Why would they not release those models, since it would help them become profitable and beat the competition becoming the one company rule them all? Oh yeah, they are so dangerous. Well Anthropic revenue comes from only few big corporate customers, so just make them Golden Partners etc. and give them access to super-models. I doubt Google and Meta wants to crack web security.
Besides, if we start imagining super models US companies have in secret, why would we expect Chinese companies have them? Why would not Moonshot have their own super intelligence hidden in the cellar?
Vertical integration. They want to control all sectors, that's why OpenAI and Anthropic are slowly rolling out solutions for law, medicine, design and so on.
I think for few years MS has tried to be a bit more developer friendly. The Windows Terminal (and no, not what opens running Command Line or Powershell, it is a different application: https://en.wikipedia.org/wiki/Windows_Terminal) is kinda good actually. It’s not Alacritty, but a lot better than any previous terminal emulator. I have also poked around their developer tools and they do seem to make using windows a little less hellish experience: https://learn.microsoft.com/en-gb/windows/dev-environment/
And PS is pretty good, although I haven’t used used PS7 much (if it is a completely new application, why does it share a version number? And why does it not come as the default shell with Windows? Why do they have to make everything so confusing?). It is probably nicer for extended scripting than bash, but honestly I rather use Python for that. For windows admin stuff it is great, but mostly I work with network stuff and it is way too verbose to my taste for simple commands. ping is a lot easier to remember than Test-Connection. I reach for cmd commands like netsh int set ip rather than tab through the verbose syntax of PS for simple network commands. I know you can set up aliases and bash aliases like ls work, but I still feel it os pretty impractical for quick terminal work.
Using it on linux feels so blashemous that I will not entertain such vile and fevilish thoughts. I use linux to get away from the sinful awful things I must do for my livelihood, I tather not use Win applications on my own system unless I absolutely need to. For any serious scripting on Mac… Maybe just don’t?
I stopped using Windows Terminal altogether because it somehow tends to crash the processes running inside its tabs. And I have a lot of terminals open.
An example issue: https://github.com/microsoft/terminal/issues/16363 . This exact error message with different codes happened to me across many machines at many different times. Never seen anything like that with the classic console host.
The title buries the lead: rabbits executing hunters in the Smithfield Decretal illustrations. Seems British humour has a tradition at least to the 14th Century.
Seems Merram-Webster dates the ”lede” coming to dictionaries around 2008, so I would still contest lead being at least as acceptable - and makes more sense imho. English is full of strange spellings which make no sense, so reader is expected to know that we are not talking about Pb, but slinging old newshack lingo. It makes more sense if you know this saying, attributed to William Hearst: ”if it bleeds it leads”, which means yes, sensationalism and gore sells, but as more universal writerly advice that you sould not waste time and start with the most important thing you have to say (ever read those Atlantic/New Yorker -type articles that ramble on for ten paragraphs before you even know what it is about).
Rabbit giving justice to hunters being the lead(ing thought, ie. most important idea in the piece that should go first) is of course me being sarco.
How is that an example of burying the lead? (I presume lede is the correct spelling, but burying lead for others to find in their drinking water amuses me.)
EDIT: saw the picture and yes. That certainly was catchy, but I the hn title is missing the subtitle: "until the black death made them commonplace".
Though lede is just a variation of the word lead to get a unique word for the leading section of an article. As long as we all agree that it's pronounced like in leading and not like the metal the difference in spelling doesn't feel significant
If there were open source Dante that actually worked it would be very cool. Don’t know how much usage it would get since the hardware will always lock it in. But for small studios and research it might be cool. Interesting project nevertheless.
There is AES67 support in PipeWire as of 1-2 years ago. I've tried it with some Dante hardware that I own, running the hardware in AES67 mode. It didn't work very well for me, even after lots of fiddling. If the experience would have been better, I probably would not have started working on ETH-68. It might be better now, I hope that it is.
ETH-68 doesn't have nearly as broad of scope as AES67. It is a much simpler system, so easier to realize for a 1-person team working on weekends.
Just a few other comments:
1. AES67 usually means Dante hardware which is notoriously expensive.
2. Dante devices running in AES67 mode often (always?) have their capabilities reduced. At least in some cases you are limited to 48 kHz, and I believe you are also limited on latency compensation settings. Someone with more experience could chime in.
> These open models still did not beat February's Mythos / Fable 5.
On what task? By who? On what benchmark? How do you measure in you own workflow the “betterness” or “more goodness” of these or any models? If you don’t say those things you’re just writing a bad ad copy.
> It's plausible that open models are 6 - 12 months behind, and there is no "good enough".
Anecdotally, a lot of people - including myself - seem to really notice much difference between the model now or six months ago. So there really seems to be good enough. It depends on the task you use them for and how you measure the output. For most tasks you really do not need frontier capability. Also how do we know how much of these “big improvements” come from the harness and tooling rather than the raw capability of the model?
Did you see any person claiming an open model was more intelligent than Fable 5?
It's always some sort of "I don't notice the difference".
And honestly, if you don't see a difference between the SOTA from 6 months ago, which would be GPT 5.4, and today's Opus 5.5, you would have to be downright blind. Not sure what else to say - the results are obviously different for any kind of meaningful output.
> Also how do we know how much of these “big improvements” come from the harness and tooling rather than the raw capability of the model?
By simply running the old models in the latest harness. Which none of the people who argue "it's all the harness" ever do.
What kind of evidence would satisfy you. Referring to benchmarks is apparently not sufficient because "you won't notice the difference in your everyday tasks", but saying "I do notice the difference in my everyday tasks" is just vibes.
I have 500+ hours of experience building native mobile apps with AI between GPT 5.4 half a year ago and today, in addition to my regular software engineering job.
The difference between GPT 5.4 and Opus 5.5 is obvious.
What do you do, where apparently you cannot see a difference?
I honestly can't imagine, unless it's like sorting your emails.
I struggle to see how a software engineer fails to see that anecdotal evidence does not matter here at all. I can say I built seven fully functioning operating systems last month, or that I am the fastest runner in the world or that my daddy is the strongest man in the world. None of it matters without data. I can say it is warmer in the living room and you can say that no, it is much warmer in the bedroon, without having something definite to measure and something accurate to measure it with, the whole discussion is pointless - just vibes. You haven't even said how you have anecdotally experienced the difference between the older or the open weight models. So excuse me, but then you will fail to convince people on your claims, especially when the whole discussion seems to be in the middle of some bloody information warfare at the moment.
Anecdotally, a lot of people - including myself - seem to really notice much difference between the model now or six months ago. So there really seems to be good enough.
I have similar experience than you, with a big caveat, opus 5.5 would have still broken badly a custom ERP.
And on that specific kind of software, ultimately a big CRUD, there really isn't that much of a difference between GLM5.3 and Opus/OpenAI.
You see the differences when you get to different class of software.
I also have data entry applications that use LLM to actually parse documents, it's all Chinese models self hosted because the economic calculus beated a hosted API by about 5x
For mobile apps, I find that nowadays with Opus 5.5 the UI looks better, the UX is better, it can implement more tricky animations and gestures, and it can do all of that with far fewer iterations and feedback than eg. GPT 5.5 would have required.
Also vision capabilities were improved significantly with GPT 6 Astra or Opus 5.5, even compared to GPT 5.6 Sol.
There was no way the old models such as GPT 5.4 would have done a comparable job when asked to align an implementation to a visual reference.
Even for basic websites with no interactive functionality, this should make a significant difference.
I even find that for what amounts to a web search Gemini 3.8 flash is better than the big models, stuff like the links to updated tax codes in different countries, or whatever MS is doing with Azure/o365 or AWS with their plethora of products.
I think use cases are the real reason why people have such different experiences, I too find that Opus5.5/Astra/6 are better for UI/UX now, it wasn't the case a year ago, at some point Gemini pro 3.5 was the best one at that.
That's also why I use all of them and try to not be locked to a single harness as well.
I tried it a bit and I like it! It is very fast via openrouter (significantly better than Kimi K3) on webui. Very verbose and starts to forget instructions after awhile it seems, but it gave me quite a lot of good info during a half an hour chat on C and embedded programming. I think I will keep urimg this as my main assistant for few weeks.
Antti is a wonderful teacher, he is one of my lecturers. His courses are pretty hard, since they go into very very bread and bones of CS (algorithms, memory allocation, etc), but I always feel I 10x’d my understanding after them.
I don’t think he needs or wants any more money. These guys already have so much moolah that it is virtually meaningless to them. They want power. Remember when Zuck tried to run for president?
It is described in Sarah Wynn-Williams’ book. According to it, Zuck even battled to change the stock structure in order to take 2-year hiatus from CEO duties:
…without losing control of the company if his absence is “in connection with his serving in a government position or office.” I assume the two years would cover the period Mark would be taking on the—presumably time-consuming—endeavor of campaigning to be president of the United States.
Especially in a world that's been programmed to believe wholeheartedly that money is literally the most important thing in the entire universe, and these people who worship it as a god will never be satisfied with any amount of money; even when the last human alive owns all of it, they'll still want more. It's a form of mass-insanity. The Cult of Money.
I think the AI provider whose model does “bad things” should be held responsible of such things, going to a “bad place” - such as a prison - if the “bad things” are bad enough. Let’s see how that affects your IPO.
America needs a corporate death penalty. Not for people, but for organizations. The people themselves certainly should be prevented from being an executive anywhere else, but yes, corporations should be allowed to be killed.
Maybe it is for better. I used Orion as my main browser for awhile on iOS, but then it started getting weird bugs. Now I have both Safari and Firefox open all the time and watch yt on Brave.
I try to see if they got Orion to behave again, but find a bit difficult to trust it as the main browser.
Besides, if we start imagining super models US companies have in secret, why would we expect Chinese companies have them? Why would not Moonshot have their own super intelligence hidden in the cellar?
reply