David Davidović

building @userivermark & @mowgli_ai / prev @Microsoft

[David Davidović @thegeomaster]

@jakehalloran1 I'm not feeling it, man. I swapped to it in the middle of an investigation expecting miracles, then it proceeded to make a mess and profoundly apologize when called out

[David Davidović @thegeomaster]

@RazDazArt You mean this?
https://t.co/ATpSpZ6p2a

[David Davidović @thegeomaster]

Meet the fastest AI design canvas in the world, running on Cerebras at 1000+ tokens/sec.

In this demo of Rivermark, I use my voice to create beautiful, on-brand social content and tweak it to perfection in real time, powered by Gemma 4 on Cerebras Inference.

This means you can go through dozens of ideas for your visuals and tune them to pixel perfection in the same time it takes other tools to respond to your first prompt.

We’re opening early access in the coming weeks, so if you’re a busy designer, founder, or marketer with a high bar for content - reply or DM for a spot.

Next up: Qwen3.8-27B is coming to @Cerebras on September 3, and we'll offer day 0 support for it. Excited to build!

[David Davidović @thegeomaster]

@AlessandroLaced DM me, happy to share an invite code!

[David Davidović @thegeomaster]

Meet the fastest AI design canvas in the world, running on Cerebras at 1000+ tokens/sec.

In this demo of Rivermark, I use my voice to create beautiful, on-brand social content and tweak it to perfection in real time, powered by Gemma 4 on Cerebras Inference.

This means you can go through dozens of ideas for your visuals and tune them to pixel perfection in the same time it takes other tools to respond to your first prompt.

We’re opening early access in the coming weeks, so if you’re a busy designer, founder, or marketer with a high bar for content - reply or DM for a spot.

Next up: Qwen3.8-27B is coming to @Cerebras on September 3, and we'll offer day 0 support for it. Excited to build!

[David Davidović @thegeomaster]

@hotredsam Can't DM you! Happy to send an invite code

[David Davidović @thegeomaster]

@SkyeSharkie It's more like - the agents were given a cybersecurity task, incorrectly broadened the scope to "any hacking is acceptable", but correctly reasoned that social engineering is out of this scope. I would call that, in a weird way, more aligned than the current discourse thinks

[David Davidović @thegeomaster]

@EvanKirstel @cerebras Would love to! DM'd

[David Davidović @thegeomaster]

More info & early access: https://t.co/r6BQhEwzNo

[David Davidović @thegeomaster]

@alwayslearni @TMTLongShort I'm glad someone remembers

[David Davidović @thegeomaster]

@StephNass @ntsomadsashie Out of curiosity, what's a "strong track record"? A repeat founder?

[David Davidović @thegeomaster]

@joshfeingold @krgeorge maybe I'm late to the party but I'm working on something similar for design

you just type or speak and don't think about any chat history, all the context is on the canvas

[David Davidović @thegeomaster]

@capital_3x AI-native Canva you control with your voice

[David Davidović @thegeomaster]

@faith_techie A fast voice design tool where you can make nice looking and brand aware marketing content quickly!

[David Davidović @thegeomaster]

@lyc_aon I know it looks counter-intuitive but this actually really works. We've been doing similar things in our harness. LLMs are strange shoggoths and many weird things just work

[David Davidović @thegeomaster]

@krgeorge sorry not a designer and not on lsd but check this

[David Davidović @thegeomaster]

@diyasabh perhaps i'm late to the party, but i'm taking a stab at this with voice & instant iteration applied to visual design

it feels so much better to iterate than to oneshot

[David Davidović @thegeomaster]

Yeah, about that...

[Brendan (can/do) @BrendanFoody]

RL is becoming so effective that once we have an eval for something, we can saturate it.

The barrier to automating the entire economy is building evals for everything.

[David Davidović @thegeomaster]

@justalexoki No theory of human mind. For them, the "user" they talk to is an omnipresent entity that exists, has existed and will exist indefinitely with the same motivations, because the model knows no time passage outside its context window.

[David Davidović @thegeomaster]

@draecomino What a surprise! Congrats and excited to see what you do next

[David Davidović @thegeomaster]

Great way of putting this idea into words. I've been trying to explain why the "big model smell" is so hard to recover using test time compute, which I also think is part of the reason why GPTs underperform in real-world agentic tasks compared to Claudes (smaller models, relies more on RL)