I wonder about the possibility that AI “clankers” and slop are being weaponised to attack the open internet to push human “data generators” into walled gardens where they can be properly farmed?
I mean, from an incentive and capability matrix, it seems probable if not inevitable.
Will this administration start costing serious people money enough that the other billionaires will get involved to kick out the nationalist idiots? That'd be nice. It's too bad we need that to be the case but.. would be a nice side effect.
The rich in America make their money through stock market pumps and dumps. Just keep the minimum wage frozen and the tax cuts coming and they won't say a word.
As long as the Federal Reserve and the Bureau of Labour Statistics keep saying the right things, the market will go up. And if they don't say these things, the president will replace them with somebody who will.
No, because an unstable regime and desperate economy increases the relative power of billionaires even if it decreases their absolute wealth somewhat. You don't see oligarchs abandoning russia en masse due to the constraints the system places on their growth.
So even if this ends up direly for some of them individually, they all personally are confident that won't be them. And as long as it isn't this aligns with their general interests.
Not only is it even faster, they now disguise themselves to look like professionally written PRs with advanced understanding of the tech, while being filled with junior level bugs. So you have to super scrutinise it.
Which makes me wonder what the point of even taking PRs is, the reviewer could just run the AI themselves and do the same review but not have to go through the process of leaving comments and waiting for the submitter to resolve them.
I'm imagining a funny possible outcome of this: Code linters/formatters get abandoned so personal style quirks can shine through, making code look visibly "not AI". If the quirks are consistent then it could also hint against it being faked.
I'm old enough to remember the global melt down in 2000ish and 2008. Oh and 1991 in the UK - lol, that's when I graduated. Take your money out of AI and stuff it under the mattress (gold if you are a magpie or blue stocks).
I have actually just spent out on a fairly handy GPU to stuff into one of our backup boxes at work. It has gobs of RAM and a fairly useful pair of CPUs and sits idle during the day.
AI via LLM is a thing but it isn't worth silly money and I think that a wind of change is on the way.
All these GPUs have to be used for something, if it's not AI then its bitcoin, if its not bitcon it's NFTs, if its not NFTs then its Porn.
This is a hardware driven bubble that will mutate into the next big computing hype. Server farms have to be doing something because they cost money. So whatever the next big hype is, big-tech will jump on the bandwagon.
They really don’t though. If what they produce becomes less valuable than the electricity it takes to operate the data center they’re housed in, the biggest concern will be figuring out the cheapest way for them to do absolutely nothing.
OSS maintainers may need some kind of response like the one I've written here that can be strategically dropped on the worst "bad AI" contributions. I certainly wrote this for myself to make my job easier, anyway.
Are they comparable tools in a team context? The ease of linking documents to people and collaborating? Seems like solo, sure. But in a group.. Obsidian doesn't replace Notion at all.
obsidian’s great for solo use, but for team collaboration, anytype is closer to notion-built-in group sync, permissions, and offline support. you don’t need plugins or external services for multi-user editing; it just works across devices
Here's one Claude-vibed project that makes me money that I run in addition to my saas, which is Elixir. I'm not strong in TypeScript and this is an Astro static site, so Claude has been really helpful. Backend is Supabase (postgres) and a few background jobs via https://pgflow.dev (pgmq) that fetch and populate job openings and uses some AI steps to filter then classify into the correct categories (among other things, there's also an job application flow and automated email newsletter): https://jobsinappraisal.com
This is clearly low quality, non-idiomatic AI-generated Elixir code. So the likely answer is that "you" did not use this at all; AI did.
I review this kind of AI-generated Elixir code on a daily basis. And it makes me want to go back to ~2022, when code in pull requests actually made sense.
Apologies for the rant, this is just a burnt out developer tired of reviewing this kind of code.
PS: companies should definitely highlight "No low-quality AI code" in job listings as a valid perk.
> It's conspiratorial thinking to assume that everything that happens in the world is perfectly executed by omniscient villains with 20/20 hindsight.
Because the original comment isn't doing this. It's not talking about everything, it's talking about one specific thing in a very plausible scenario.
It wouldn't even need to be a very complicated or widespread "conspiracy": Just Musk and a few VC guys in a Signal or Telegram thread saying
> someone should just buy Twitter and downrank all these crazy leftists
> Hmm
> I'll help line up financing.
> Ok!
This isn't flat earth, chem trails, lizard people, or weather weapons. It's not even Illuminati, Masons, or Skull and Bones. We've seen some of these chats already.
Can you provide more about this idea? I see the Boring company as being pretty feckless, and at the same time extremely boastful. They have gotten hopes up in a number of places about solving city traffic problems, only to go dark when the rubber (should have) met the road.
But I don't see any of those having impacted the California High Speed Rail. Rather that has been harmed by lots of different groups throwing roadblocks up, sometime for ideological reasons (lots of this from State and National Republicans, sometimes with reasons, but often more political), and a whole lot of NIMBY (see: Palo Alto). What do you see the Boring Company having to do with that?
As a side note: there are some really poorly thought through parts of the project, for example they don't have a plan for actually making it over the mountains into Los Angeles. I still want it to happen, but...
The CHSR thing is a bit apocryphal (no evidence, just according to his biographer) since hyperloop never really competed in any way with CHSR. He did, however, play a very big role in fucking up a potential Chicago connection between downtown and O'hare, as the Boring company actually did win the bid to use the abandoned cavern below the Washington Red/Blue line stop, promising to run a hyperloop up to the airport. It never went anywhere, and the cavern below block 37 remains abandoned.
> Last week, the Boring Company won a $48.6 million bid to design and build a “people mover” beneath the Las Vegas Convention Center. The payout represents the first actual contract for Tesla CEO Elon Musk’s tunneling venture. And Las Vegas, a tourist city that wants to be seen as a technology hub, will get a new mobility attraction with the imprimatur of America’s leading disruptor.
> “Las Vegas is known for disruption and for reinventing itself,” Tina Quigley, the chief executive officer of the Regional Transportation Commission of Southern Nevada, said when the partnership between the Boring Company and the Las Vegas Convention and Visitors Authority (LVCVA) was announced in March. “So it’s very appropriate that this new technology is introduced and being tested here.”
He's provided evidence of being an impulsive fool for even longer. I defended Musk as a useful idiot for a while until be fully showed his true colors, but it has always been clear he's not a wise man.
(His vigorous and pathetic efforts to get out of the purchase also push against it being a big master plan, FWIW.)
There were dozens of social networking companies at the time that FB was founded. If Zuck didn't exist those same or similar workers would have been building a similar product for whichever other company won the monopoly-ish social media market.
Since LLM’s aren’t deterministic isn’t it impossible? What would keep it from iterating back and forth between two failing states forever? Is this the halting problem?
Varying the deployment type (chip model, number of chips, batch size, ...) can also change the output due to rounding errors. See https://arxiv.org/abs/2506.09501 for some details on that.
The two parts of your statement don't go together. A list of potential output tokens and their probabilities are generated deterministically but the actual token returned is then chosen at random (weighted based on the "temperature" parameter and the probability value).
I assume they use software-based pseudo-random-number generators. Those can typically be given a seed-value which determines (deterministically) the sequence of random numbers that will be generated.
So if an LLM uses a seedable pseudo-random-number-generator for its random numbers, then it can be fully deterministic.
There are subtle sources of nondeterminism in concurrent floating point operations, especially on GPU. So even with a fixed seed, if an LLM encounters two tokens with very close likelihoods, it may pick one or the other across different runs. This has been observed even with temperature=0, which in principle does not involve _any_ randomness (see arXiv paper cited earlier in this thread).
I'd suggest the problem isn't that LLMs are nondeterministic. It's that English is.
With a coding language, once you know the rules, there's no two ways to understand the instructions. It does what it says. With English, good luck getting everyone and the LLM to agree on what every word means.
Going with LLM as a compiler, I expect by the time you get the English to be precise enough to be "compiled", the document will be many times larger than the resulting code, no longer be a reasonable requirements doc because it reads like code, but also inscrutable to engineers because it's so verbose.
Yes, in general, English is non-deterministic, e.g., reading a sentence with the absence or presence of an Oxford comma.
When I programmed for a living, I found coding quite tedious and preferred to start with a mix of English and mathematics, describing what I wanted to do, and then translate that text into code. When I discovered Literate Programming, it was significantly closer to my way of thinking. Literate programming was not without its shortcomings and lacked many aspects of programming languages we have come to rely on today.
Today, when I write small to medium-sized programs, it reads mostly like a specification, and it's not much bigger than the code itself. There are instances where I need to write a sentence or brief paragraph to prompt the LLM to generate correct code, but this doesn't significantly disrupt the flow of the document.
However, if this is going to be a practical approach, we will need a deterministic system that can use English and predicate calculus to generate reproducible software.
Interesting, I'm the opposite! I far prefer to start off with a bit of code to help explore gotchas I might not have thought about and to help solidify my thoughts and approach. It doesn't have to be complete, or even compile. Just enough to identify the tradeoffs of whatever I'm doing.
Once I have that, it's usually far easier to flesh out the details in the detailed design doc, or go back to the Product team and discuss conflicting or vague requirements, or opportunities for tweaks that could lead to more flexibility or whatever else. Then from there it's usually easier to get the rest of the team on the same page, as I feel I'll understand more concretely the tradeoffs that were made in the design and why.
(Not saying one approach is better than the other. I just find the difference interesting).
Sure, we cannot agree on the correct interpretation of the instructions. But, we also cannot define what is correct output.
First, the term “accuracy” is somewhat meaningless when it comes to LLMs. Anything that an LLM outputs is by definition “accurate” or “correct” from a technical point of view because it was produced by the model. The term accuracy then is not a technical or perhaps even factual term, but a sociological and cultural term, where what is right or wrong is determined by society, and even we sometimes have a hard time determining what is true or note (see: philosophy).
What? What does philosophy have to do with anything?
If you cannot agree on the correct interpretation, nor output, what stops an LLM from solving the wrong problem? what stops an LLM from "compiling" the incorrect source code? What even makes it possible for us to solve a problem? If I ask an LLM to add a column to a table and it drops the table it's a critical failure - not something to be reinterpreted as a "new truth".
Philosophical arguments are fine when it comes to loose concepts like human language (interpretive domains). On the other hand computer languages are precise and not open to interpretation (formal domains) - so philosophical arguments cannot be applied to them (only applied to the human interpretation of code).
It's like how mathematical "language" (again a formal domain) describes precise rulesets (axioms) and every "fact" (theorem) is derived from them. You cannot philosophise your way out of the axioms being the base units of expression, you cannot philosophise a theorem into falsehood (instead you must show through precise mathematical language why a theorem breaks the axioms). This is exactly why programming, like mathematics, is a domain where correctness is objective and not something that can be waved away with philosophical reinterpretation. (This is also why the philosophy department is kept far away from the mathematics department)
Ah yes, that makes a lot more sense - I understood your comment as something like "the LLMs are always correct, we just need to redefine how programming languages work"
I think I made it halfway to your _actual_ point and then just missed it entirely.
> If you cannot agree on the correct interpretation, nor output, what stops an LLM from solving the wrong problem?
Yep. I'm saying the problem is not just about interpreting and validating the output. You need to also interpret the question, since its in natural language rather than code, so its not just twice as hard but strictly impossible to reach a 100% accuracy with an LLM because you can't define what is correct in every case.
It seems to me that we already have enough people using the "truth is subjective" arguments to defend misinformation campaigns. Maybe we don't need to expand it into even more areas. Those philosophical discussions are interesting in a classroom setting, but far less interesting when talking about real-world impact on people and society. Or perhaps "less interesting" is unfair, but when LLMs straight up get facts wrong, that is not the time for philosophical pontification about the nature of accuracy. They are just wrong.
I'm not making excuses for LLMs. I'm saying that when you have a non-deterministic system for which you have to evaluate all the output for correctness due to its impredictability, it is a practically impossible task.
As much as many devs that haven't read the respective ISO standards, the compiler manual back to back, and then get surprised with UB based optimizations.
Repeatable builds are not a requirement for determinism. Since the outputs can be determined based on the exact system running the code, it is deterministic - even though the output can vary based on the system running the code.
This is to say every output can be understood by understanding the systems that produced it. There are no dice rolls required. I.e. if it builds wrongly every other Tuesday, the reason for that can be determined (there's a line of code describing this logic).
While I don't disagree with your comment, I would say that a that a large language model, and a Docker build with a complex Dockerfile, where not every version is exactly pinned down, are quite similar. You might have updates from the base image, you might have updates from one of the thousands of dependencies. And each day you rebuild the image, you will get a different checksum. Similar to how you get different answers from the LLM. And just like you can get wrong answers from the LLM, you can also get Docker builds that start to behave differently over time.
So this is how it often is in practice. Then there is the possibility to pin down every version, and also some large language models support temperature 0. This is more in the realm of determinism.
