Until now we've been using Sonnet 4 to power an editing agent in ApostropheCMS. Sonnet is a good price/quality/speed compromise, but sometimes when giving it a large set of instructions it would miss half of them. At least until we told it to go back and try again.
In my early tests tonight, Sonnet 5 is a LOT better out of the box. It's one-shotting complex instructions. It also recovered independently from bad instructions that led to an uninformative 400 error by using its schema-fetching tool to figure out there were was too much input.
If I have to gripe about something: it interpreted another impossible instruction by quietly discarding the input in question. But, the way it did it is... kinda exactly what anybody else would do, if they weren't in a position to change the implementation.
Flies though (50-70tps is impressive for a model this smart)
I went through roughly the same process to get it working on my M2 Macbook Pro... at awful speeds of course, since models like this one are mostly bound by memory bandwidth.
Yeah it is ugly. I'm not convinced binding future leadership with corporate docs is even possible. Unless the shares are controlled by other reputable organizations from the start? Maybe?
Everybody who understands the technical problems is proposing a government fix. There is another option, foundations / NGOs could do this. Of course, openai has shown how quickly that can pivot into something completely different.
Recently I got a PR that is (1) genuinely correct and useful, and (2) not from someone who uses the software in question at all. A happy outcome, but clearly the economics of open source contribution have changed.
Although I no way suspect this particular individual of anything untoward, of course it's always possible it could be part of one of the long-term goodwill-generation campaigns mentioned by the Ladybird team. Generating credibility by making seemingly difficult genuine contributions over a long period, then abusing that credibility. But in our particular project we're not in the habit of delegating approval authority, so I'm less concerned about that.
I wonder about this too. The other objections miss the point: if it's faster, and otherwise the same, and doesn't require different hardware, then why not just announce that the standard tier of MiMo-v.25-Pro is now ridiculously fast and raise the price? What does "limited high speed resources" mean if it runs on the same hardware as the rest of their pool?
I think the answer is that there's a tradeoff here where additional throughput for a single person can be achieved only by tying up more resources than a normal request would, even when you take into account the fact that the normal request takes longer to finish. I'm not an expert, but some of the optimizations they describe, particularly the parallel prediction stuff, sound like they could take up extra resources.
Isn't that nitpicking? It's a smaller representation of the data, if you have a certain appetite for decompression time. It could conceivably be worth it. I think it would make a great level 2 cache for older chats.
In 1994 I was 2 years out of school. I'd written one windows shareware application and a whole lot of unix-y things. People were excited about the internet but most people didn't have access. Unix shell accounts via dialup were common though.
One day I was encouraged to write a Windows Sockets emulation layer for ordinary dial-up shell accounts like those offered by netcom. The idea was to allow the use of the recently released Mosaic browser without an actual internet connection. I figured sure, no problem. I'll use curl or some other tool in the shell account to do the actual fetching of URLs, transfer styles over zmodem, and simulate all the tcp/ip calls in the DLL.
I couldn't even get started. The reason is that I couldn't understand how the different Windows applications could all share memory allocated at runtime in the winsock.dll.
I asked a highly experienced ex Microsoft person, and he just said what are you talking about. There's no API to allocate shared memory.
So I gave up. 6 months later someone else did it.
Around then I realized the truth: Windows 3.1 had no memory protection at all. Specifically all global variables in DLLs were shared by default. The hard part wasn't sharing memory among users of a DLL. If anything, the hard part was having good discipline to avoid sharing it.
Since I'd only used multiuser Unix in school, and I knew Windows supported multitasking (even if only the cooperative kind), I just couldn't wrap my head around the idea that I'm multitasking operating system could exist without memory protection.
I believe there was a science fiction story all the way back in the early '80s describing a scenario where physics gets reclassified as a soft science or an art form because it is no longer feasible to prove anything.
You're right that it's not literally frontier. But like recent Qwen releases, it is a lot more capable than anybody thought models of this size could be a year ago, like capable enough to set a ceiling on what you can charge for AI for certain applications. Others still clearly justify a stronger model, but this trend may continue, etc.
Also me: just a month ago I bought a Bluetooth keyboard and mouse for my phone because they are completely sufficient for the work emergency use case along with the termux app
The thing about a diagonal measurement is it doesn't tell you if it's going to fit on a shitty airline tray table or not. Some laptops with a larger diagonal measurement are not too deep. Others are way too deep.
(No, no, I take this stuff seriously too, but it had to be said)