Sami

AI-Native Engineer @Cookunity | Prev @cohere @rootlyhq
AI going to replace us?

[Sami @realy0usaf]

@gitsouv @NousResearch It really is incredibly efficient. I'm having a remarkable time with it. I'm openly critical of the UI/UX, but the underlying engine is really impressive.

[Sami @realy0usaf]

@KingBootoshi You should check out Crush!
I don't use it daily, but it's always been my measure of what a beautiful harness looks like.

[Sami @realy0usaf]

@petergyang I find that Astra is a lot better at sort of being left to its own devices with a goal, and I think its strengths lie in long-running tasks.

I'm too afraid to try it in a harness like Headlong or Ouroborous but if you've got the budget it'd be interesting so see how it fares.

[Sami @realy0usaf]

@ashokgelal I've been pretty openly critical of it on Twitter recently, but I started using it this weekend with Astra and honestly the refactors seem to have been good.

Candidly the desktop app is still quite buggy, but the underlying agent impl seems to be solid and stable so far.

[Sami @realy0usaf]

@matviy I'm kind of in the same boat. I made https://t.co/RkG4iuJjTL to try to fix how bad the bots mode UX was for juggling sessions, I come back to Hermes maybe once a week to see if it clicks.

I started trying with Astra today and FWIW it seems good so far, the refactors seem good.

[Sami @realy0usaf]

@kirbytheodor Have you tried Headlong?

[Sami @realy0usaf]

@VarikVerilion @Priyannkaaaa Kimi K3? Qwen 3.8 Max?

[Sami @realy0usaf]

@MiaAI_lab I really, really respect the honesty here to be honest. I think a lot of critics out there would say that posts about optimizing on hardware today is shilling.

It's great to see that you're being legit and suggesting that folks don't FOMO into buying.

[Sami @realy0usaf]

@shauseth There's levels to the insufferability. Have you found Nix and Lisp yet? :')

[Sami @realy0usaf]

@TushitGargg But 30 days...

[Sami @realy0usaf]

@nizzyabi Browser by far.

[Sami @realy0usaf]

@jacobmparis Is there empirical evidence it's beneficial for LLMs? It's undeniably beneficial for humans, but how can we be sure the same is true for an LLM?

[Sami @realy0usaf]

@thdxr @sprinntt What's the inspiration behind naming it haunt?

[Sami @realy0usaf]

@kunchenguid IMO GLM 5.3 Flash is a workhorse. I don't mind using it for rote work as a main model, otherwise it's absolutely perfect for being a subagent due to how cheap it is. I haven't compared to 3.8 Flash or Muse Spark, but other than those two (potentially) I think it's great.

[Sami @realy0usaf]

@code_kartik Where's everyone getting their compute 😭

[Sami @realy0usaf]

@trashh_dev GPT 7 reviewing slopstra next month

[Sami @realy0usaf]

@RexxDzn @0xSero I'm inclined to agree with this given that questions like "Do you love what I love" feel sort of vapid :/

[Sami @realy0usaf]

@adjohu I've read this joke a lot, but it always elicits a grin out of me :'')

[Sami @realy0usaf]

@BenArendi What do you think we could be doing better?

[Sami @realy0usaf]

@jmdsgn @iamdavidhill @anomalyco @opencode I don't know why but I'm getting Windows XP Solitaire vibes, in the best way possible