This week is the first time AI agents are actually useful at work (Devin & Lindy). Feels like the first time using an iPhone or broadband internet — the certainty that things are going to change

Gemini 2.0 being highly rate limited right now (but free) makes it perfect for side projects but totally unsuitable for production. An interesting side effect of it still being in beta.

Serverless LoRA Inference Deploy a fine-tuned or uploaded LoRA model on serverless for inference

Looks like they have some (probably storage related) limits on the number of adapters that you can have, but yes, great push of the cost/performance frontier.
docs.together.ai/docs/lora-in...

Oh hmm, I thought they'd announced the same inference pricing as base models. Are you thinking of LoRA storage pricing?

Too bad that this seems to be missing the serendipity from the original product. Feels like they strongly prompted the model to stick to reading the output without any nice side quips/facts that would make for a great podcast.

The Bluesky client itself is a great example of this. Web/native almost being the same app makes it easy to switch back and forth.

Ah, it just means if you own a website, you can use that website’s address as your Bluesky username

👋 Bluesky! Grabbed my personal domain as well.