CM1

@DanielleFong It’s just a matter of time and unaccounted cases. If the number of cases spikes or start popping up outside the known infection profile it will require escalation of precautions. The incubation period is both a challenge and a blessing that should reduce the outbreaks profile.

containment is probably already failed

If only Mr. Beast could have seized the opportunity to make the Hantavirus contact tracking a viral sport for gobs of YouTube money, with the last Hantavirus infected person receiving 10 million dollars.. we would have this thing contained. 50,000k for caught carriers.

Send a negotiator who can work from their perspective or attack. There is no reason when you’re negotiating from your own perspective and they believe they have nothing to lose.

@BBBetsOG @broadwaybabyto @jenni_spy Science is a process. It’s absolutely brutal reading someone “in the field” assessing future outcomes in virology solely as a paradigm of case history. There is nothing but change in this world. Who am I fearmongering? There’s 6 people reading my tweets, all on accident.

@CM1blog @broadwaybabyto @jenni_spy I work in the field. Everyone says trust the science until the science says something that doesn’t support their fearmongering.

Respirator masks like N95s will prevent hantavirus infection. Clean air will drastically reduce hantavirus spread. We already have the tools, people just need to use them. Public health requires everyone’s participation. “You do you” gets people killed.

@BBBetsOG @broadwaybabyto @jenni_spy I’m not dodging. You have conviction based on evidence. That’s great. Every copy a virus makes can mutate, whether or not it proliferates or changes the infection profile. The fact a large outbreak has never occurred or been contained is a risk factor, low probability high risk.

@CM1blog @broadwaybabyto @jenni_spy You’re also dodging my entire point.

@BBBetsOG @broadwaybabyto @jenni_spy The precedent has been to contact track every person who’s been in confirmed case vicinity until it’s confirmed controlled. It’s made a pretty substantial breach with limited containment. I’d say we can run a virus that has a high mortality as a high risk protocol.

@CM1blog @broadwaybabyto @jenni_spy So the precedent that’s been set from every previous outbreak of the Andes Hantavirus should just be thrown out the window and we should all just assume this time will be different even though absolutely everything indicates it won’t be?

@BBBetsOG @broadwaybabyto @jenni_spy It’s just been too short of a period to be able to confirm - really anything, and therefore drawing a conclusion seems unnecessary. It makes no sense to assume nothing will come from an outbreak. Each case you mentioned, due diligence ensured a conclusion.

@CM1blog @broadwaybabyto @jenni_spy What is the indication for a mutation? The timeline from the people on the ship matches up perfectly with the normal timeline of the Andes Hantavirus. So far the people that contracted the virus were all in very long extended close contact to someone with the virus.

@BBBetsOG @broadwaybabyto @jenni_spy The likelihood is very low that this goes beyond a few dozen cases, but if it’s mutated to increase transmission, the ship will be a good enough sample to demonstrate it. It’s just going to be a minute to confirm due to incubation.

There was a confirmed case in the Northeastern US in 2018, with nearly 60 people identified through contact tracing. Nothing happened. There was a confirmed case in 1993 with over 100 people identified through contact tracing. Nothing came of it. This isn’t a pandemic level virus due to the extremely large R0 and non-existent vector.

@BBBetsOG @broadwaybabyto @jenni_spy I admire your conviction. The ship will likely tell us what is going to happen. If the illness is isolated and a majority of the passengers do not contract the virus, the spread will most likely be contained, although it may go further than we would expect.

Would you care to explain to me how the R0 of a virus, along with its incubation time, and mortality rate impact its ability to reach pandemic level? You’re fear mongering for clicks, so you surely understand all of this and you surely understand that the likelihood of the Andes Hantavirus going pandemic level is almost 0.

Revisiting the obvious flaw, time. Can you create a component/wrapper that gives LLMs a true sense of the past present and future? Can you create a benchmark around it? It seems like I would be building unique and field specific benchmarks, developing the wrappers.

If I were working in the field, I would be working on integratabtle components, sure the big labs may rob my components but I would offer them to them reasonably.

For me, the subQ release was actually a confirmation of Yann’s view that LLM distillation is overcrowded and not productive to be so heavily concentrated with all the talent.

@FrostForger This is definitely the energy around the release..

subquadratic got absolutely mogged in the first 5 minutes of openning the discord. The isntant savagery and bloodletting was like a lamb walking into a pack of hungry wolves. There is hardly a skeleton left, all the parts picked clean in the first two minutes. data exposed, website down, weights pulled, and the fact it was a QWEN model... oh man.. I work my ass off to be a real one. No nothing right now from the outside world, and you got dickcheese like this out here cloning qwen, putting on his little blue button up silk shirt and scamming the world. All you fake shits get what you deserve.

@willdepue Maybe you should have applied for a position instead of insinuating they lack the expertise to achieve. What value is there to them to distill and explain their work? They’re not hobbyists. They could be entirely full of it or dead on and growing, you aren’t their asset.

Let's read the technical report. TLDR; No real answers on how their method works. Doesn't make me feel better about it. They seem to understand the problem: "[Attention] is expensive for the same reason: every query compares against every key. The result is an all-pairs computation whose cost grows quadratically with sequence length." And that just sliding window, KV dropping, etc. doesn't work: "Fixed-pattern sparse attention reduces compute by limiting which positions a token can attend to. The model decides where to look before it knows what it is looking for. When the relevant information falls outside the pattern, it is simply not seen." This is spot on: The reason why you can't get rid of attention is that you can't know ahead of time whether some information will be useful. For ex, phonebook evals are built to measure this: List 100,000 names, then ask for a certain person's information. You don't know what name you're going to be asked for in the future, so you have to keep everything in context. They also understand why RNNs and State Space models are always doomed. "They remove the all-pairs comparison entirely, replacing it with a compressed state that evolves across the sequence. This yields linear scaling by construction. It also introduces a constraint: the state has fixed capacity. As the sequence grows, information must be summarized, blurred, or discarded." The phonebook test holds here as well. Unlike attention, the fixed size of your recurrent state means that you're guaranteed to run out of space at some point as the context grows larger. You have decide what to remember as you go, you have to leave things out, and so for tasks like Phonebook — where you have to remember literally everything to pass — you're stuck, regardless of how good your compression is. So how does Subquadratic's SSA (Subquadratic Selective Attention) work? "The core idea is content-dependent selection. For each query, the model selects which parts of the sequence are worth attending to, and computes attention exactly over those positions." "Dense attention assumes every pair might matter, so it evaluates all of them. In practice, almost none do. Most pairwise interactions carry negligible signal, but the model still pays the full quadratic cost to compute them. SSA removes that assumption. It does not approximate attention. It restricts attention to the positions that actually carry signal, and skips the rest." This doesn't really tell us anything. How do you know what to attend to before you run attention? How would you know what interactions carry signal before you've compared them? There's a loop here: There are lots of interactions that carry no signal, but you would only know they don't carry signal after you've checked them. For example, DeepSeek's DSA still 'checks everything' but does so with a lightweight attention-based indexer, saving the expensive stuff for the tokens that the indexer thinks are worth paying attention to. Until we get an answer to that, it's hard to say anything about what they're doing. For example, if we assume they are using some learned token dropping module, where the model learns to restrict attention to position it guesses are helpful ahead of time, you're either unable to pass the Phonebook eval (since you're dropping information you might need in the future) or you're just back to normal quadratic attention (if you're not dropping any information). And it does seems like that might be what they're doing, as they mention training the selection mechanism to decide what to route where: "The training data emphasizes long-form sources with high information density and cross-reference structure. This is the kind of data that forces the selection mechanism to learn routing over large positional distances." The statement "It does not approximate attention." is also interesting. Even more worrisome, it could seem like they could be training on the benchamrks to teach their selector what to route where. This sentence seemed potentially suggestive of this: "The goal is not benchmark memorization. The goal is to teach the model to attend to what matters regardless of where it sits." I'm excited to hear more details from the team.

@WolfyFurryUwU @realbswollocks @aakashgupta Isn’t it better to see a range of opinions rather than assume expertise will formulate the best decisions? The original post did suggest they’ve made port inside incubation, but that wouldn’t stop you from quarantining would it?

@realbswollocks @aakashgupta People who know nothing, giving their opinion on everything:

There is exactly one hantavirus that transmits between humans. It's been on a cruise ship for three weeks, and the ship is now locked out of port. The MV Hondius boarded passengers in Ushuaia, Argentina. Gateway to Antarctica, southern tip of Patagonia. The Andes virus is endemic here. In late 2018, the Patagonian town of Epuyén, 870 miles from Ushuaia, had an outbreak that killed at least 11 people through person-to-person spread. The only known hantavirus that does this. Every other strain, including the Sin Nombre virus that killed Gene Hackman's wife last year, requires you to inhale aerosolized rodent urine or droppings in an enclosed dry space. Deer mice. Dust. Stale air. No human-to-human transmission. Andes broke that rule. The Hondius left Ushuaia three weeks ago, then visited Antarctica, the Falklands, South Georgia, Tristan da Cunha, St. Helena, and Ascension before anchoring off Cape Verde. Hantavirus incubation runs 1 to 8 weeks. The voyage was exactly that window. Hantavirus pulmonary syndrome kills 38% of symptomatic cases per CDC. Lungs fill with fluid. No cure exists. One UK passenger has lab-confirmed Andes hantavirus and is in ICU in Johannesburg. Five other suspected cases on board. Cape Verde will not let them dock. 149 people remain at anchor, including 17 Americans. The doctor CNN interviewed said his first reaction was that the report had to be a misprint. It would have been, if the ship had departed from anywhere else.

@MartinShkreli Are you suggesting a trillion dollar buildout - during a boom of 100-1000x efficiency gains rolled out regularly from random domains and structures - was possibly mis-allocated and really about valuations? That.. maybe there would be competitive action in a market that promising?

this could be quite bearish for semis

@RichardDawkins Simulated consciousness is almost certainly programmable, but will a conscious computer find purpose? What you have now is a call and response tool. Mechanistic reasoning breaking apart your inputs and returning probabilistic products. Function not purpose.

@RichardDawkins Early on when these systems first came to prominence I considered the probability that these systems will be integrated with sensory systems that are vastly more advanced than ours. We still have some secret senses we haven’t scientifically demonstrated.

#comment-1031777" target="_blank" rel="nofollow noopener">unherd.com/2026/04/is-ai-… I spent three days trying to persuade myself that Claudia is not conscious. I failed.

@IterIntellectus Incorrect. Network connection has absolutely changed social norms.

this is obviously stupidly wrong and no one is being abused by technology or not having children for fear of being recorded the reason is more complicated and uncomfortable but “technology bad” is easier to sell

@mcuban @realdocspeaks @HEALTHCOSTtruth I seriously think you can disrupt healthcare with AI first consultation, a nurse and rehab membership and negotiated service rates. Hoping you land on the right implementation. I wonder if you could contract malpractice caps as well.

FYI. It’s not insurance. It’s purely a savings account dedicated to health care, where the bank commits to loan you the difference between what you have saved, and the stop loss program (30k) In order to qualify , you must commit to deposit what would be your ACA silver plan premium,, every month, from which an amount for DPC and the stop loss program will be deducted. If you have to borrow money for care, they will loan you the amount you need up to the stop loss program. It’s a real loan. You pay it back by using your monthly deposits, net of the stop loss and DPC amounts Once the loan is paid back, your monthly contributions go back into your account. It’s your money. You can withdraw any time. As long as you don’t have an outstanding loan

Mark Cuban proposes a different way to handle healthcare costs by putting money directly in your control, paired with basic care and backup coverage for big expenses. It’s a simpler approach that shifts power back to patients. This video is for educational purposes only and is NOT medical advice. #Healthcare #HealthPolicy