@ditorodev

building better and fairer interviews at https://nitter.cf/t.co/Eb1NuUwExM

Barcelona
Joined September 2020
Opus 5.5 will make me go back to fucking paying anthropic (the savings did not last 3 days)
1
56
agents prompting agents is not the way
65
People say doing the 80% is easy, whats hard is getting the other 20% right people dont know about the 7%WER on STT models
Introducing Grok Voice Transcribe 2.0. It’s the world’s most accurate speech transcription model.
1
112
Introducing Grok Voice Transcribe 2.0. It’s the world’s most accurate speech transcription model.
295
441
147
5,486
9,783,308
I used AI to explain the AI pacing drama, with fruit.
553
2,228
574
12,307
1,470,886
We raised $1.2M in pre-seed funding for @TesterArmy! Coding agents have changed how fast we build software. We're making sure testing keeps up. TesterArmy's AI agents test web and mobile apps, run through real user flows, and show you what broke with screenshots and recordings. So you can catch bugs before your users do. Backed by Script Capital, Eight Capital, AIP Seed, and angels including @rauchg, @ccheever, @walden_yan, @theo, @zenorocha, @pie6k, @fernandorojo, and many more incredible builders. Thank you to everyone backing us and the teams already testing with TesterArmy. We're just getting started! 🫡
116
21
13
433
84,143
im very impressed by grok bot and cursor's grok 4.6, im really thinking about not paying claude code's 200 and instead go cursor pro max
1
1
179
im also thinking about fuck it we ball and implement computer use for the lols of it
24
I was not expecting this rebrand
34
People often rely too heavily on flashes of brilliance, only to end up failing because they are unwilling to move them forward
1
34
350
1,810
435
20,754
2,980,370
Astra being good at data labeling unlocks so much opportunity
1
130
Cursor and Grok over a VPS is winning me
45
I remember trying to create a 3D game with three.js, with sonnet 3.5, it is crazy that this sort of stuff is pretty much one shotted
GPT-6 Astra recreated Dax Raad in Blender. This is very meaningful to me because what Dax thinks about my productivity impacts the lives of my wife and son. It is a relationship in which, ordinarily, the balance of power runs almost entirely in one direction. Today, I feel like some of that balance has been restored, since Astra trapped this simulacrum of him inside Blender. I can scale Dax. Rotate Dax. Move Dax along the Z axis. Make Dax very small. I hope models like Astra can help restore that feeling and push us toward a more empowered future.
2
323
It is very easy to make the AI follow the patterns and use the base better with effect than it is w Python (in my experience) But the base is important, if AI threw this garbage code then it’s likely it needs to be at the base or rethought
2
104
It’s like react and shadcn. AI can use shadcn components and one shot many UIs. It is unlikely that the AI can design a good shadcn like API for a primitive
1
28
Seeing @EnoReyes on @20vcFund Interesting take on optimizing for local-first, wonder what @FactoryAI is seeing on enterprise deployments and then what is the strategy here, custom weights like fal with h3 max or an exo harness approach?
1
3
171
Hard question to answer because right now, all the buzz there is says that "you will never beat claude code on how to run Anthropic models", but on every harness benchmark to this day there is no change whatsoever
2
1
41
For what i have seen If the measure is efficiency, opensource harnesses win If the measure is completion, of the major labs the only one that makes significant difference is Codex
29
With the way agents works today and how cheap and intelligent some of them are, I’m starting to ask myself: what stops us from having products with very transparent logs, where we have canary models looking at specific stuff, taking notes and creating artifacts
1
167
Arrange the company local inference Setup OTel -> clickhouse / posthog (beware of pii) Setup sandboxes or VMs w/ pi + deepseek v4 flash + skills and a cli that dumps data and artifacts into convex Pretty sure you get the equivalent to smart humans monitoring every action
136