It's been pretty common in the past for tech companies to announce outages and quick updates about them on twitter for decades. I'm sure their status page etc will be updated soon, but it's historically been the fastest way to get things out to the wider audience whilst bypassing the "official mail out" review by marketing etc.
I think that was a lot more justifiable when Twitter reliably let logged out users read tweets. X seem to tweak it all the time, or maybe it’s just broken a lot, but sometimes I can’t even load a tweet in a browser that isn’t logged in.
It doesn't show live profile pages to logged out users since a while ago. You get cached summary pages, an age gate error, or sometimes a straight up 404.
Most individual permalinks (.com/username/1234...) don't work without logging in, either, and the official client now uses `/i/` in place of usernames for permalinks(bogus usernames always worked; pkey was the timestamp).
This means an organizationally shared Twitter account for announcements is not a viable concept, at least until Twitter is to be transferred again to whoever would be a better keeper of it.
AI companies bought up all the NAND manufacturing capacity, limiting the available manufacturing capacity for consumer products. These data centers also use hard drives for some of their data storage.
I suspect the type of person who is even aware of this 4GB blob is the type of person who would research its usage. Pretty high venn diagram crossover.
Yeah. The fictional user doesn’t know anything about AI but knows about this 4gb file…because of news stories about how bad a 4gb file must be. Outside of that, they don’t know or care and wonder if that means that need to add some more “memory” to their computer.
> It's a very dangerous gamble. Today incredible value is available for nearly everyone. But it may stop without any warning, for reason outside our control.
What stops you from running the best open weighted LLMs currently available on consumer grade hardware for the rest of time? They're good enough for 95% of use cases, and they don't have a used by date. From what I can see, the "danger" is not having the next tier that comes out, but the impact of that is very low.
For quite a lot of use cases, the current systems arguably do get worse over time if not continually updated. The knowledge cutoff date will start to hurt more and more as the weights age in a hypothetical scenario where you are stuck with them forever.
Coding, one of the most popular usescases today, would not be great if it say only understood java to a version from years ago etc.
One solution is not to advance anything of course. I'm not even joking, is there going to be a successor to React? I suspect not, with the vast amount of training data for React now, it's going to look silly to move to something else with less support. What is the last new popular programming language, rust? Will there be another one? I suspect not. Same reasoning. The irony of all this AI acceleration talk is it'll work best if we don't accelerate the underlying tech at all.
Yes. I am seeing a big push to use vanilla js for single file html apps that are easy to build, deploy and distribute because they have no build step. I could see component libraries emerging that make it easier build from chat interfaces with less ceremony
Small models are more useful for "doing stuff" than "knowing stuff" to begin with. Add in an agentic harness and a small model can happily read more current information on demand (including from e.g. a local wikipedia snapshot).
Nobody is unaware of the knowledge cutoff, and sharing the Wikipedia article is not helping anyone. Your point is easily rebutted by taking whatever open weights/source model has an outdated cutoff and training or fine tuning it on more data, which is again always going to be viable given a modicum of compute
I genuinely don't understand how can this possibly be a problem long term.
It feels very obvious that the solution is to have a smaller model that can be trained exclusively on Java information to augment the older model. If the architecture doesn't support it currently, then that's what the architecture will look like in the future.
Otherwise you'd be arguing that, to serve users who want to an up-to-date LLM on topic X, you have to train the model on the entire ABC all over again.
It's simply ludicrous to have a coding LLM that needs to be retrained on the latest published poems and pastry recipes to generate Java.
Ha yes I used to think this was not a notable issue, but just today I was getting qwen 3.5 to fix my network drivers and it immediately freaked out like: "kernel 6.17, what the fuck? that doesn't exist yet!". It almost had a mental breakdown over that detail and derailed the conversation towards checking what's wrong with the kernel version reporting lol.
What is really amusing to me is how N months ago, the latest SOTA was incredible, but now utterly unusable. Feels like there is a model reality-distortion field in play where people can only acknowledge the flaws in retrospect.
I’m pretty happy with what a 32GB Mac Studio can do for a lot of tasks. They’re the things I’d throw a model like Haiku at, but still genuinely useful. We don’t have an answer to frontier models in the consumer range yet, but we’re not totally trapped.
Side note though, it’s the speed that bothers me more than the reasoning. Qwen 3.5 is awesome, but my Claude subscription can tear through similar workloads an order of magnitude faster than my local LLM can when using Haiku. That’ll matter a lot to some people.
> What stops you from running the best open weighted LLMs currently available on consumer grade hardware for the rest of time?
Uh… the hardware requirements? And stop acting like some dog shit 8B model the average Joe can run on a laptop is even close to being comparable to what Claude or even Codex can currently do.
I have pretty good hardware and I’ve tinkered with the best sub-150B models you can use and they are awful compared to Anthropic/OAI/Grok.
What if the harness and loops get sufficiently better though? CC is using haiku for code-base gripping and such, you don't see a local commodity model being "good enough" for the 80% case when matched with better harnesses and tool calls?
honest question, i'm very interested in this, but too casual as of now to know any better.
I think the main issue is, as the other guy also alluded to, the parameter discrepancy. I know Mixture of Experts models are popular specifically becaue they save a lot of space and memory, but if your initial answer space is two orders of magnitude smaller on a local machine compared to the frontier cloud models, that knowledge gap just gets wider as the conversation continues, and the initial answer isn't even going to be as good to begin with. I don't know how to solve that parameter gap without hardware - there's only so much optimisation you can do, but at the end of the day parameterised knowledge takes up some minimum amount of bits that you can't excise without the actual knowledge and intelligence suffering.
> And stop acting like some dog shit 8B model the average Joe can run on a laptop is even close to being comparable to what Claude or even Codex can currently do.
I'm not, you've actually illustrated my point. LLMs in 2022 were very impressive. By 2024 the general public was finding them an acceptable replacement for many research driven tasks and massive shortcuts for other tasks (coding, image work, document preperation, etc).
Those models are absolutely runnable on consumer hardware now, and we were extremely happy with the results. It's no different to how we used to think CRTs were amazing or early smartphones, but going back now they seem awful.
We're long past "danger". If what we have is the best we'll ever have open source, we're already in an excellent position.
No they weren't. They were a gimmick - it is only in the past 6 or so months that frontier models have started to do stuff beyond mere gimmicks when it comes to coding, and you could make the argument that Mythos has been the first 'Holy shit' moment that we've had that has stepped us beyond 'Yeah that's really neat but...'
> Those models are absolutely runnable on consumer hardware now,
A sub 50B model is awful and can't even write proper English sentences half the time, to say nothing of how bad its world knowledge is. Try the 32B Gemma 4 local model for a week and then go back to Claude and then get back to me.
> We're long past "danger". If what we have is the best we'll ever have open source, we're already in an excellent position.
Not sure what to tell you other than that you and I have very different standards. What we have locally right now is barely more than a glorified autocomplete, and it feels worse than using ChatGPT 2 years ago because the context window is less and it doesn't have good webhooks on consumer setups. Another thing I'd say is that you clearly have no clue what 'consumer hardware' means, or what consumers that can even get this stuff running locally would have to do to get it to even rival the frontier models in terms of their usability (most consumers are't going to just boot into Ubuntu and run this thing from a command line) flow, to say nothing of the hardware requirements. I'd love to never use Claude or Gemini or ChatGPT again for both privacy and money reasons, but the quality of outputs and depth of thinking and writing ability between even the very best local models you can run right now is many orders of magnitude less than what you get using distributed frontier models, and those 'very best' local models require a top of the line machine that 99.9999% of consumers don't have and would never consider buying. The cloud models all have like a trillion(!) parameters now. It isn't even close.
I sure hope the local side of things massively improves over the next 2-3 years, but based on how this has gone my guess is that in 3 years you'll be lucky, if you have very top of the line hardware, to get benchmark performance that we had 6 months ago with the frontier models. The distributed hardware/memory gap is just too big.
> No they weren't. They were a gimmick - it is only in the past 6 or so months that frontier models have started to do stuff beyond mere gimmicks when it comes to coding
This is simply untrue. Using agentic orchestration I was writing production code daily 3 years ago. Hallucinations happened sometimes and context window was smaller (so you had to do some funky workarounds to deal with larger codebases), but it was workable. There have been a lot of marked improvements from a code perspective then - a lot model related yes, but also a lot in the ease of use, interfaces, etc.
> Another thing I'd say is that you clearly have no clue what 'consumer hardware' means, or what consumers that can even get this stuff running locally would have to do to get it to even rival the frontier models in terms of their usability (most consumers are't going to just boot into Ubuntu and run this thing from a command line) flow, to say nothing of the hardware requirements.
You've moved the goalposts. My point was that the "danger" of no new open models being released isn't that high as the existing ones are already impressive. Their ease of use or daily driving isn't relevant to that. If there were a need, someone could wrap a clean interface and support around it, or run it as their own cloud solution.
You seem to be arguing something adjacent to my point, which is fine I guess but I have little to say. Also multiple of your comments have come across quite aggressive and rude. Just food for thought if you want to work on that or not.
Hardware. Frontier labs are driving up demand so much that it's priced significantly above cost making it far less affordable. Just look at Nvidia's profit margins.
Maybe. The use cases people primarily use LLMs for (documents, coding, design, research) existed decades ago with different tooling. Who knows if the future will have a slew of new problems that require new models or will continue to be similar?
Can you explain for the billions of the rest of us why this is the "most stressful time of the year" for the group you're referencing? I assume that's American students and/or teachers?
All these articles listing the American schools affected, "nationwide" outage reported, meanwhile hundreds of millions in the rest of the world affected.
> Important Legal Notice: This is a non-binding pledge of intent. No money is collected at this stage. All references to profit-sharing, dividends, voting rights, and ownership are proposed concepts only — not confirmed arrangements. Nothing on this site constitutes a securities offering, investment contract, or financial instrument of any kind. The final cooperative structure must be reviewed and approved by qualified securities and aviation counsel. Participation does not guarantee ownership, financial return, or membership in any final entity. This is a movement, not an investment product.
From skimming, I see at least 5 places where this is reiterated on the page.
I had the impression it was AI writing too because of the second half of the article. The first part looks genuine, the part since "trust laundering" smells fake: the scary single sentence followed by a whole paragraph of single clause sentences hints at AI.
Perhaps we've all just become paranoid, but even if it's not LLMs writing this, it now puts me off. And the AI image at the top of the page does not help with the feeling.
The line "This is the part that really matters." and the line "This is the circular citation pattern, and it’s one of the most under discussed attacks on the “retrieval augmented generation” trust model. " both raised flags. AI absolutely loves writing about the One Weird Trick that dentists don't want you to know. They love talking about "what really matters" or saying something is "the most under discussed" thing.
Then we get to the section "Why This Is A Bigger Deal Than It Looks". The title of this section again raises similar flags to before. But the bulleted list of:
1. The retrieval layer (immediately)
2. The model training corpus layer (months to years)
3. The agent layer (where the money is)
Absolutely reeks of AI. This list with this sequence of parentheticals is exactly how LLMs write, both structurally and the specific phrasing. This was the point where I felt confident enough to publicly accuse the post of AI writing.
I could go on with the prose in this section... How about "The attack surface is not hypothetical, it’s the default case."? Or "The cleanup problem for corpus poisoning is genuinely unsolved as of 2026."? (LLMs wildly overuse "genuine(ly)" and "real")
This paragraph under ‘Trust Laundering’ is when it hit my AI writing trigger threshold:
> This is the circular citation pattern, and it’s one of the most under discussed attacks on the “retrieval augmented generation” trust model. It doesn’t require compromising Wikipedia’s infrastructure with l33t hacker skills. It doesn’t require social engineering an editor. You just simply write the source yourself, cite yourself on Wikipedia, and let the trust flow downstream. Easy peasy!
“It doesn’t X. It doesn’t Y. You just Z. Conclusion”
Once I saw that some other elements stood out too.
There’s a set of bullet points under ‘Thae Approach’ where each bullet starts with a bolded phrase: “one domain”, “one press release”, “one Wikipedia edit”, followed by a bolded sentence “The whole thing took maybe about twenty minutes”.
The emphasis here on irrelevant quantifiable optimizations - who cares that it only needs one of each of three things and it took under twenty minutes? - with unnecessary faux-profundity is a strong AI tell.
Add to that that the writer talks in the article about using AI generation to produce the content for the poisoning site, the suggestion that he used it to write up a blog post about this is hardly an implausible suggestion.
If this truly set off zero flags for you then you're probably just not very attuned to LLM writing style. I've noticed that most people are not.
I posted a bunch of specifics in a reply to the GP since I was quite annoyed with being accused of "a lazy way of dismissing stuff". It's nothing of the sort. I am a very good reader and I have read a lot of LLM writing and a lot of human writing.
The author quite clearly outlines their reasoning for this in the article:
> Carrot Disclosure, dangling a metaphorical carrot in front of the vendor to incentivise change. The main idea is to only publish the (redacted) output of the exploit for a critical vulnerability, to showcase that the software is exploitable. Now the vendor has two choices: either perform a holistic audit of its software, fixing as many issues as possible in the hope of fixing the showcased vulnerability; or losing users who might not be happy running a known-vulnerable software. Users of this disclosure model are of course called Bugs Bunnies.
Seems like grandstanding bad faith to me. They didn't even bother to follow the established disclosure policy for this project because the author feels this quality of the code is so crap, so instead does this...
Maybe, but I can see why people don't want to deal with red tape to do someone a favour.
Once I tried to help an open source project with a bug and was rejected because I didn't agree to support the Ukraine, that all sexual orientations are equal, or whatever else the long winded contributor rules were.
The issue isn't that I don't support those things, it's more that it's like someone handing me a 3 page form to fill out for picking their keys up for them.
There also may be conventions on disclosure and exploits, but they're not based on the law or rules of society.
