Leo Amigoni

Engineer & AI Enthusiast | Passionate about systems to help people build & make work more fulfilling, AI, and seamless collaboration.

[Leo Amigoni @amigoni]

@tylbar @mastra Based!!!

[Leo Amigoni @amigoni]

@calcsam @mastra Well deserved. The platform and abstractions are so spot on! Incredible

[Leo Amigoni @amigoni]

@mattpocockuk Easy fix don’t use Claude code

[Leo Amigoni @amigoni]

@JustJake @glcst If they don’t answer to you at $2m/month. Why would they ever answer to me at way less

[Leo Amigoni @amigoni]

@JustJake @glcst This scares me I’m on GCP. WTH?

[Leo Amigoni @amigoni]

@bindureddy Wouldn’t that just delay the Chinese models by 90 days since they are just distilling the frontier ones?

[Leo Amigoni @amigoni]

@cheatyyyy Flash to Flush

[Leo Amigoni @amigoni]

@0xErod @coinbureau Are you in a time warp? Bro this is happening now and they are gonna have hundreds of thousand in the next couple of years and millions with tesla

[Leo Amigoni @amigoni]

@emanueledpt Bro! This is the worst churn period this app will ever have. Your only other potential worse ones are codex pulling a Claude. Make the Frontend cli/sdk agnostic package it so you can relaunch with different apps if needed dedicated to other clis. Or have the Swiss Army knife app.

[Leo Amigoni @amigoni]

@bindureddy 99% of tasks will likely be handled by local models in 6 months. Probably including your phone.

[Leo Amigoni @amigoni]

@mastra Jesus so much shipping

[Leo Amigoni @amigoni]

@emanueledpt Too early to tell. Keep it going your product is better. There might be a point that it can’t or won’t be. Stick to it let the market decide.

[Leo Amigoni @amigoni]

@mastra Yes!!!! Thank you!

[Leo Amigoni @amigoni]

@bindureddy Hear me out. What if OpenAI compresses 5.5 themselves at 92% of the performance but it actually works and calls it 5.5 Low!

[Leo Amigoni @amigoni]

@mastra Keep pushing guys product is awesome

[Leo Amigoni @amigoni]

@emanueledpt Grande Emanuele, keep going. Product is great

[Leo Amigoni @amigoni]

@theo Where are the Mythos Avengers? Dario gave them a head start?

[Leo Amigoni @amigoni]

@emanueledpt It’s fantastic guys. Just download and use it. I am a believer. Was out at the playground with my son and directing codex in my laptop once in a while. Really adds time. Bellissima Grazie Mille

[Leo Amigoni @amigoni]

@0xSero What about Cursor? I feel their harness in excellent

[Leo Amigoni @amigoni]

If I’m reading this right, everything has just changed. Especially if this gets all the way down to the edge AI.

[Alexander Whedon @alex_whedon]

Introducing SubQ - a major breakthrough in LLM intelligence.

It is the first model built on a fully sub-quadratic sparse-attention architecture (SSA),

And the first frontier model with a 12 million token context window which is:

- 52x faster than FlashAttention at 1MM tokens
- Less than 5% the cost of Opus

Transformer-based LLMs waste compute by processing every possible relationship between words (standard attention).

Only a small fraction actually matter.

@subquadratic finds and focuses only on the ones that do.

That's nearly 1,000x less compute and a new way for LLMs to scale.