Hacker News: jeudesprits

GLM-5.3 is now open-weight

jeudesprits — Fri, 28 Aug 2026 15:20:13 +0000

https://twitter.com/Zai_org/status/2093354097122455713

https://z.ai/blog/glm-5.3

Comments URL: https://news.ycombinator.com/item?id=49479878

Points: 660

# Comments: 222

Muse Image and Muse Video, the first media generation models developed by Meta

jeudesprits — Tue, 07 Jul 2026 23:24:15 +0000

Article URL: https://twitter.com/AIatMeta/status/2074577662840832382

Comments URL: https://news.ycombinator.com/item?id=48825392

Points: 4

# Comments: 0

How Apple Replaces Frameworks [video]

jeudesprits — Mon, 02 Feb 2026 12:53:23 +0000

Article URL: https://www.youtube.com/watch?v=SuQGQ1vh9k0

Comments URL: https://news.ycombinator.com/item?id=46855459

Points: 1

# Comments: 0

Kimi K2.5 1T runs on 2 M3 Ultras with mlx-lm in it's native precision

jeudesprits — Tue, 27 Jan 2026 21:35:18 +0000

Article URL: https://twitter.com/awnihannun/status/2016221496084205965

Comments URL: https://news.ycombinator.com/item?id=46787243

Points: 5

# Comments: 0

Kimi-K2-Thinking 1T at 28 tok/SEC with MLX tensor parallel

jeudesprits — Fri, 19 Dec 2025 01:45:54 +0000

Article URL: https://twitter.com/exolabs/status/2001817749744476256

Comments URL: https://news.ycombinator.com/item?id=46321370

Points: 5

# Comments: 1

Kimi K2 1T model runs on 2 512GB M3 Ultras

jeudesprits — Sun, 14 Dec 2025 13:04:34 +0000

Article URL: https://twitter.com/awnihannun/status/1943723599971443134

Comments URL: https://news.ycombinator.com/item?id=46262734

Points: 234

# Comments: 121

Confidential Compute Open Network

jeudesprits — Sun, 30 Nov 2025 18:27:31 +0000

Article URL: https://cocoon.org

Comments URL: https://news.ycombinator.com/item?id=46099059

Points: 12

# Comments: 1

New comment by jeudesprits in "Welcome to Cocoon – The Confidential Compute Open Network"

jeudesprits — Wed, 29 Oct 2025 13:08:11 +0000

Welcome to Cocoon — the Confidential Compute Open Network

Cocoon is a decentralized network for executing AI inference securely and privately.

In this network, app developers reward GPU owners with TON for processing inference requests.

Telegram will be the first major customer to use Cocoon for confidential AI queries — and will invest heavily in promoting the network across its global ecosystem.

App developers who want to run inference through Cocoon are invited to contact us via DMs to this channel.

Please specify which model architecture you plan to use (e.g., DeepSeek, Qwen), along with your expected daily query volume and average input/output token size.

GPU owners who want to earn TON by contributing compute power can also message this channel using the button below.

Please indicate how many GPUs you can provide and include details such as type (e.g., H200), VRAM, and expected uptime.

Cocoon is ready — launching in November, once we’ve gathered your applications.