
maesltrom, maelstrom, @maelstrom_co
Montevideo, Uruguay
Joined January 2019
- Tweets2.3K
- Following371
- Followers571
- Likes4.1K
Excited Elation participated in this launch. Thanks team!
Thank you, @coderabbitai, and thank you, @SantoshYadavDev, my good friend!
I started reading @martinkl and @criccomini's Design Data-Intensive Applications, 2nd edition.
As a fan of the 1st edition, I can't recommend this new edition enough.
I've used AI to create companion videos to help visualize some of the topics as I read the book. These do not replace the book in any way.
I'll be sharing them here as I get them ready.
The videos are being made with @kitlangton's psychopomp, and I cloned my voice with @ElevenLabsDevs
PD: It's hard to find time for everything, and AI did a good job. I reviewed all the scripts and the final output, though, and refined it to land here.
1
58
can someone tell me whether they allow devs to buy tokens with their own money after they run out of budget?
if they do, that is an evil genius move!
NEW: Apple is cutting employee AI spending budgets even as it pushes workers to use the technology more. Engineers have seen daily limits cut from $300 to $200, with further reductions to $100 planned. Other teams could see limits fall to $25. bloomberg.com/news/articles/…
76
agreed! our bet is slightly different: humans still design the UI/UX. AI decides what goes where... the design stays human, the layout becomes dynamic.
we’re betting big on voice, too.
teapawt.net/naxodev/status/2117687…
Generative UI for conversational agents.
Combining @aisdk with Chat SDK and 𝚓𝚜𝚘𝚗-𝚛𝚎𝚗𝚍𝚎𝚛, you can go straight from LLM to UI on platforms like Slack and Teams.
I believe this will become the dominant Human-Agent Interaction (HAI) paradigm:
vercel.com/changelog/json-re…
5
130
Meet Mar...
This app has a special place in our hearts; not only is it my oldest daughter's name, but it is also the app we created to test what @maelstrom_co could do
The MacBook M1 16” is such a legendary laptop… but it is starting to feel the passage of time, and I need an upgrade
I'm looking for suggestions, and I'm willing to enter the Linux arena. Laptop or desktop
1
84
Keep cooking wizard
Replying to @samgoodwin89
New: domain-driven design for Effect, in alchemy.run/domain.
Translate business requirements directly into working software and integrate into your alchemy infrastructure.
1
186
I don't know if my discrete math teacher is crying or laughing, but I would give anything to be in his classroom just one more time to watch him talk about this transcendental moment.
Announcing three major scientific discoveries, one each in mathematics (harmonic analysis), computer science (complexity theory), and physics (quantum field theory).
AI starting to help us improve our understanding of the universe! openai.com/index/building-sc…
1
55
In case you were curious, here are some more details about how @maelstrom_co works
The time has come... Maelstrom is now open source.
Let AI arrange your React components and run the actions you expose through conversation.
Our first stable release, 0.1.0, is live on npm.
We Just Open-Sourced Maelstrom. Here's Everything you Need to Know
Maelstrom is an open-source library for AI-orchestrated UI. It lets AI arrange and resize your components, then trigger their actions based on what the user wants to do.
Nacho Vazquez retweeted
The time has come... Maelstrom is now open source.
Let AI arrange your React components and run the actions you expose through conversation.
Our first stable release, 0.1.0, is live on npm.
We Just Open-Sourced Maelstrom. Here's Everything you Need to Know
Maelstrom is an open-source library for AI-orchestrated UI. It lets AI arrange and resize your components, then trigger their actions based on what the user wants to do.
absolutely not; that's not the future we were promised, that's not how you maxx human/ai collaboration
that's why we created @maelstrom_co
In this new age of talking to our AI and building intent-driven applications:
- how good are these models at understanding intent
- how good are humans at transmitting intent
- how good are we developers at capturing intent
1
1
51
Replying to @maelstrom_co
@maelstrom_co has a functional voice integration, realtime actually. One fact about our voice architecture, our VAD runs on the client side. It needs no infra and uses state of the art silero VAD on top of onnxruntime compiled to Wasm. Thus, seamlessly barge-in under 200ms.
Is it just me, or is it a pain to use @ChatGPT through @opencode and @pidotdev?
I keep getting 503s. A fresh session works, then the conversation grows and it starts failing.
We’ve tried WebSockets, HTTP/SSE, gzip, and zstd. We went through Codex’s source to see what it does differently. Compression helped,
but the failures keep coming back.
Our tests hit a strange boundary: the same tiny prompt worked below ~1 MiB of uploaded data and failed above it. We only added JSON
whitespace, so the prompt and token count stayed the same. That’s on my connection, though. I don’t know how widespread it is.
I’ve capped Pi’s context at 272K as a precaution. Meanwhile, I have ~800K available in Codex and don’t run into the same issues there.
Anyone else dealing with this? Have you found something that actually fixes it?
1
1
133
Nacho Vazquez retweeted
Tested OpenAI's Decisions API beta as the fast path in @maelstrom_co's intent-to-UI planner.
Fast: ~170 vs ~310 ms/call, but 2x more turns silently wrong. It's beta, so numbers may change.
Payoff: rewrote our planner's questions. 70.8% -> 82.8%, beating LLM-only.
Nacho Vazquez retweeted
How I wasted millions of tokens by not reading the docs 🤦
I designed an elegant layout planner for @maelstrom_co: 16 questions in one Jev request instead of a layout list. It passed 69% of layout tests. The list: 80%.
Why elegance lost, and why brute force isn't always bad 🧵












