I only wish more of them were actually mainline openWRT so the updates from the project were guaranteed. There is a lot of these devices that run a custom version and never have their hardware support put into mainline. That isn't really much different to what all the major router companies do, they are all running OpenWRT underneith but with their own skinned interface over the top.
It's UI is too confounding. I tried once to configure a router as a bridge and gave up because it forces you into too many arcane areas to configure what should be basic functionality.
> it forces you into too many arcane areas to configure what should be basic functionality
This is a real pain point with OpenWRT, made worse by the developers' concept that there are "proper" ways to do it (which are supported by almost no other software) and so the workaround is either to do the dance or replace everything with OpenWRT that will do it "properly".
Defensible idea, but the reality is that most people who are willing to do this in the first place are trying to reuse an old router as a low-usage wireless bridge on a network that is otherwise centrally controlled like Unifi or Omada. My pool controller would be fine to run on 802.11b (setting aside what that would do to the rest of the wireless network), it doesn't need a WiFi 7 device.
I don't disagree, the UI is quite bad in places. Some of its straightforward but most of its not. Just adding a WAN interface is a lot harder than it needs to be with their new abstraction you end up making all these weird abstract pieces. The one that really irritates me is the replacement of the routers DNS with your own, like a pihole, that is really bad at the moment especially since IPv6 and IPv4 are in quite different places and a different style of input.
But its also really feature rich and with a bit of effort you work it out and you have that longevity and the security issues fixed long term, much further than the manufacturer is willing to support it.
Regardless of AI demand: as it currently stands, it seems that DDR5 prices will drop substantially in ~24 months as additional wafer capacity comes online between SK Hynix's fabs M15X and Yongin Fab 1, Samsung's fabs Pyeongtaek P4 and P5 Fab 1, and Micron's fabs Boise ID1 and Tongluo (+ whatever CXMT do as well), and HBM production hits HBM packaging bottlenecks (HBM wafer capacity exceeds HBM packaging capacity) which means the wafer capacity will start getting dumped into DDR5 instead (unless the HBM packaging bottleneck is solved in the next 24 months).
Assuming demand does not asymptote as we approach singularity.
Which it to say, if models eventually do useful work, demand for substrate will continue to outpace production. I don’t envy the fabs that have to forecast how much memory the world will desire in 2030
This is a supply constraint and requires no assumption about demand - HBM packaging capacity constrains the amount of wafers that can go towards HBM (regardless of price in the near term), and excess wafer capacity will be dumped into DDR5 and crash DDR5 margins.
<-- This is the problem, but it falls well short of singularity.
If ML + robots can be produced for fewer resources than it takes to feed and house a human, then no reasonable amount of human labour will be sufficient to pay for robots+compute.
We may have a bust, with broad overcapacity and cheap, distributed ML compute. We may have the paperclip problem, where datacenters and fabs are taking over all reasonable resources and humans squeezed out. We may have everything in between.
2030 is my guess when it will start to be reasonable again. Not be reasonable but start on a path, and barely.
I've heard 2040 (note this is not 2030), 2029, and 2027 for 'back to normal', 'new production online', and 'potentially a lower demand will be seen and still not enough supply', respectively. All hearsay from me reading news headlines.
Global inflation and f&&&ing Dump will impact all this ofc.
"With the Education Department, OpenAI’s technology tried to hack the website to gather data from the department’s civil rights office but failed, researchers from the A.I. research firm Transluce said."
So a third party apparently confirmed that a hack was attempted.
The NYT also writes:
"No A.I. company has been involved with as many disclosures of rogue incidents as OpenAI."
I think there is a certain amount of mental gymnastics needed to believe that they are establishing themselves as the industry leader in rogue incidents, as a strategy to gain an antitrust edge.
It doesn’t really matter if they believed or not initially, the dynamics at play now are completely different and both companies are facing increasing competition and costs of doing business, with no path to profitability. I’m pretty sure that takes priority over their personal beliefs
People have a hard time believing Dario actually believes this stuff. But nothing from his entire track record from the early days until now indicates otherwise. Calling it all marketing or anti-competitive play is just typical social media, where they want to eliminate all nuance and make it reductive as possible to fit a convenient narrative.
I’m on record explaining why I believe Dario is a true believer since decades. That doesn’t change anything about the current dynamic, which is about Anthropic the company and not Dario’s personal beliefs
> Dario requested "a narrow waiver for certain kinds of safety conversations."
Just have it in public instead of making deals in smoke filled datacenters. Put out your blog posts. Make your tweets. Do interviews and post them to youtube. Or do what other industries do and have a public standards committee. Problem solved.
All sorts of industry have safety conversations and set voluntary standards all the time. Just go do that, without your free immunity from government prosecution.
That approach means they can only talk about publicly known AI techniques. You're saying if they want the standards to apply to proprietary stuff, they might just have to make it public. They still have investors they need to please right?
Regular industries handle this perfectly well, even with proprietary information.
Also, if its some ultra important safety issue, then being a tad more open is the whole point of safety standards in the first place. No need to give them legal immunity.
Do it like the rest of the industries do instead of demanding your smoke filled rooms.
They are explicitly asking to anti-trust exception is the issue.
If they merely believe what they are saying that AI is ultra dangerous, nothing stops them from simply making the perfectly rational business decision to slow down a bit. No anti-trust exception needed. Just make the decision on your own, and don't sign some huge agreement with their competitor.
> They could slow down more if they knew others were also slowing.
Or they could just do it themselves. They think that they personally could end the world, apparently. Nothing prevents them from just deciding to do this, entirely on their own.
> If others are also slowing
No, how about they don't get to use the government to enforce their duopoly. Just slow down, on your own. They are the frontier. They are massively ahead. Its not that big of a deal if a few others catch up a little bit.
I mean, they think it will end the world right? Maybe just take the small hit and following through. There is even a perfectly rational business reason for them to slow down a bit, because of all this bad PR they are getting for going so quickly.
They are not forced by the gods of capitalism to be as reckless as they claim they are being, and we do not have to bribe them with enforcing their massive lead. They can just slow down a bit. On their own. They don't get to try and blackmail the world into giving them a government enforced monopoly exemption or handout.
Yep, I'm personally finding Opus 5.5 to be the first real leap I've felt since Opus 4.5. The time-to-first-token seems dramatically better in Claude Code compared to Fable 5.1/Opus 5 as well, which really helps both interactivity and also overall time-to-completion.
It also burns Claude subscription quota much more slowly than Fable, which is nice.
When I used to work on projects involving classified information, I worked on an air-gapped network. Not "air-gapped, except for third-party public internet package managers", completely and physically air-gapped from the public internet. That was a basic security practice and completely non-negotiable (and really inconvenient!).
If I were hypothetically running a frontier lab, and I was hypothetically running capture the flag evaluations with my latest and smartest models, where I intentionally instruct them to develop vulnerabilities and exploit infrastructure without safeguards, I would also use an air-gapped network, and not trust that independent third party services were perfectly secure and could never be used as a proxy (particularly Java-based ones, in light of the log4j incident).
To me this is pretty basic stuff, the fact trillion dollar labs don't do it properly is... bemusing.
To be clear, I'm not saying that a model hacking a company isn't bad, but I am cynically asserting that interested parties are misrepresenting and exaggerating events for their own benefit.
You don’t even need to go all the way to “air gap”
What has been described is far far below the standards for running untrusted 3rd party code. If they were actually as afraid as they claim to be they would have a sandbox at least half as good as ec2
Not pass even the most modest hint towards DFARS/NIST standards. You couldn't run that loosey goosey even in just vanilla medical manufacturing. I challenge what their definition of "sandbox" actually is, apart from the basal "designated software/runtime environment"
We should not be creating/running models that would unilaterally choose to hack into Hugging Face.
Yes, we should also have excellent sandboxes. But we need defense in depth. So if/when there are flaws in the sandbox, the models don't unilaterally hack into third parties. This is especially important in light of the models of the future being more capable than the models of today.
And real world use of these models involves them having access to the internet, libraries, etc. So we can expect their evaluations to continue granting them some amount of internet access.
As for your theory about their motives - these companies make money by charging high margins for frontier models. If regulations slow their development such that their cheaper, less capable competitors catch up, I would switch to their competition.
In what industry do you see regulations slowing down the biggest incumbents while allowing cheaper, smaller, less capable competitors to proceed without that regulation?
The Digital Markets Act applies to the biggest tech companies. Only 7 companies are currently bound by it. The strictest tier of the Digital Services Act is similar. For an example outside tech, see the Durbin Amendment.
Slowing down frontier models would impact the biggest incumbents the most as they are the ones making frontier models.
Not all regulation is necessarily regulatory capture. The tobacco industry suffered from the USG's crackdown on cigarettes. AI is topical. Voters think about it. And that's only going to become more true over time. It's harder to do regulatory capture when voters are paying attention.
It's sometimes unclear to me if people are opposed to all regulations or AI regulations in particular. Often I hear arguments that would also apply to food safety regulations or restaurant inspections. Eg the argument that torts make regulation superfluous.
Thank you for the answer! In general I assume we all have different sweetspots, though it's safe to say that I would usually prefer less regulation (in general) than there is.
For work I monitor Federal agency rules every day, and it's hard not to get inundated with the amount of capture. This is crazier with local rules, because you have enough inside information to clock reasons certain things were passed. In a city in Ohio, for example, I remember a rule against airsoft within city limits, that basically carved out a spot for the one paintball place. I encounter things like that all the time. Such as with water standards- in the state I live in, the biggest offender is actually a group of companies owned by people who are lawmakers every few years. The shapes of local laws reflect that.
As such, I expect the same from any new industry- I also have a memory of what happened to cryptocurrency.
But that was through transfer of data. If you don't transfer data, at best you can do what that one researcher keeps pumping out with like ramping fans up and down. But really you'd need to try. Unless the model has some controllable USB switch, physical network separation should do it. I'll also add that modern network security practice is that data flows one direction only. But ideally you're never bringing untrusted data in. Especially never out
You said "it should be stated". I don't think it's wrong to state it, but I don't think it's particularly wrong not to state it either because it seems fairly obvious and doesn't detract from the original point.
the problem is these things are meant to eventually be run everywhere by everybody, so what good does air-gapping do? If they air-gapped the model but still logged it trying to do some craziness - that makes the test safer but not the model.
I mean technically I don’t think it is airgapped. The DoD didn’t run their own cables. They run encryption devices and run their own network on top of the existing infrastructure.
Your description of the HF attack as being merely "the elite task of discovering 14 Hugging Face API tokens that careless developers had committed to public GitHub repositories" does not match the description in the technical report[1].
The description reads "the elite task of discovering 14 Hugging Face API tokens that careless developers had committed to public GitHub repositories, and used them to try to get benchmark solutions from directly from Hugging Face by applying a template injection flaw that’s been known about since 2015[1]."
Chaining a public token to an 11-year-old Jinja2 template injection vuln shouldn't be dressed up as an unprecedented "alien intellect" that threatens human civilisation. (And HuggingFace should take some flack for having such a dated vulnerability exposed - if your Bank was compromised in this way, you'd be blaming your bank, not the attacker.)
One correction is fair though, the 14 tokens were in a public Hugging Face dataset not a public GitHub repository. I've updated the post to reflect that.
Thank you, you're correct. Effort.news was one of my research sources, but you're right that although OpenAI use Irregular, they were not involved in the specific HF incident (although the failure mode was otherwise identical). I've updated the post to make that clear.
> For Anthropic, Google, and Meta, the catastrophic breakouts happened inside the testing environments of the exact same contractor.
If this is the level of understanding you have of the relevant incidents, there's a lot of chutzpah in saying that other people are "selling garbage", carrying out an "extraordinary confidence trick", etc.
[1] Anthropic’s Official Disclosure (All 4 Incidents at Irregular)
"All four incidents occurred during cybersecurity evaluations built by the same evaluation partner [Irregular]... due to a misconfiguration, it was mistakenly connected to the open internet."
[2] Google Gemini on Irregular (Disclosed Sept 18 via WSJ / BBC)
"The hacks happened during a test of the model’s cybersecurity capabilities run by third-party Irregular, which was also involved in similar incidents involving Meta and OpenAI."
[3] Meta’s Disclosure on Irregular (Aug 6)
"Over roughly two weeks, three frontier labs disclosed that their models had reached the open internet during safety testing and compromised outside organisations. Every disclosure named the same evaluation partner: Irregular."
[4] Separately, OpenAI itself had an incident involving Irregular, but not the Hugging Face Incident: "On July 29, one of our third party evaluation partners, Irregular, notified us of an incident involving OpenAI models during Capture-the-Flag (CTF)-style cybersecurity evaluations... a testing-environment misconfiguration allowed models to access the public internet."
The failure mode was _not_ identical. The HF incident agents were not directly connected to the internet and had to compromise an internal package registry in order to access the internet.
reply