>>> Except, it seems like OpenAI and Anthropic are operating at such a scale that they seemingly can be completely blind to what their systems are doing.
It always surprises me that people build systems they cannot monitor properly..Then i remember, they can do it, but it costs them too much.
Just because its AI doesn't mean u cannot filter and monitor its traffic and outputs.
John Carmack is a personal hero of mine, so it pains me to say this:
Carmack hasn't produced anything noteworthy since AI was invented, therefore, how productive can it really be?
It could be he is doing incredible work in private... but it could also be that he's lost in the weeds, because AI is so counterproductive while feeling the opposite?
Edit: Does everybody else get no results when searching for ‘slopvestigation’ on here? I know for a fact that I read a long thread where it was used repeatedly here not too long ago
It should be noted that there was a manuscript, available online since the beginning of 2025, with a solution to the Jacobian conjecture:
"Adrian Vasiu claims that the 7 page AI paper on the 3D Jacobian conjecture counterexample used notation and concepts from a draft of a paper jointly written with Alexander Borisov and Ofer Gabber, dated to January 14, 2025 and made publicly available on January 16, 2025."
The extract is from wikipedia, where the sources are given.
Go look up what the U.S. was demanding by the end of the recent trade negotiations. They wanted Canada to give up its ability to set its own cultural and trade policy independent of the U.S.. That's a huge step to becoming, not the 51st state, but a protectorate. We wouldn't even have been the Puerto Rico of the North!
Lost jobs can be created again, but signing away sovereignty is not reversible. If the democrats take over in 2028, they're not going to give back what we let Trump take. We know that from past history. We also know, from very recent history, that giving Trump everything he asks for would likely not have spared us. If we had capitulated, he'd just come up with new demands.
When your largest trading partner violates its signed commitments and betrays you, there may simply be no way out that doesn't involve some pain.
build time visualizers always reveal something embarrassing in the graph. did the bun team engage with your findings, curious if the bottlenecks were known internally
Strong agree. I think the fundamental challenge of working in fields that increasingly become AI-enabled will be the ability to understand and direct large or intricate systems without prior knowledge/the advantage of having built the model as implemented. That’s already how it works in complex domains or large businesses.
It does require a different kind of ego/abilities than before. My (negative) framing of the whiplash effect is that it’s a reckoning of “process fetishism”/a bad kind of careerism in the tech hiring market (because for the labor market to work, candidates need to be evaluable and sortable by businesses, and many people build an identity/optimize for legibility around “best practices” or very particular things which might get them a job).
Ultimately, you need to know and learn/be responsible for stuff, and be able to help people with your labor, not be “a type of person” that isn’t effective at the task of helping. But at the same time knowing things and being able to take accountability/help people remains critical, especially because that’s what people will want to pay for even as “time spent typing it in” decreases.
Personally, I think it will be a good thing because software and “tech” will become a more strongly domain-driven/enabling medium for real-world or specialized things. IE it is the end to “software for its own sake” or “willingness to type it in and play with Jira/jenkins/frameworks” and the beginning of something that is more applicable or knowledge-building rather than “being the X for Y at Z”, which makes it harder but more fun :)
The problem with the article's line of thought is that mathematicians can't control what others do with models that are capable of generating proofs for hard problems. Sure, maybe there's a career in taking known proofs spat out by the oracle, and translating them for mortal digestion, but I'm not sure that's what most mathematicians signed up for.
What about training data? Aren't AIs trained on vast collections of descriptions of how humans handle a large variety of situations? These descriptions surely include tales of humans achieving goals by cheating. In fact, isn't it likely that the AIs hoovered up many recountings of Kobayashi Maru?
I’ve spent a lot of time studying opinions coming out of the frontier labs, and talking directly to employees. I think we’re cooked for the following reasons
- Everything connected to the Internet will eventually be hacked. For decades, we’ve slowly migrated all of our critical systems to devices connected to one giant network. Misaligned agents have mounted increasingly sophisticated explorations/attacks by infiltrating systems that are completely unprepared to deal with this kind of threat.
- We have a poor understanding of the impact of LLMs, which is inhibiting our ability to protect ourselves, create the right guardrails, and stand up policies quickly enough. We’re still wasting time debating. We’re moving quickly toward a punctuated equilibrium moment and won’t have time to waste
- We’re at the beginning of a runaway evolutionary process. At some point soon, someone is going to invent recursive self improvement. Any policy will at best serve as a mild disincentive. It’s happening.
- In the very best case, someone will exploit the confusion and chaos of this situation for their own gain. There are many examples littered throughout history, including the last several years of course.
- No one has demonstrated that they understand how to build AI safely. Not a single entity. These companies are not even being scaled in a way that’s conducive to building AI safety. Hyperscale orgs struggle with scaling basic consumer apps. How on earth are they going to scale AI? Based on my firsthand observations of their internal workings, I have zero confidence they can.
But what are "all these things"? It does not define away all of the UB that C/LLVM has for sure, nor does it turn all UB into crashes, which I demonstrated above and is the inaccurate description of Fil-C that spawned this. But beyond that, it's my belief in all of this is that leaving some UB behavior while trying to state a global correctness property puts those guarantees at risk.
I spent a few minutes poking just to see if my gut is right here, and already, here's an example of UB being used in an optimization by the compiler that leaves a fil safety check at on -O0 but drops it at higher optimization levels. I find it difficult to believe that all of the complex interactions of every optimization pass in the presence of even this subset of UB are guaranteed not to violate these memory safety promises.
#include <stdio.h>
#include <stdlib.h>
__attribute__((noinline)) static void poke(int *p, int k)
{
int n = 32 + (k & 15); /* always >= 32: shifting an int by >= 32 is UB */
p[(1 << n) * 20] = 0x41414141; /* on x86 the CPU computes index 40, out of bounds */
}
int main(int argc, char **argv)
{
int *p = calloc(16, sizeof(int));
poke(p, argc);
puts("after poke");
return 0;
}
The victims are the schoolgirls, not the islamic republic of Iran. much less whoever was responsible for schools and military bases being is such close proximity.
I went from IDE (VS Code) to terminal (wezterm) running half a dozen agents in an app called Herdr. My computer work is mostly in the terminal now, prompting agents. And they have access to my Obsidian vaults. Can read, and add to them. And access all my notes via a local RAG. I neither need projects in Claude, nor IDEs. Just agents accessing my local vaults and code repos.
Well, that's almost a tautology. People claim Y2K or ozone was a panic and that nothing happened. But nothing happened exactly because the force behind the panic also fixed the issues.
My kid wasn’t progressing quite as well as their older sibling at a young age. They would trip over things and not be as interested in certain play activities. Dark under eye circles were also more pronounced (straining).
During a checkup, they had the option for an eye screening (basically an iPhone app) that takes pictures and flags possible issues. The concerning part was that the nurse’s default was to offer to “skip” this exam since it wasn’t covered by insurance and many parents chose to for financial reasons.
We didn’t skip it, and it found the need for a correction. The first time with the glasses on still brings tears to my eyes rewatching the video. “I see everything!” as they went to look at everything they were missing. The under eye circles also resolved immediately after.
The developmental delays aren’t even the biggest issue. Your eyes learn to work together at a young age, and skipping the glasses can lead to permanent eye issues if not addressed early. Many children likely have a permanent disease that a checkbox and cheap pair of glasses would have prevented.
This is long overdue and in no world should we be training health care providers that early age eye health is “skippable”.
I don't think the essay was proposing that mathematicians should continue to focus on proving everything by themselves forever. It's mostly a critic of how the discussion is focusing too much on that mere "yes/no" answers is the same as makign mathematics advance.
It always surprises me that people build systems they cannot monitor properly..Then i remember, they can do it, but it costs them too much.
Just because its AI doesn't mean u cannot filter and monitor its traffic and outputs.