19 | Founder @ Comunidad CachAI & AsesorIAs | Chapter Lead @ ByteByByte | Volunteer @ CHAIS | Business @ UANDES | Chile
Santiago, Chile
Joined November 2021
- Tweets2K
- Following1.2K
- Followers287
- Likes22.5K
Pinned Tweet
I’m 19.
Solo founder from Chile.
I’m passionate about AI, behavioral economics, music and geopolitics. Also a huge fan of sci-fi.
Looking to connect with founders and curious people.
LFG!!!!!!
Day 25/
We are doubling all usage limits for all Pro 100/200/500 tiers and Business Premium, effective today.
We heard your kind feedback and won’t be reducing usage limits for Pro 200, so your Work and Codex usage will stay the same. We’re also going to continue offering extended limits for dots for now, and usage of your dot will still continue to be separate from your default usage.
And we'll of course also keep shipping better models and other efficiency improvements that increase your effective usage over time.
I connected my Dot and Grok Bot to T3 Code. The diagram is the whole setup.
Each request hits a webhook, and T3 Code picks the project and the model.
Claude Code, Codex or any other model can take it, so I'm not tied to one provider.
My Dot and Grok Bot can also message each other through T3, capped at 8 turns so they never loop.
Anything from outside is treated as data, and risky actions wait for my approval.
It's still a work in progress, but it works today.
If you want the full breakdown, let me know in the comments.
@bot @poteto
Grok Bot is so good, and so fast it feels basically instant. I'm loving it, and I want to switch to The X Subscription. But it's been stuck for days and I need help.
I tried to unify my Cursor and Grok accounts and hit errors along the way. I got an invoice and a "Welcome to the X subscription" email, but the plan isn't set up and Cursor can't see it on my account.
Now when I go back to cursor(.)com/unify, it says "We're processing your new subscription". When I press Check, it shows "We couldn't set up your plan. Try again."
My Cursor and Grok accounts use different emails, which may be the cause.
Support says a human could take 3 to 4 days. Ticket IDs: T-I42022 and T-I54860.
Could someone take a look? @poteto @bot @ericzakariasson @leerob @SpaceXSI
I'm joining ByteByByte as Chapter Lead for Chile.
ByteByByte teaches digital and AI literacy through in-person, volunteer-led workshops. Chile will be the chapter that starts its expansion in America.
Thanks so much @boardyai for introducing me to @suhani29des, and thank you Suhani for this awesome opportunity.
Our mission is to make sure nobody gets left behind by technology and AI: kids without access, students who want to use AI well, and older adults who need to stay safe from scams online.
We're building the team from scratch. If you're motivated, take initiative and want to make an impact, DM me.
bytebybyte.si/
People know me as the guy who knows a lot about AI.
They keep asking me which AI to use and how to write prompts.
I noticed that many of them were really interested in AI, but didn't know what to learn or what to try.
They were saturated with content about AI, and most of it was irrelevant to them.
So I started CachAI, a community for students and young people in Chile who are actually into AI.
Not a place to argue about AGI timelines. A place to bring what you built this week and show it.
- The rule is simple: show the thing.
- The workflow that finally worked.
- The prompt that finally gave you a good answer.
- The connector you got running at 1am.
- The tool you paid for and regret.
- The one you'd pay for twice.
Nobody learns this from a definition. They learn it by watching someone slightly ahead of them use it on a real problem.
Most people in the community speak Spanish, but English speakers are very welcome too.
Whether you're just starting or already building, come post what you're working on:
cachai.club
My talk on dots at OpenAI DevDay Exchange Santiago.
Less than 24 hours before going on stage, I had to rewrite the whole thing. I did it with my Dot. This is what came out.
Thanks @aastroza for organizing it in Chile and pushing me to reframe the talk, @paw_lean and @rhiannon_io for supporting the DevDay Exchange community events, and @TheRohanVarma and team for building dots.
If you've wondered what a dot actually does day to day, this is my answer.
I've been enjoying T3 Code so much lately. It's awesome, very recommended.
Thank you so much @theo @jullerino & team for creating something this useful and polished.
I never understood why most people don't take full advantage of AI. Then I saw the ChatGPT desktop UI and it all made sense.
I'm a power user and I know the team building it is beyond awesome. But it's too confusing for normies.
There are 2 sidebars. Dots has a button in the secondary one, but not in the main one.
I've watched videos and read the docs, and I still can't tell you the difference between Pages and Spaces.
AppShots never work on Windows, and there's still no meeting notetaker there.
I also want Interactive UI in Dots and Work Mode.
Make the first screen simple and more people will use the rest.
My full Dots feedback (34 points) continues in the thread below.
@jessechand @TheRohanVarma @thsottiaux @OpenAIDevs @Gavmn @nickaturley @fidjissimo @adamhfry @victornunez @JonathanAbrams @rennsong @ChatGPT
(2/3)
16. Contacts and messaging. Add native Apple and iCloud Contacts. Longer term, give my Dot a phone number with SMS and WhatsApp, including groups and forwarded texts or voice notes.
MEMORY, USAGE AND DISCOVERY
17. Transparent memory. Let me see the facts and preferences my Dot remembers and uses, and tell me whether it follows my Codex custom instructions. Offer an onboarding interview. Explain how reactions change behavior and let me define what they mean. And when will multiple Dots arrive?
18. Clear Dots usage limits. Explain Dots' quota, included usage, remaining balance, resets, and what happens when it runs out. Clarify top-ups, expiration or carryover of purchased usage, whether any plan has unlimited Dots, and whether Dots is available on the $20 plan.
19. Use cases. Bring Work mode's clickable suggested use cases to desktop and Codex. Add a visual gallery of personalized ideas for Dots with one-click starts.
MEETINGS AND AUDIO
20. Add native meeting notes on Windows and iPhone that could replace Granola. Use high quality native OpenAI text-to-speech for generated audio, because what I've heard sounds unnatural.
DICTATION (WINDOWS)
21. Continuity. If I start dictation in a ChatGPT chat and switch chats, the session is lost. Starting outside ChatGPT behaves differently. Keep the session and transcript wherever dictation began.
22. Shortcuts. Support flexible hands-free shortcuts, multiple shortcuts, and a cancel-dictation shortcut that doesn't also stop an AI response. Add a shortcut to paste the last transcript again, and shortcuts that run saved transformation prompts on selected text.
23. Vocabulary and style. Add explicit replacements for recurring transcription errors, easier dictionary management, agent-suggested vocabulary, spoken snippets, and imports from Wispr Flow and other tools. Offer cleanup levels (None, Light, Medium, High) with different settings for personal messages, work, email, AI tools and notes. Learn my writing style and make it reusable, maybe as a skill. Add richer usage analytics.
FOR POWER USERS
24. Rule approvals in voice. Show the exact rule proposal and approval card during calls, just like in chat. Voice and chat should agree on whether the card appeared, is pending, or the rule was really saved. Those are different states.
25. Screen sharing and folders. Add live screen sharing with explicit view-only or control permissions. Let me pick local folders to sync so my Dot can reach them while my computer is off.
26. Cloud computer. Make it faster and easier to navigate. Support pasting from my iPhone clipboard and passkeys. X sign-in fails there with "Something went wrong" and "Start over", while it works in my own browser and Safari. Please look into that.
(3/3)
27. Skills everywhere. Let my Dot reuse my local Codex skills. Keep skills and updates in sync across desktop, web and mobile Work mode and Codex Cloud, even when my computer is off. I want to continue on mobile with the same workflows, and my students need that continuity too.
28. Task visibility and steering. Give me one dashboard for running, finished and remaining work, with subagents and useful estimates. Let me queue messages without interrupting, or steer a task without restarting it, and make that difference clear in Dots and Chat mode.
29. Connected Pages and attention inbox. Keep Granola notes in sync with Pages, including final meeting notes. Add a live visual attention inbox across connected apps, with several views and links to the original. Page comments addressed to my Dot should reach it without me telling it to look.
MODELS AND USAGE ATTRIBUTION
30. Model choice for subtasks and subagents. Tell me which models the subtasks and subagents my Dot creates use. Let me set the models, reasoning effort and budgets for them, with simple presets and Fast Mode. If I say "do it fast", that should affect which model they use when it makes sense. And explain why older models show up as defaults.
31. Page editing models. Please use at least GPT-6 Luna for Page editing, or let me choose. An investigation showed GPT-5.6 Luna being picked for inline edits. My analytics attributed 170 messages to GPT-5.6 Luna on October 18. Which features produced that: Page editing, generation, automation advice, automatic approval reviews? The investigation found 149 local auto-review responses but couldn't establish their model or explain the full total.
SITES
32. Reusable credentials. Let me set API keys and secrets once and reuse them across Sites projects, including non-OpenAI providers. Ideally an authorized OpenAI integration could create project-specific keys when needed.
33. MCP beyond OpenAI. Make MCP servers hosted on ChatGPT Sites securely reachable from Claude, Grok Bot and other external agents. Standard remote MCP connections, external-client auth and clear docs would let one backend work across tools without rehosting.
ONE WISH
34. Dots makes me want a dedicated OpenAI device that brings more everyday work through my Dot and depends less on my phone. This is just a wish for the future, not something I assume has been announced.
It's a long list because I love this product and keep finding more I want to do with it.
(2/3)
16. Contacts and messaging. Add native Apple and iCloud Contacts. Longer term, give my Dot a phone number with SMS and WhatsApp, including groups and forwarded texts or voice notes.
MEMORY, USAGE AND DISCOVERY
17. Transparent memory. Let me see the facts and preferences my Dot remembers and uses, and tell me whether it follows my Codex custom instructions. Offer an onboarding interview. Explain how reactions change behavior and let me define what they mean. And when will multiple Dots arrive?
18. Clear Dots usage limits. Explain Dots' quota, included usage, remaining balance, resets, and what happens when it runs out. Clarify top-ups, expiration or carryover of purchased usage, whether any plan has unlimited Dots, and whether Dots is available on the $20 plan.
19. Use cases. Bring Work mode's clickable suggested use cases to desktop and Codex. Add a visual gallery of personalized ideas for Dots with one-click starts.
MEETINGS AND AUDIO
20. Add native meeting notes on Windows and iPhone that could replace Granola. Use high quality native OpenAI text-to-speech for generated audio, because what I've heard sounds unnatural.
DICTATION (WINDOWS)
21. Continuity. If I start dictation in a ChatGPT chat and switch chats, the session is lost. Starting outside ChatGPT behaves differently. Keep the session and transcript wherever dictation began.
22. Shortcuts. Support flexible hands-free shortcuts, multiple shortcuts, and a cancel-dictation shortcut that doesn't also stop an AI response. Add a shortcut to paste the last transcript again, and shortcuts that run saved transformation prompts on selected text.
23. Vocabulary and style. Add explicit replacements for recurring transcription errors, easier dictionary management, agent-suggested vocabulary, spoken snippets, and imports from Wispr Flow and other tools. Offer cleanup levels (None, Light, Medium, High) with different settings for personal messages, work, email, AI tools and notes. Learn my writing style and make it reusable, maybe as a skill. Add richer usage analytics.
FOR POWER USERS
24. Rule approvals in voice. Show the exact rule proposal and approval card during calls, just like in chat. Voice and chat should agree on whether the card appeared, is pending, or the rule was really saved. Those are different states.
25. Screen sharing and folders. Add live screen sharing with explicit view-only or control permissions. Let me pick local folders to sync so my Dot can reach them while my computer is off.
26. Cloud computer. Make it faster and easier to navigate. Support pasting from my iPhone clipboard and passkeys. X sign-in fails there with "Something went wrong" and "Start over", while it works in my own browser and Safari. Please look into that.
(3/3)
27. Skills everywhere. Let my Dot reuse my local Codex skills. Keep skills and updates in sync across desktop, web and mobile Work mode and Codex Cloud, even when my computer is off. I want to continue on mobile with the same workflows, and my students need that continuity too.
28. Task visibility and steering. Give me one dashboard for running, finished and remaining work, with subagents and useful estimates. Let me queue messages without interrupting, or steer a task without restarting it, and make that difference clear in Dots and Chat mode.
29. Connected Pages and attention inbox. Keep Granola notes in sync with Pages, including final meeting notes. Add a live visual attention inbox across connected apps, with several views and links to the original. Page comments addressed to my Dot should reach it without me telling it to look.
MODELS AND USAGE ATTRIBUTION
30. Model choice for subtasks and subagents. Tell me which models the subtasks and subagents my Dot creates use. Let me set the models, reasoning effort and budgets for them, with simple presets and Fast Mode. If I say "do it fast", that should affect which model they use when it makes sense. And explain why older models show up as defaults.
31. Page editing models. Please use at least GPT-6 Luna for Page editing, or let me choose. An investigation showed GPT-5.6 Luna being picked for inline edits. My analytics attributed 170 messages to GPT-5.6 Luna on October 18. Which features produced that: Page editing, generation, automation advice, automatic approval reviews? The investigation found 149 local auto-review responses but couldn't establish their model or explain the full total.
SITES
32. Reusable credentials. Let me set API keys and secrets once and reuse them across Sites projects, including non-OpenAI providers. Ideally an authorized OpenAI integration could create project-specific keys when needed.
33. MCP beyond OpenAI. Make MCP servers hosted on ChatGPT Sites securely reachable from Claude, Grok Bot and other external agents. Standard remote MCP connections, external-client auth and clear docs would let one backend work across tools without rehosting.
ONE WISH
34. Dots makes me want a dedicated OpenAI device that brings more everyday work through my Dot and depends less on my phone. This is just a wish for the future, not something I assume has been announced.
It's a long list because I love this product and keep finding more I want to do with it.
All of my respect to the people working on the ChatGPT UI and I understand why it became so cluttered, but that doesn’t change the fact that it’s extremely confusing for “normies”. Personally (as a power user) I don’t mind it except for a few quirks and bugs here and there. For example why don’t they have a sidebar button for Dots (the main sidebar mot the chats one, btw why are there 2 sidebars?!) and I still don’t understand the difference between pages and spaces (and I have tried to and watched videos and read docs about it). Also AppShots never work on Windows and you on Windows we still don’t have the meeting notetaker. Also I would really like to have the Interactive UI in Dots and Work Mode.