Maelstrom

Coding, economics, politics and other stuff

[Maelstrom @strugglercss]

@brayozin_ Na vdd isso tbm existe em empresas privadas. Funcionários velhos defendendo sistemas legados com unhas e dentes.

Departamentos de TI que não deixam novas tecnologias avançarem pq "pode ter problema de segurança" mas o legado tá lá rodando com 1000 CVEs 🫠

[Maelstrom @strugglercss]

@kcosr Because

[Maelstrom @strugglercss]

@rauchg @Zai_org Indeed. There should be a "Deepseek moment" in markets right now, especially considering it was trained using Huawei GPUs.

But everyone seems to be in cognitive dissonance right now.

[Maelstrom @strugglercss]

@BenjaminDEKR Most people don't even have a desktop computer, only a laptop. You cannot fit 700 GB VRAM in a laptop. Not with current technology.

[Maelstrom @strugglercss]

@loktar00 Yes, for sure. People who did code before AI especially can see the value on it.

People who didn't might it still see it as just a chat interface.

[Maelstrom @strugglercss]

@spiritbuun As long as it can still do tool calling without breaking the syntax, I'll take it.

[Maelstrom @strugglercss]

@Grummz Chinese models are winning, with GLM 5.2 we got to the point where American models just cannot compete anymore, and it's all because of Anthropic.

[Maelstrom @strugglercss]

@NVIDIARTXSpark Gaming? Don't you mean "edge computing"? 😏

[Maelstrom @strugglercss]

@leolinshumor Tome seu IPCA + 8% investidor

[Maelstrom @strugglercss]

@rezoundous A little bit below if you consider frontend. Ahead if you consider terminal/scripting capabilities.

[Maelstrom @strugglercss]

@bridgebench Indeed. They have broken through, now trading blows with Anthropic itself. Their only bottleneck is inference compute now.

[Maelstrom @strugglercss]

@danielbergholz Me. It is the real deal. By anthropic billing has fallen 90% overnight and I am like 200% more productive cause can achieve Anthropic-level intelligence for much cheaper without being afraid of going homeless over an API bill.

[Maelstrom @strugglercss]

@ZixuanLi_ @ArtificialAnlys Some people doubted, you guys shipped it. Have been testing it and it completely replaces Claude usage except for file reading or image parsing.

Great model at a very much needed time. Congratulations z ai team!

[Maelstrom @strugglercss]

@ZaiforStartups You guys really shipped it this time.

This is the frontier model, only behind Opus 4.8. Tested it in practice, it's true. Not just benchmaxxing.

[Maelstrom @strugglercss]

@xeophon Still, using it in practice, it got stuck in loops that proprietary models just didn't.

They completely fixed it in GLM 5.2.

[Maelstrom @strugglercss]

@uwunetes @sawyerhood Now that's something only a very competent software engineer would know. Thanks for the info.

[Maelstrom @strugglercss]

@AIMelGibson @initjean People used to have that same pessimism in the crypto mining boom in the 2020s. Things will level themselves out sooner or later.

With these high prices China can spend billions in R&D and manufacturing to develop their chip industries and still be profitable.

[Maelstrom @strugglercss]

@Elettra251 @NielsRogge @DarioAmodei @Zai_org No regular person can run 😂 lmao

[Maelstrom @strugglercss]

@pupposandro This chart does not lie. It does feel like it really sits right between Opus 4.7 and 4.8.