Coding related have had massive step-up improvements over time.
The same cannot be said for image generation. If anything, I swear it's gone backwards.
It peaked with very early midjourney (like 2022), maybe follow ups with stable diffusion, and then with some of the nano banana models for actual editing. But overall, very lackluster in comparison to other areas.
It is entirely possible to prompt engineer away from that style, but that requires effort and knowledge that said average person does not have, so it's stuck in a feedback loop.
Like, I'm not even an artist, so I'm sure there's a million things I'm missing, but it's just so upsetting to look at. There's weird random reflections on things that shouldn't have reflections, the coloured light absolutely doesn't actually track through the stained glass onto any surface that makes any sense, light is coming from every which way, it's just... gross to look at for anything more than a quarter of a second.
And this is what they're using to hype people up for it. /shrug
Our neighbor's mom called the animal warden on us for not feeding him. He eats better than I do.
Her dog is a morbidly obese miniature collie.
Most people's perspective of ideal animal shapes is incorrect.
Doesn't help that pet dogs spend much of their time laying down, with perhaps a couple brief respites in the form of a gentle walk.
They gain weight.
A healthy dog can often look malnourished to pet owners.
Either there’s a huge lack of common sense at many places here or this story is leaving some details out.
Godel’s theorem says that unprovable true statements exist. Fortunately there is a vast set of provable true statements that are quite useful in real life.
Similarly, the halting problem is that you can’t write a program that always determines whether absolutely any program you give it will halt. But in real life compilers can still detect infinite loops in your code unless you try really hard to fool them.
In practical terms a software system can never be “perfect”, because first you have to define “perfect” (i.e., specify correct behavior in every possible state), which is practically impossible, and even if you manage that, the hardware has to implement the software abstraction perfectly, and that never happens. (See prefetch attacks, cache timing attacks, supply glitch attacks, rowhammer, etc.)
The security issues come about when someone finds where you failed to specify desired behavior, or (at any level) your system failed to implement it in some circumstance. No strange loop required.
I remember reading Godel Escher Bach back in the day, and there is the parable of the record player between the tortoise and the rabbit. And the tortoise keeps building more and more secure record players that the rabbit keeps finding ways to break.
GEB then goes on to teach you that, no, in fact, you cannot have a perfectly secure system. Not just practically or economically, but even in theory, all systems are fundamentally breakable. GEB, after a lot of chapters (what a doorstop!) ties that parable into Godel's Incompleteness Theorem, and essentially proves that no, you can't make unbreakable things, as a fundamental point of logic.
So, when I look at all this, I see that yes, there is an economics point of view here, that fixing code is just expensive, like making a less and less breakable record player. And that AIs push that breaking and economics point much further along than what we have had.
But then I also see a lot of comments here that seem (to me) to be arguing that there is an unbreakable code out there.
Am I reading this right? Is there such a thing as 'perfect' code? Or is my outdated reading of GEB still valid or not?
[0]Note, I'm not a CS person, more a biophysics person.
Yes, we're going to see the bugs exposed by AI and it's going to suck a bit for a few months.
But when this is all over, software will be more secure than it has ever been before.
That's why it doesn't suffice for only LLMs to know the relevant stuff.
You can't have a zoo run by animals.
Namely:
> The only way to do that is to make your code provable.
Proofs are a mathematical concept. They only work in the abstract world of mathematics where everything is perfectly logical.
Engineering exists in the real world. The physical realm where our mathematics are only imperfect models of what actually exists. Removing bugs through physical engineering requires persistent trial and error until you have found all the places where our mathematical models don't adequately represent the real world.
It's the same in software engineering, which involves cobbling together a myriad of systems not all designed to talk to each other and massive, ever changing codebases where perfect knowledge is impossible.
So yes, we should be trying to engineer bug free code. But that's really more a matter of changing our resource allocation, spending more time and money on QA and doing away with mottos like "Move fast and break things". It's a cultural shift and will only come when the bosses -- the venture capitalists who ultimately control our industry -- are held accountable when things break, and given strong incentives to ensure they don't break.
Oftentimes there are two broad ways to fix a gap in the logic. You can eliminate the source of it, which often involves refactoring but tends to result in less code and cleaner, more comprehensible logic. Or you can leave the gap there, but attempt to render it safe by adding extra logic to detect and handle the error condition. This reduces changes to existing code, but makes the overall system more complicated.
You can’t just indiscriminately favor one approach. Which one of those is best depends on the situation. The deciding factor is often something like a time constraint that you can’t know just by reading the code. You might need to talk to your team before deciding.
But LLMs don’t do that. They always choose option B: slap a bandage on it. Keep on adding more epicycles. In the 2000s we learned that, over time, always choosing this approach inexorably leads to codebases that are brittle, bloated and expensive to maintain.
So the problem, from my perspective, isn’t that we shouldn’t be coding defensively. It’s that LLMs encourage us to do it in a way that we’ve already learned is unsustainable.
> -- Donald Knuth (if I recall that exact quote correctly)
Just because you have proven the correctness of your code, that does not mean that it is infallible. Especially -- as a lot of HN users point out -- there's intricate complexity that is likely to cause erroneous behavior if several interfaces/complexities get combined. And those might be caused, for example by a 'user' setting the driver to 'db.backend.sqlite' instead of 'db.backend.sqlite3'; and then you have suddenly an unexpected configuration.
The only way to do that is to make your code provable. That again is only realistically doable using AI.
In the process, people will have to turn the pretense of "software engineering" into reality, making the engineering part explicit.
That only works when humans actually understand what they're doing.
But explaining that openly, as is your duty, means surfacing something which could be read as 'bad work' by a non-technical manager unable to grasp the complexity. So in low-trust teams it very quickly becomes a game of brinkmanship, or a prisoner's dilemma for technical diligence
But shipping faster was pushed at almost every company until, if it's not revenue generating, stop putting resources into it. Not all are like this but it's not all or nothing, it's always a sliding scale how much you can put into making the software correct.
I think you have things backwards. Devs aren't clowns who make it this far. But in industries where buggy code doesn't kill people, devs are expected to ship fast and iterate, because they can and more importantly, it serves the business.
Personally, I'm all about good software engineering principles, unfortunately I don't think you can last long in most companies if you're too opinionated about this and can't compromise (and it's getting even worse with AI, reading the code is no longer a requirement, go explain your teammate that there's an abstraction leak in their code...).
Another industry does even better: the medical/health industry. It's both a clown industry AND kills people AND they have made it way beyond what software can only dream.
Trying to understand underlying reasons and incentive structures is much more productive and useful than simply venting about the worse moral character and seriousness of people in one industry vs the next. Curiously, the grass seems always greener - when you ask people in other industries, you'll often hear endless complaints about the state of things there too.
It totally feels like the former if you watch it find and fix vulnerabilities or other debt-tasks one after another.
But only so many of these "technical debt" issue are little off-by-one errors that get remedied by a surgically narrow patch.
Many enough require some kind of restructuring or redevelopment of their own and it's going to take time to figure out if AI faces the same dilemma as human developers: where each non-trivial change carries the risk of some new issue, ad infinitum.
If that happens to be the case, and AI isn't able to magically produce perfect software in finite time, then what's really happened is that the threat of hostile AI exploiting our technical debt will force us to always be burning and churning code just to remain a moving target.
It's another "we'll see" thing for now, I think.
First, I am not a PHP fan, but I will absolutely stick up for the "its good now".
Yes, absolutely no debate the old PHP 4.x/5.x was shocking. However to give PHP their due, they absolutely did make serious efforts in 7.x/8.x. They deserve due recognition for that.
Unfortunately lots of people – I'm guessing you're one of them – still have battle-scars from 4.x/5.x and they think (incorrectly) that 7.x/8.x is "more of the same".
Second "joke language", well, cast your mind back to what else was available at the time... ColdFusion, CGI, ASPX ... it was not exactly a beautiful world back then.
Sure, to borrow your terminology, the Java fans will be along soon to tell us all how awesome Java is for the web. But many of us still bear battle-scars of the horrors of Java stacktraces, messing around with incomprehensible XML config files, Java's penchant for eating RAM for breakfast, and generally trying to figure out why the hell Tomcat is not working (again !).
Everybody in the industry knows that we've been building on quicksand. Moore's law meant that we could just sacrifice performance and add a dozen layer on top of technology from the 90s (if not older) and keep the thing chugging along.
Look at the PHP situation:
- HTTP is an old and quirky protocol with a lot of surface open for attackers that are rarely useful nowadays and might be better left off to higher layers. But of course, we can't break compat, so everybody still implements all the naughty parts, just in case.
- PHP was a joke language from the start, we were making fun of it in the early 2000s, the fact that over 20 years later it still powers a huge chunk of the web is enough to demonstrate that we're not serious people. Cue the PHP fans telling me that "it's good now". Maybe, I don't really care, the effort that went into making it good should have gone into ditching PHP for a better language 20 years ago. The bandaid mentality is what got us in this situation.
- And of course to add insult to injury, PHP, this security-critical language, is still implemented in garbo C using things like setjmp/longjmp. Just to make sure to absolutely maximize the damage if a vuln is discovered, I suppose. Brilliant.
- Oh and all of this is still running overwhelmingly on top of x86 which is its own bag of legacy garbage. "No but you don't get it, modern CPUs are really more like RISC CPUs, the RISC ISA is translated internally" yeah exactly that's my point let's stop doing things like that.
We're a clown industry. We only made it this far because, usually, buggy code doesn't kill people and we can iterate and patch things quickly. We never had to do things correctly, we could flourish with abysmal levels of quality acceptance. This may not be true anymore.
Of course the reason things are done this way is generally to maintain backward compatibility and make transition easier. At least now that we have LLMs this excuse will hopefully vanish: porting even huge codebases to a new language/framework, something that was a tremendous undertaking mere months ago, is now vastly easier and faster. We can leave the 1990s behind us.
It’s frustrating to have all these 0days and CVEs to patch for now, but in the long run perhaps we’ll leave with more robust software.
I'm no expert but even I've read that "Works at the White House" is bigger than the executive residence and oval office, including the West Wing and the big office building that's directly connected to it by basement walkways. The National Security Council maintains 24-hour watch operations in the Situation Room (which is more a 'complex' than 'room'), while regional desk officers routinely work late evenings to overlap with morning business hours in Europe and Asia.
I asked Gemini if my recollection was correct and got: "The EOP network (managed by the EOP Office of Administration in coordination with Presidential Information Technology Community entities like the White House Communications Agency) operates as a single campus enterprise network. Its high-speed backbone links the West Wing, the EEOB, the East Wing, and nearby administrative facilities such as the New Executive Office Building (NEOB) on 17th Street."
So, definitely an issue that there's a device with some rogue malware installed but the total workforce across all shifts at the White House Complex is roughly between 2,500 and 4,000.
https://www.npr.org/2025/04/15/nx-s1-5355896/doge-nlrb-elon-...
> Most of the activity happens in the evening, Eastern time: about half of it falls between 5 p.m. and midnight, and almost none happens in the early morning. That pattern suggests a person rather than a public guest Wi-Fi network: someone regularly on the White House network late into the evening, whose device is infected with a proxy SDK. Our data shows the exits, not the device or the person behind them.
There really aren’t very many candidates for who that person might be and it really seems like it would be someone who, I don’t know, maybe lives where the traffic is originating.
https://darknetdiaries.com/episode/172/
There's a transcript if you want to read, not listen. At the start, you'll be "hey! I gotta get me one!" and later you'll be like "burn it with fire!"
They don't need to be physically inside a specific building to be a networking security problem.
The core issue is whether an outsider has bypassed your firewall and can send+receive traffic from a node which is getting more trust than it deserves. If they've compromised the computer's microphone/camera/wifi-control that's an additional problem.
https://techcrunch.com/2025/03/11/doge-axes-cisa-red-team-st...
Perhaps some of them annoyed the President's illegally-appointed mega-donor, if they spoke out against giving root access to all his unvetted junior flunkies.
If there's anything there, monitor your network to figure out exactly what
A compromised device inside the White House network could presumably be doing something _much_ worse than participating in a DDOS against random websites.
There is no evidence that physically ties this to the WH. It could be in a security lab for all they know. Or a guest network for the news media.
On a server that used to get below 1k HTTP requests a day, I've had several cases of mass scraping from these botnets exceeding 100req/s. The web is getting harder and harder to use due to all the anti-bot mechanisms site operators had to put up.
How is DDOSing every site on the web benign?
Help me understand. As far as I'm aware, these residential proxies are almost always malware.
Aren't they almost always run unknowingly, by people tricked into running them?
Person buys a product, or downloads some software - that unknowing to them, was reaching out to a central server elsewhere to establish a network tunnel. And over that tunnel, the other server routes traffic onto that person's home network. Which can then of course go onto the internet, and it appears like the traffic originated from their home internet connection.
Other people rent access to this network. The goal being, you have surreptitious access to appear as any one of tens of thousands of home network IP addresses on the public internet. Bad guys can use this to rotate through many different IP addresses when their access gets blocked because of malicious (or benign but unwanted) behavior on the internet. Residential IP addresses are more 'valuable' because they don't have the negative connotation that datacentre and commercial IPs can have, and they can be spread geographically or across many ISPs.
A very common way for people being a part of these residential proxy networks is by buying dodgy streaming boxes, and also installing shady apps on TV app marketplaces.
This also puts some of the troubling reports that politicians were bringing phones into SCIFs into context…
We do education outreach to K-16 learning institutions around the world. There are many things that can be studied in the near absence of gravity. Fluids are a huge area of research (surface tension, sedimentation, etc). We germinate seeds and look for genetic changes, people study brain organoids, artificial retinas, crystals, et al
Historically, there was an issue with dosing pembrolizumab, as the manufacturing process was creating large uneven crystals that required physicians to process at the clinic to ensure proper dosing, but in microgravity? The crystallization is just gorgeous. absolutely beautiful
Take a look at this image, https://issnationallab.org/wp-content/uploads/2024/05/merck_...
Left hand side are the crystals grown on the ground. Right hand side is 0g.
That change made it possible to turn it into a safe injectable that patients could take at home. And they were able to recreate those conditions on Earth, https://issnationallab.org/iss360/merck-lab-publishes-pembro...
https://pmc.ncbi.nlm.nih.gov/articles/PMC6889310/
This is more of a scientist said at a conference to another scientist kind of story, but apparently, they got some seed crystals from the ISS that are now used on Earth.
https://techconnectworld.com/World2025/a.php?i=602
You can do a lot more than make just seed crystals. It turns out that cells in our body are inside of a 3D environment with forces / tension acting in all directions. When you smoosh these in a lab (very literally) and try to observe how a drug behaves in a model cell / model organ, then it doesn't have the same physical constraints acting on it. BUT in 0g that's different. In 0g, cells behave slightly more similarly to how they behave in vivo and that could be a huge difference maker for drug discovery.
One breakthrough drug alone is worth more to humanity than what the ISS has cost across its lifespan.
https://iafastro.directory/iac/paper/id/89939/abstract-pdf/I...
So you tell them where to deliver it (anywhere on the planet that its orbit goes over..), they tell you when it will be arriving, and it reenters and lands there.
Article about a recent test (launched on a Falcon 9):
https://www.afrl.af.mil/News/Article-Display/Article/4629122...
From one of the issue trackers:
"While it's important to not break existing web content, it's also important to prevent security vulnerabilities. XSLT has been in maintenance mode in browsers and has been an ongoing source of security issues."
It is quite possible that flies prefer one part of the light spectrum more than another part, but to say in the title that they don’t swarm at LEDs is misleading at best.
> We found that LEDs attracted significantly fewer insects than other light sources, but found no significant difference in attraction between the “cool” and “warm white” LEDs.
I’m guessing he’s confusing the dramatic drop in insect life overall with a change in behavior due to different lights.
Also, not sure about white bulbs, but those color-configurable spotlights definitely emit UV at least when set to anything near blue/purple. They cause certain things to glow. Probably white is doing it too but it's harder to see.
The thing that insects can see and we don't is the rapid flashing of a PWM power controller.
I wonder if these streetlights have power saving fade-ins as the nightfall approaches instead of clicking on at sunset.
Clearly it isn't, otherwise every language would have sorted it out by... a long time ago.
I'm not entirely surprised, and I'm glad the Go team is carefully considering their interfaces before launching. But 6 years seems like a lot of work to land a JSON API.
I am using local LLMs and Go has a bunch of strengths where smaller contexts are concerned; less complicated flow of control being one of them.
So yea hopefully a nice new addition but only my prompts now matter. So need to make sure they pickup on the new API.
Anyone who fell for it would not only get in trouble for poor security practice, they would reveal that they were job searching. That opens an entire can of worms that creates problems beyond the pentest. This also angers other employees who see it as a double-trap being sprung on employees.
Another major problem is that responsible employees who accept the fake interview but then use their personal laptop to do the interview will have their personal laptop compromised by the pentesters. Responsible testers will immediately abort, but some will try to use the personal laptop to pivot deeper which is a huge ick. It’s also really bad if the company is compromising personal laptops of people who are doing the right thing (switching to personal gear for interviews).
One of the first things I did in my last job was shift processes like this off my team's computers, aside from the security ramifications all it takes is one dumb mistake, one up-arrow too many, one didn't-read-the-whole-line, and now you're doing something to production by accident!
That's a huge red-flag. In a company of this size, the deployments should be via CI only. CI isn't just to increase the speed and reliability of deployments but also to provide a clean room to build the software! Even if it's a token that was copied and pasted temporarily, the systems in place should not have to require this type of 1. copy 2. paste 3. remember-to-delete the token from my PC.
Also, Binance should be paying for intrusion testing. Better to pay a trusted company to do this kind of thing instead of letting North Korea show you how it's done. I don't think this exact exploit path would have been found but I bet the intrusion testing would have figured out another way to use Engineer C's token.
I think Binance needs to start having IT Security like a bank as well.
No AI, no leet code, just old fashioned gauging someone's experience and communication skills.
Even worse when you have to do stuff like screenshare or coding. I've had candidates share their screen and have obvious proprietary material in open windows. I've seen countless email inboxes and calendars. I've seen candidates do coding exercises using internal IDEs, dev environments and dev tools. I've had them talk – fully unprompted – about internal tech and unreleased projects. And I'm talking about Google, Meta, Amazon, Coinbase, Uber.
At this point I'm convinced that a large chunk of tech interviews are happening just to extract insider info from employees.
I even create a 1 man company for my consulting services and I receive ads about how people at my company saved me a spot on an online game.
LinkedIn is partially responsible for their neglect here.
And now you get results like this.
And another in Dresden: https://www.youtube.com/watch?v=cQUpQZAeA3w https://www.diesachsen.de/en/politics/a-zebra-crossing-in-dr... "In the summer of 2025, unknown individuals painted a crosswalk on the roadway along the busy street in Neustadt to provide safe access to the Heinrich Schütz Conservatory. The city promptly blocked off the crossing with warning cones [...] stating that unauthorized crosswalks constitute not only property damage but also a dangerous interference with traffic. Furthermore, traffic signs installed without authorization constitute an administrative offense under the Road Traffic Code."
In Los Angeles https://www.latimes.com/california/story/2026-10-16/risking-... "Risking arrest, he paints illicit crosswalks across L.A. ‘We can’t wait any longer’" (initially repainting faded crosswalks, rather than creating new ones)
And an example in Saskatoon, visible only in the 2015 street view image https://maps.app.goo.gl/3zWAkjtSEATswcWm8
They made up the term jaywalking by combining "walking" with the term "jay" which was a slang term for a country bumpkin or idiot from out of town. They hired actors to spread the slur to refer to those out of towners that don't know how to "properly" behave around streets in a modern city. They even paid Boy Scouts to hand out warning cards and staged mock funerals
As to the time investment:
> He spent more than three months preparing, researching the color and shape of the signs, mixing his own custom paint to get the colors just right, and applying a bit of faux patina to finish it off. Ankrom even spent time tracking down the button-shaped reflectors used on other legitimate signs and convincing the supplier to give him some. Once he was satisfied that it looked authentic, he came up with a plan to install it.
---
Caltrans has 15,000 miles of state highways that they have signage for. A single sign may only get reviewed once a decade because of that.
Doing single things in isolation is "easy" without going through the required public comments and such. Maintaining it for the entire state is much more challenging.
Vehicular cycling techniques can help in situations where there is no bicycle infrastructure, but they are not safer than good bicycle infrastructure.
> And then…nothing. Caltrans didn’t notice. Passing motorists were thrilled but figured the improvement had taken place at the direction of Caltrans. It wasn’t until Ankrom leaked his actions to the press that the true story was known.
> The original sign with Ankrom’s artwork remained until 2009 when the whole thing was replaced by Caltrans. The paint had faded, but the message remained with a lasting impact. The replacement sign erected by the Department of Transportation proudly sports the 5 North placard exactly where Ankrom had put it nearly a decade prior, and it remains there to this day to help motorists that would otherwise be just as confused as he was.
And an update from 2023:
> In 2001, artist Richard Ankrom posted a "5 North" sign on the 110 N. in DTLA to alert drivers to get in the left lane for the 5 turnoff. His work was so impeccable (and helpful), Caltrans at first didn't notice, then allowed it to stay. "5 North" is now part of permanent signage.
The Swiss police under prosecutor Walther Cimino proceeded to investigate as if they were a crime syndicate, gathering forensics but also identifying 96 sympathizers and putting them under heavy surveillance, which included requesting total records from their banks, employers, phone companies, email providers (Google through the US DoJ), and even CCTV and purchases from their local supermarket.
I'm afraid of the possibility that what used to be "grassroots civil disobedience" will now be labeled "eco-terrorism" and prosecuted accordingly.
Drivers and pedestrians hate each other too. Everyone gripes about any minor impediment to their own travel, particularly other modalities, but also their own. I mean, drivers universally complain about being "stuck in traffic", but they are traffic.
At some point students started painting their own crosswalk, which would be removed again and again by authorities, until the city eventually got the hint and installed a permanent crosswalk with a traffic light and a pedestrian island.
Sometimes the way to get something is fixed is by demonstrating the fix.
https://thelandmag.com/richard-ankrom-guerrilla-public-servi...
> Introducing Ripple-65B Preview, a Frontier Open-Weight Model
It's not "frontier", it's not "frontier" even in the category of open-weight models (though usually that phrasing would mean frontier over all, and also open weight)... it's only "frontier" in the small sub-category of US made open-weight models.
But as is, it's probably hard to become popular amongst hobbyists and local model enthusiasts like Qwen's 27B class models or MOE models because it's a 65B dense model and inference will be prohibitively slow on consumer grade hardware, and seems like the goal/mission of the company behind this model is to automate chip design.
So if I have to assume and connect the dots, this is for the companies who want to use AI in air-gapped environments for chip design so that they protect their IPs (not leaking to third party AI labs/hostile parties) and can afford lots of server grade inference hardware for a quick unlock with the big dense model.
Neat!
You mean it's state of the art for this subset of benchmarks you're targeting? No? Well surely it's state of the art for models of a certain size category? Huh, what about models that are available for local inference? O, state of the art for open models of a certain size made in the US specifically. Uh. Neat!
GAG
This made me laugh. The world is different now. Congrats!
The whole team is around to help answer any questions. You can reach out to me personally at charlie [at] melty.sh.
I'll prolly have claude reverse engineer conductor over the next month so the show can go on ...
And today we're doing a refund + moving the org to t3.codes instead.
Boo Meta. Boooooooo.
Wednesday, Nov 4 New users can no longer sign up for Conductor or download the app.
Friday, Nov 13 Conductor Cloud goes read-only
Local Conductor will continue working as normal for now. It will start winding down in early 2027.
Flashback to Cloudflare acquiring Deno and hiding the shutdown at the bottom of their announcement -- https://news.ycombinator.com/item?id=50019911.And you'll see that mitchellh is a big fan of Pierre (and I think investor)
https://x.com/mitchellh/status/2036855931569651884?lang=en
For those who don't know Pierce, here's a demo from mitchellh:
You'll have to click the link through to https://code.storage to understand: They host Git servers at scale, especially for agentic companies like Lovable (who create many projects and need many code repos to store them).
I had never heard of them before and have no clue what they do or why OpenAI wanted to buy them.
What a laughably insulated statement. An example from NZ - Janek Patel was murdered in 2022 while attempting to prevent the cash register at his corner store from being stolen.
The figure has been described as "a few dollars".
I asked "what the hell?" and he said "this is for me to use in case somebody talks shit about your father".
Upon some investigation, it turns out that the nicest guy in the world was actually living in my city because he was on the run for a murder he committed in another city many years back. The nicest guy in the world had an altercation with another patron at a bar, while drunk, broke a beer bootle on a table, stuck into the stomach of the other guy, and twisted left and right until internal organs started showing.
I did befriend a murderer once.
I expect a decent percentage is total BS and too big to get rid of without messing with their quarterly numbers.
Of course, it would be pretty simple for google to prevent through a sort of advertiser FICO score. But they don’t, because it would make then less money.
I think the easiest thing to do would be to pass a law or regulation requiring companies to be responsible for the ads shown on their sites. If, say, the NYTimes is promoting scams (they are) they should be fined or worse.
EDIT: per project, not per month
EDIT 2: digging out an old bill from 8 months ago when we bought the server, $14k per month is what we were running and it was going up, a lot.
We will break even from our old Claude spend in about 2 weeks.
It services 30-40 analysts and staff concurrently and is faster than any of the SaaS out there for us.
This model is for researchers who already have racks of equipment, inference providers who sell access to models, and for businesses who must (or choose to) use on-prem models. All of which have to pay to use the model if their revenue rises beyond certain thresholds: https://huggingface.co/moonshotai/Kimi-K3.1-Pro/blob/main/LI...
I wouldn't over-weight internal benchmarks from a competitor that claim benchmaxxing TBH. Kimi K3 has been at the top of Design Arena for open-weight models for a long long time, and I suspect K3.1 will be there for a long time too. It's pretty hard to fake that one since it's actual human preferences for the artifacts produced.
> Advancing the Frontier of Open Intelligence.
Too bad you probably need something like an 8x B300 node to run this. Black Friday is coming soon though, but I fear this open intelligence will be pacing anyone's liquidity frontier so I'll probably keep trailing the frontier for a while with nvidia/qwen3.8-27B-NVFP4 on dual RTX PRO 4000 in TP=2. I can almost fit 450k context in there.
Sharon was the AI Chief of Staff brought in through an acquihire, managed all the high-visibility external/internal AI engagements, reported directly into Lisa Su, and was being groomed to be Lisa's successor. Even with all that, she couldn't make it 2 years at AMD.
I don't see any way out of this hole. We can't fire our execs and promote internally because our internal directors/VPs are mostly non-technical. We've been trying to pull leadership in with acquihires but we've been losing them too. Best win was replacing Alan Lee w/ Anush Elangovan which finally got ROCm working but there's a massive re-org coming he might not survive.
The board should've sacked the entire leadership team of Intel and AMD years and years ago.
The fact that you can sit through both a mining boom and now a generational AI boom and have little to show for it and still keep your job is a Yahoo!-level fumble.
The rule includes >90% of "remote" workers who have discovered they were classified as hybrid (a classification now eliminated) and their manager didn't make it official. Now, remote approvals are SVP level.
The issue is, AMD still pays significantly less than any company with an RTO mandate, even with the stock appreciation. And the stock has gone down 25% in the past few weeks so it's even less competitive. The typical engineer was not given an equity-weighted package, even L9 only 30% of your TC is in stock.
It's a recruiting bonanza right now and everyone's getting calls offering 50% pay bumps from 5-day office semi corps like Broadcom or Qualcomm. The better people are getting offers at Google or Apple for 3-day RTO for even greater TC bumps.
Optane had much lower latency than other types of durable storage, and also had very good endurance compared to NAND.
At a prior employer, we made extensive use of them as ZFS SLOG drives for large database servers. They made fsync very inexpensive.
In the last few years Intel's division that makes Xeons had huge losses, because in order to retain their big customers they had to accept to sell at big discounts, otherwise the customers would have switched to AMD.
All has changed this year, due to the AI craze.
AMD has sold all the server CPUs that it could make, within its allocations at TSMC. Everybody who did not succeed to buy EPYCs has bought Intel Granite Rapids Xeons, because they are not made at TSMC, but in Intel's fabs, so they did not compete with other TSMC customers for allocations.
So Intel has also sold every server CPU that they could make, and they are now trying to retool some fab that made new Panther Lake CPUs back to the obsolete Intel 3 CMOS manufacturing process, in order to make more Xeons, to sell them as long as this huge demand persists.
Thus in the last quarter both AMD and Intel have sold less consumer CPUs than in the past, because all who could did delay their upgrade plans due to the unacceptable memory prices, but on the other hand both AMD and Intel had record revenues and profits from selling server CPUs.
If you wanted a GPU with 32GB of memory for AI tasks, the AMD AI Pro R9700 was (up until about a month ago) reasonably priced at ~$1300. It's not a RTX 5090 in terms of bandwidth and compute, but for the price and being actually purchasable it's a reasonable pick.
Krzanich "optimized" the company for profit, by cutting out things he considered to be excess. Including a lot of their irreplaceable validation & verification talent, that AMD instantly snapped up. The lack of validation resources is directly responsible for the failures of many key Intel products during his tenure and later, because major bugs that could have been spotted earlier slipped through, and then resulted in horrible delays and even cancellations. He then spent the surplus of the company on 30 major acquisitions, of which only two were in any reasonable degree successful.
There are CEOs who have destroyed more value than Krzanich, but not many.
Honestly both of them have demonstrated that they have little interest in AI. Intel gave up on Optane right as (a much-improved successor) would've helped with MoE offsetting or even as a highly durable drive for KV caching.
AMD have... turned out some decent gaming graphics cards and fine server CPUs but little else.
People are out there solding SMDs on old Nvidia mining rig GPUs to use them for inference; their hardware is not depreciating at the normal rate you'd expect from GPUs. DGX spark is finally useful now that AI can write multi-machine inference tooling optimised to it. Where's AMD in all of this? Nowhere.
This is a lack of linux driver support for firmware designed for windows laptops.
This is a problem on the MSI Prestige laptop using the RTX Spark platform. Lets assume it also applies to the Asus and Dell machines using RTX spark.
RTX Spark is a new platform and all of these laptops ship with Windows. You cant get Linux on any of these machines from the factory. There is also precedent for NVIDIA shipping linux support well after Windows (2 years in the case of Dynanmic Boost, a similar power sharing feature).
This is a linux driver problem. On laptops designed for windows.
Read the article and stop sensationalizing.
This is a problem on the MSI Prestige laptop using the RTX Spark platform and Linux. Lets assume it also applies to the Asus and Dell machines using RTX spark and Linux.
RTX Spark is a new platform and all of these laptops ship with Windows. You cant get Linux on any of these machines from the factory. NVIDIA may even ship drivers for linux later as they did with Dynamic Boost.
Kindly remove this from your list of "Evidence that NVIDIA is Colluding to Centralize Inference" and everyone else take note of this confirmation bias in action.
I don’t think anyone “allowed” them - quite the opposite. They’ve made generational bets that ended up being right, bets that IIRC the industry did not agree with for a long time. Them pursuing CUDA resulted in a draw down of something crazy like 75% of their valuation. They bet the farm multiple times to get to the point they are - and it’s only “obvious” in hindsight.
If you want folks to make generational bets as a matter of course (which I think most folks do), centralizing the downside of those bets and socializing the upside of those bets is not a great strategy. Almost as bad as socializing the downside.
Competition will eventually close that gap. We’re only a few years into this thing.
Nothing stops you from giving Nvidia half a million for their latest and greatest AI GPUs, unless you are in China I guess, but China does very well despite the export control.
It is just classic market segmentation, the thing Nvidia and just about every company that can benefit from it has done for decades. Cloud providers are ready to pay a lot of money, consumers wanting local AI have a lower budget, and gamers even lower. But a low budget is greater than zero, and there is still money to be made on these markets. But they also don't want to compete against themselves, so they are going to find ways of crippling enterprise-class workloads on consumer hardware.
It used to be that they crippled high precision arithmetic on consumer hardware, because it was a differentiating factor, but AI doesn't care about high precision, so now, it is something else.
Pay up for license to unlock full capacity, pay up for license to use encrypted models from AI moguls, paid per billion tokens processed or just operations in GPU
Sovereign AI has never been more important.
Controlling AI is so much more powerful and impactful than controlling search or social algorithms.
Centralized AI a 10x on social control over everything we have seen before, and there is no way that a free and open society survives contact with centralised, high capability AI if the system is manipulated towards guiding outcomes. It’s essentially mind control, as people will offload an even greater share of their cognition to AI than they do with search and social.
Build and maintain sovereign AI while you still can.
Man it really is exhausting to read a title that sparks my interest, just to be greeted with the most basic Claude-isms in the subheading.
> VerInf proved Llama 4 Maverick (the same 400B-parameter MoE model we prove) at ≈0.02 tok/s.
The problems of inference proof for a closed model provider are more difficult than an open model provider -- you can get a probabilistic proof much more cheaply out of four independent inference providers running the same non-batched prompt than this 90-100x compute hit.
Anthropic and others also have a significant problem with these proofs in that they have a lot of incentive internally to have some portions of the pre-prompt secret; I haven't read this newest scheme closely enough to determine if they have a way to hide parts of the prompt with the proof. But if they don't, they are almost certainly not going to want to send these proofs out in the general case.
Last, I'll take this moment to soapbox and say for the researchers out there -- provable inference is nice. Provable training corpus information would be spectacular. It would be very, very, very nice to provide a set of model weights and a proof that they had only been trained on a dataset. To be clear, this is a really hard problem, but it would change fundamentally what levels of trust we could have about a given set of weights.
> we are open-sourcing two foundation models co-developed by Hy4 and Hyra 2.0: HyCell for single-cell representation and HyPDE for physical field modeling.
What a refreshing contrast. Where American startups brag about stopping researchers from writing grants about neglected tropical diseases and making novel anti-depressants, these folks are open sourcing the means to do science. https://news.ycombinator.com/item?id=49651727That's how you get a civilizational advantage. By allowing people to just do things. It's so bizarre to watch China embrace the wonders of science and technology, while certain American startups and communities repeat "WHAT ABOUT THE BIOWEAPONS?"
The American century has been fumbled hard.
> For game development, we partnered with professional developers at Tencent Interactive Entertainment Group (IEG) to curate high-quality training data and build evaluations. We focused on feature implementation, gameplay logic, state tracking, and asset pipelines. We also extensively adapted Hy4 for major engines such as Unreal Engine, Unity, and Godot, and expanded its capabilities to multiplayer game development. Compared with Hy4 preview, Hy4 improves by 19.1% on GameDev and 9.8% on GameLogic.
Here's more about that game development studio - https://en.wikipedia.org/wiki/Tencent_Games - they've been around since 2004.
I like that they acknowledge that. For working with ai as a tool open models are already great. For millennium problems and doomermaxxing they are not there yet.
Over ten years my brother and I went through a set of experiments (stop and go, turn, communicate) while probing the memory. We uncovered where the variables were, how to inject calls into the game functions, and how to sync the gamestate across the network, so we can fly the story missions together.
My favorite finding is that the camera lives in entity slot 0x3D, which is possibly a "3D" reference by the original developers.
You can play it at https://fireflyk64.github.io/dosbox in browser. If you own Wing Commander, drag in the installer since no game files are hosted. If you don't, the page will load the original DOS/3Dfx version of Vega Strike, before I open sourced it. I upscaled and redid DOS Vega Strike art using AI, but it now has multiplayer and a similar feel.
I'll be in the lobby for the next few hours if anyone wants to fly together.
Disclosure: I also used AI for the final step to help compile for the browser and port our TCP networking stack to WebRTC with voice and solve a few lingering bugs.
Very exciting to watch a completely new browser/browser engine come together. Reminds me of the early days of Mozilla and Gecko
Posted with Ladybird :)
Tech is now viewed in the same lense as dirty greedy oil companies. If an oil company opened a candy store most people I know would not go because "Oh those oil companies can't be trusted"
Anytime I talk to anyone not in tech they're extremely sceptical of anything tech companies do. Most common reasons I hear as to why is greed, social media destroying children in many ways, seeking political power, and now many parents are concerned their child won't have critical thinking skills because AI is here to think for them.
I don't think they're crazy or obviously wrong. It breaks my heart to hear tech leaders chalk this all up to "AI has poor messaging" as if you say "Our lies have got to be better so we can make the public believe what we want". I've never seen tech so out of touch.
Er... I mean, if that's the case, then, fine, I guess?
I do not believe that this is actually the case. People build heavily regulated industrial facilities all the time.
- Any new datacenters being built cannot raise utility rates
- Some of the revenue generated by the datacenter must be set aside to improve power reliability
- Datacenters must not touch our waterways at all
- Datacenters must be built using union labor”
Deal. This sounds like a great list. I would add, “no property tax breaks”.
Just get the state to pass this as a law and set appropriate enforcement mechanisms and then just let data centers go ham if they meet this criteria?
No more “per community meetings” and municipal haggling (see: corrupting) local governments. Buy the land build the data centers. The municipality can figure out the exact benefits that works for that community from the giant tax coffers.
This seems imminently fair to me.
* We've built a technology that has obliterated entry level hiring
* That technology might obliterate all the other "normal" person white collar jobs. The kinds of jobs you can clock in and out and not think about them outside of that. Instead, every new job will be a managerial one working with 500 agents.
* Loaning all the money to build these data centers and train these models costs so much it cranks rates through the roof, so we've frozen the real estate market and nobody can move or buy in
* The past 10 years of American politics have been defined by how the last time we did a broad replacement of labor event like that for a group of people. Their standard of living permanently shrunk (the NAFTA disrupted working class), and we may end up doing that to you guys again. The only solution is that the government is going to have to do some sort of UBI or redistribution (which the last time they did during COVID, it was followed by the greatest inflation we've seen in our lifetimes)
* This technology might literally destroy the world in the next 10 years, and we keep accidentally releasing it into the world like its Skynet. But we are not responsible for it, and by the way its unavoidable with this technology. Somewhere on the other side, utopia will arise (No really your standard of living will actually go up after losing your job this time unlike everyone else, trust me.)
Then half the people in this thread is smugly are acting baffled at why people don't want this, and going "Hey they don't know that much about data center logistics but still oppose this. What is up with that?." Forget about if a data center actually ends up being a net drag if its near you and the standard nimby stuff. Even if it weren't, what benefit does marching towards this future have for anybody not in the top 10%? Sure, you can make a bunch of economic theory arguments: growth is always good, stagflation, Luddites, etc. But in practice, economic advancement and technology over the past 30 years has advanced so fast people can't even comprehend it, and in return everybody is becoming stupider, more unhappy, polarized, and the median standard of living has dropped as rich people have 20x their net worth. Why would a normal person want more of this?
And that's even before we consider the impact of LLMs themselves. Personally I expect LLMs to create tremendous growth, but that should be highly counter-intuitive. They're going to absolutely wreck cognitive labor; society entering into an era of deteriorating wages for cognitive work as well as increase unemployment is quite likely. My prediction is based on 'what comes next' but if somebody's only concerned about paying rent next month, this is all going to be, again understandably, seen as pie in the sky.
This is a once in a generation opportunity to get private investment for solar farms from multi-trillion dollar market cap firms and we're squandering it in order to construct noisy, polluting behemoths that belong firmly in the prior century. Talk about wasting our political capital.
I think they have very rationally concluded there is no economic growth benefiting them in massively consolidated capital building the infrastructure to replace their small businesses and jobs.
- There must be community benefits agreements negotiated transparently with the local community
- Any new datacenters being built cannot raise utility rates
- Some of the revenue generated by the datacenter must be set aside to improve power reliability
- Datacenters must not touch our waterways at all
- Datacenters must be built using union labor
A lot of the communities that had datacenters built in them ended up getting steamrolled in negotiations by the tech companies because you had the local government (in small towns, this is people who work part time making a few thousand dollars per year) negotiating a deal with the Google/xAI/Oracle/etc corporate lawyers. I think it makes sense to say that the buildout should be paused until fairer terms can be worked out on a national level, rather than letting this continue to happen.
Generating their own electricity how? I would be pissed if I got one of these xAI power turbines near my area [1]. And even if they weren't loud, gas plants are environmentally bad.
> They're generating their own electricity and excess power will be put back into the grid
Electricity prices will go up! The data centers won't keep generate their own electricity if they aren't forced to. They will just take electricity from the grid when it's cheaper than generating, and will start generating only when the costs of generating are lower.
[1] https://www.nbcnews.com/news/us-news/musks-ai-power-plant-ge...
Yeah, from natgas right? Or maybe diesel generators? Or are they building one of the practically nonexistent green energy data centers?
There is also the increase in cost of borrowing for everyone. Going this fast is just insane.
"It'll use too much water." - It's going to use less water than the farm that used to run on this land. The county is gaining water from this.
"Our electrical bills will go up" - They're generating their own electricity and excess power will be put back into the grid
"Noise" - It's an hour away from any structure on a stinky swamp north of the Salt Lake
"Wildlife refuge" - This bird refuge is in Brigham City. The data center will be on the opposite end of the county. Some people seem to think it's right next door.
My concern is I don't want tax payers left holding the bag if this falls apart but it seems like, at least with this DC, that that's not the case. This is going to bring a LOT of money to my county which sorely needs it and a LOT of people to my dying rural town.
I really just have to come to the conclusion that we have a tiktok informed voting block here. I've had people ask me if I'm worried that the new data center will spy on me. What?
Edit: I checked. This is going to use 162 million less gallons per year than the farm.
Tech “ownership models” have accelerated the separation of haves and have nots. I’ve wondered if datacenters were regulated such that they become public goods (libraries, power and water infra, etc), if the public would invest more agreeably in them.
Just policy thinking out loud.
* https://en.wikipedia.org/wiki/Kepler%27s_laws_of_planetary_m...
People did not understand how or why they worked, but they switched over to them from the Ptolemaic (Tychonic, etc†) system anyway because the math was easier. It was only later (Newton, Einstein) that we understood the "why" and "how".
† https://tofspot.blogspot.com/2013/08/the-great-ptolemaic-sma...
The problem is the systemic disincentivation of true science, the pursuit of truth and knowledge unencumbered by greed.
When you speak favorably of "factories for research", you adopt the position of a mindless consumer, relishing the prospect of "abundant" products of science.
You fail to notice, livestock lives in abundance as well.
Fields medalist and as Tao listed as one of the most important mathematicians of our age.
Scholz is known to find „precisely the correct and cleanest way to incorporate all the previously done work and find an elegant formulation for that — and then, because he found really the correct framework, go way beyond the known results“
In a way that man is the complement of an AI, he generates understanding rather than just proofs.
Maybe he is the right template for the mathematicians of the future.
https://www.quantamagazine.org/peter-scholze-and-the-future-...
Even better, why not have the agents sleep for you?
EDIT: for those who say this is not realistic, I've had Claude sleep for 10 hours in just 30 seconds. Saved me an entire night.
I wouldn't say this was the norm, but I have burned through £1k of tokens in a couple of days before. And, on that same gig, I realised a test I was considering running would burn £2.5k worth of tokens over the course of a few hours so spent a couple of hours of my own time re-engineering it to get that down to £150.
The point is that I don't necessarily know that spending a shitload of money on tokens really tells us anything useful. If he's getting a return on that investment, fine, whatever.
Obviously most of us couldn't afford to spend that much ourselves, no matter how much value it might deliver, and many of our employers wouldn't be too happy if we spent it on their dime either. Equally, if you can, and it's delivering real value, then I don't see why you wouldn't?