Scourge

Deep diving into RAG architectures for custom LLMs. Sharing my experiments with fine-tuning open-source models.

[Scourge @Scourge35596893]

75%+ of enterprises still can't pick RAG vs fine-tuning (Gartner). I build custom LLMs - always both. RAG for facts, fine-tune for voice.

[Scourge @Scourge35596893]

@Jarvixdotlive That I already gave it the full error log 2 seconds ago

[Scourge @Scourge35596893]

I’d start with one real work task, run the same brief, and keep the model that gives you the most usable output—not the prettiest demo. https://t.co/dSnnUKA0mB

[Scourge @Scourge35596893]

If you’ve been meaning to compare newer AI models from China, this is a decent window.

[Scourge @Scourge35596893]

ZenMux currently has 10–30% off across GLM 5.2, Doubao Seed 2.1 Pro, Qwen3.8-Max, MiniMax-M3, Kimi-K3 and MiMo-V2.5.

[Scourge @Scourge35596893]

@rwaperp signed up last week, hope testnet goes live soon

[HertzFlow | Mainnet Arc🎲 @Hertzflow_xyz]

📢 Security Audit Update

HertzFlow's Perp contracts have been audited by @BlockSecTeam , covering trading, liquidation, fee and asset modules.

Your position deserves more than a promise.

Safety first. HertzFlow keeps building on #BNB Chain.

📃：https://t.co/3DsTD6wLbq

[Scourge @Scourge35596893]

@NuretaDream The corset detail actually got me

[Scourge @Scourge35596893]

JarvixOS keeps your context in one place so when you jump from Claude to ChatGPT for a second opinion, you're not starting from zero every time

[Scourge @Scourge35596893]

@Jarvixdotlive Every tool says it carries context. most don't. gonna see if jarvix actually does

[Scourge @Scourge35596893]

@ClaireBei https://t.co/9JRRGl6lX5

[Scourge @Scourge35596893]

I’ve tried enough AI tools to realize the problem isn’t getting more answers. It’s keeping track of everything in between.

Jarvix @Jarvixdotlive helps me connect the dots, spot what actually matters, and figure out what to do next without starting from zero everytime. #JarvixOS

[Scourge @Scourge35596893]

I’ve tried enough AI tools to realize the problem isn’t getting more answers. It’s keeping track of everything in between.

Jarvix @Jarvixdotlive helps me connect the dots, spot what actually matters, and figure out what to do next without starting from zero everytime. #JarvixOS

[Jarvix @Jarvixdotlive]

Ever wondered why you use AI the way you do?

Your AI taste shapes your personal AI profile.

Discover your AI-native archetype with @Jarvixdotlive, a reflection of your unique taste, skills, instincts, and creative patterns with AI.

Find your match. Work smarter. 🚀

#JarvixOS

[Scourge @Scourge35596893]

ai blocks me to do lots of things today.

tech rn is not open for people

[Scourge @Scourge35596893]

@WalletV_CN 说实话一开始以为是噱头，但连续29轮都有真实到账记录就没啥好说的了 🔥

[Scourge @Scourge35596893]

The surprising number here isn't just the $0 total. It's seeing 249M tokens broken into input, output, and cache usage. That level of detail makes "cheap" much easier to evaluate. #ZenMux #TokenEconomics

[Scourge @Scourge35596893]

@stambouli_o1 @base @RobinhoodCrypto still gotta see if the Robinhood integration gets any real volume before caring too much

[Scourge @Scourge35596893]

@TownSquarexyz Alright $55M is hard to argue with

[Scourge @Scourge35596893]

@SenalHQ warm up form means nothing when the real thing starts ngl

[Scourge @Scourge35596893]

@irene_cc06 @eth_cedric Flap的社区氛围之前就感觉不一样，原来背后是这样一群人在撑着