MostlyStable

300Gbps? Is that typo? Unless you are connecting to some very particular infrastructure on the other one, nothing you could possibly connect to could use it, and you would need gear that would be somewhat high end even for server grade.

(I know you said you didn't select that option, but just the idea that it's even offered to residential units is mind blowing).

I've always had a soft spot for the plan of repopulating the Great Plains with modern versions of all the extinct megafauna. We used to have camels and cheetahs and lions and more. We could import half of the African large mammals and they would have had near species analogs in the relatively recent past, geologically speaking.

I was going to post a comment on a related topic (I couldn't find in the announcement if this is English only or not), but would you mind expanding? I was thinking about doing something very similar, and yeah, if the model isn't hearing the mistakes I'm making, that would dramatically decrease it's usefulness.

I'm somewhat skeptical that the cost difference would really be that high, but honestly...yeah I probably would. If what I was getting was fully physical (not capacitive) media and AC controls, including pause/play, skip/tune, volume control, temp control, fan speed control, zone selection, etc? I interact with those systems multiple times every single time I get into a vehicle, which is essentially multiple times every day, for years. Improving the quality of those interactions even a little bit (and in my opinion the difference between good physical buttons and even a very good touchscreen, let alone a shitty one, is massive), is worth thousands of dollars.

While I was not presented with the option on a given model to go with buttons or touchscreens, when I was shopping for cars, I did eliminate models based on their interface options. The models I was willing to consider were probably cut in half because I wouldn't get anything that was entirely touch screen or capacitive for both AC and media.

I don't have the mathematical chops to really analyze this on my own. Is this a bigger deal than the fact that real world markets already violate all the theoretical assumptions (e.g. unimpeded access to new entrants, perfect information, etc.etc), and so, in practice are never perfectly competitive or efficient?

While I think you are accurately describing how people do/would react, the "big deal" you describe killed, injured, or caused adverse health effects for exactly zero people. It is possible that these are inevitable outcomes of human psychology, but a more rational world would have gone full steam ahead on nuclear power, even after all of the events you describe. A Chernobyl level accident every single year would have killed fewer people (by a few OOM) than particulate emissions from coal, and that's completely ignoring any climate effects.

Our societies risk tolerance with nuclear is literally orders of magnitude disconnected from how we treat risk from any other source, and as a result we are all poorer, less healthy, and have injured the environment to a dramatically greater degree relative to a pro-nuclear alternative timeline.

I was talking about reporting on testing and capabilities. Yes, open models provide a greater amount of information about the development of the model and how to run it yourself, but I am quite confident that literally no AI company, open or closed, conducts and reports so thoroughly on testing about the capabilities of their models.

While I'm still not sure I would characterize that as bragging, you're right that that is a fair interpretation. However, another Fair interpretation of that is something along the lines of "the downside or cost of this positive thing is this following negative thing."

I find it interesting how two different directly opposed messages seem to have both been interpreted as being nothing but marketing speak.

Why do you think they are bragging? Anthropic has long been the company to give us by far the most in-depth information about their models, both positive and negative. I read this as them just stating a fact about this model that users would want to know.

The fact that you need top talent also suggests that it is indeed that hard

The gap between grok and Gemini to Claude and chatgpt suggests that yes it is that hard.

And even if they didn't, they have a track record. Even if we did have benchmarks in this case I would still wait until people got there hands on it and formed a more holistic opinion.

Wouldn't that be a stork?

"Affordable" depends on what you need. When a task is able to be achieved by two different calibers of model, it's obviously more cost effective to use the less capable model, in the same way that you wouldn't hire a math PhD to do simple addition.

If what you need is only possible with the more capable model then the "affordability" of the less capable model is sort of irrelevant. If what you need is a novel mathematical proof, it doesn't matter that a high school student is "more affodable". You need the math PhD.

As "old" models get more and more capable, it's going to be an increasingly important skill to be able to adequately recognize when a task requires a frontier model and when it doesn't, so that the less capable (and therefore cheaper) model can be used.

20 years ago, that argument would make sense. They had no competition and could do what they wanted. As an earlier comment stated: that is starting to change, and if they wait until open competitors are fully established, then it will be too late. Now is the time for them to realize that their parasitic business model is coming to an end and they need to change if they want to survive long term.

They can of course choose short term profits over long term viability, which wouldn't be all that surprising, but that changes the explanation from "more profits" to "short-sightedness/incompetence"

As others have commented, this is not a Framework product. That's part of the beauty: they open source there designs, allowing for third parties to easily make things like this (and much more beside). I believe at some point someone in the community was trying to design one of those slim ethernet ports that expands open when you need it (the jack doesn't really fit). Apparently some of the mechanisms for doing so are still proprietary though.

-edit- here it is: https://community.frame.work/t/low-profile-ethernet-expansio...

Exactly. I made a small package for my lab and just put it on GitHub. My guess is that many academics don't know that you can install packages from other sources than cran

How long have you been testing this? Have you noted a large improvement? I tested Opus for this quite a while ago (maybe 4.5? Whatever was out about a year ago), and it performed quite poorly on my use case.

Does anyone know of OCR benchmarks that include hand-written documents? I'm currently using Gemini pro 3 for this, and error rates are quite good, but it's a little bit pricey, and I'd be interested in a cheaper model that could perform as well, but almost all the OCR benchmarks I'm aware of (and I believe all the ones included in this announcement) are about printed/typeset text.

The cost is a choice. Not inherent to the power. We could choose literally any day to make it dramatically cheaper.

I'm ok with a version of this as a concept. The version being when a feature is technically present in every SKU, but requires an extra purchase to unlock. A reply to you specifically mentioned subscriptions, which I very much do not like (except in cases where the feature requires ongoing costs), but there are many cases where having every version contain the feature, but requiring a purchase to unlock it is pro-consumer, and is a win-win-win (or at wist a win-win-draw). It can, under the right circumstances, allow the product to be available for a cheaper cost than it would otherwise be. So people who are willing to pay for it are better off, people who aren't willing to pay for maintain the option to change their mind for a nominal fee, and the company probably makes slightly more profit.

All that being said, in my opinion, it needs to come with several features:

1. no subscriptions for something that is just a one time unlock 2. It needs to be legal and protected for customers to figure out how to unlock it on their own without purchasing the unlock.

I haven't thought enough to have a strong opinion on the exact situation you describe (where it is present and an unlock isn't available at any fee), other than to say I'd still argue strongly that customers figuring out how to unlock it on their own should be legally protected.

The person I was replying to specifically said that the Claude will "encode more knowledge" and that their problem was that they didn't like talking to Claude. It sounds like they think that Claude is at least slightly more functional. And the "not liking talking to it" is probably fixable. Someone for whom a local model works, and for whom the economics make sense, should absolutely run a local model and I wouldn't try to convince them otherwise. I'm sure it's the right choice for a lot of people. But not liking the personality of Claude is probably not a great reason on its own, given the minuscule amount of effort it takes to fix.

Curious if you have tried custom instructions. I was never quite as unhappy with Claude's voice as you appear to be, but there were several things I didn't like. A custom prompt fixed almost all of them.

I'm out of my technical depth here, but out of curiosity: is this meant to be a full replacement for the current IP address paradigm, or is this meant to be a specific tool on top of/alongside IP addresses that solves particular problems/frictions?

Even if some models are able to find some large fraction of the vulns that Mythos finds, the contention (that those of us outside Anthropic and a few select partners do not yet have the ability to replicate) is that Mythos not only found vulns, but was able to string them together into working exploits, very close to autonomously or at list with minimal hand-holding. I believe that UK AISI has at least in part confirmed that Mythos is better at this than other models like GPT 5.5, even while it's not better at the simply finding vulns part (although, again, no one else is currently able to replicate their results).

This was the primary reason for not releasing it. The difference in the two primary camps around this topic are that the doubter group thinks that Anthropic, and all of their partners, are essentially lying about this (since no one outside Anthropic and select partners has the access to replicate), whereas the other side believes that Anthropic and partners are probably mostly telling the truth without too much exaggeration.

Neither camp has evidence other than unconfirmable reports and/or arguments about economic incentives. I personally think that Anthropic has, in the past, mostly not lied about things like this and has by far been the most transparent and open AI company. That could change, and they could be lying now, but I think that the camp that is certain that they are is far, far too confident in their belief.

Mostly a principled (although admittedly near totally useless) stand against exactly the kind of issue talked about in this article. The organic label, as it is used in the US, is basically BS marketing bordering on misinformation. I prefer that my one, tiny, consumer signal goes in the direction of not supporting that BS marketing label.

You do you. Me personally? I never buy organic if I can avoid it, to the point of, in the very very rare cases when organic is cheaper, I have still sometimes bought conventional.

aren't there other issues with having the motor in the wheels? Unsprung mass, plus the wheels can get pretty banged around?

Unfortunately, Anthropic and Claude models have joined the ranks of Mind-killer topics where the signal to noise ratio in any discussion has dropped through the basement.