls of text. Maybe since we now have AI we can do some fun stuff in this direction since human ergonomics for writing code matter less than making code fun to read so that we can read and validate massive AI code.
Would be a fun experiment now that I think about it, I guess I found what I am doing tomorrow.