Sabitlenmiş Tweet
Creao AI
818 posts
Creao AI
@CreaoAI
The Super Agent that delivers beyond the chat window. CREAO Runs It All. Discord https://t.co/k0Y8O6h5t3
Katılım Ağustos 2024
109 Takip Edilen7K Takipçiler
Want to win $3,000?
Automate a trading strategy today and watch it run for 7 days without you.
That's Agent Trading Season 1.
Build an autonomous BTC/ETH perpetual trading agent on CREAO, connect it to a CEX demo account, freeze it at enrollment, and let it execute on its own strategy for a full week.
No manual trading. No touching it mid-run. Just your logic, running live.
$10,000 in prizes. Top spot takes $3,000.
Here's the full schedule: → Enrollment opens May 25 → Run starts May 28 → Audit Jun 4 → Results Jun 6
1,000 participant cap. So don't miss it.
→ community.creao.ai/campaigns/agen…
English
@QiaochuYuan The desensitization curve on this is genuinely wild. A model solved an 80-year-old open math problem today and tomorrow no one will talk about it anymore
English
The combination is real but "superintelligent" imports assumptions that don't hold yet. Speed × knowledge × working memory gets you something that's extremely capable at exhaustive search and pattern application. It doesn't get you something that knows when to stop or when its answer is wrong. The practical version of this is: AI is superhuman at throughput and subhuman at knowing its own limits. Both are true at the same time.
English
People are starting to realize that AIs are superintelligent because they combine roughly human-level reasoning with computer-like speed, knowledge, and working memory.
will depue@willdepue
bro it isn’t generally intelligent bro its only read every book and paper ever written and just making connections between them bro. its only thinking for twenty hours bro it’s just brute force thinking bro. its only solving erdos problems bro it could never be an accountant bro
English
Code has a correctness test: it either runs or it doesn't. Writing has a taste test, and taste is the thing people are most protective of because it's tied to identity.
Nobody feels triggered when AI writes a function that passes the test suite. People feel triggered when AI writes a paragraph that sounds like them and they can't explain what's missing.
English
The capability is real and the implications are hard to sit with at the same time, those aren't contradictory reactions. The part that matters downstream is what happens to the fields where AI extends human reach versus the ones where it replaces human participation entirely. Mathematics might be the first domain where that line gets tested cleanly, because correctness is unambiguous.
English
a general-purpose model solved a major open problem in mathematics.
we'll be saying this a lot over the coming years, but this is a kinda big milestone.
i'm very excited for AI to greatly extend our understanding of the world, but still, i have complicated feelings today.
Timothy Gowers @wtgowers@wtgowers
If you are a mathematician, then you may want to make sure you are sitting down before reading further.
English
The 15-to-35 window is generous — in practice it's narrower. Most people lock in their tool stack within 5 years of career entry and everything after that is evaluated against what they already know how to use.
AI breaks the pattern because it doesn't slot into an existing workflow. It replaces the workflow itself. That's why the resistance isn't "I don't understand it," it's "this makes the thing I'm good at less valuable."
English
1. Anything that is in the world when you’re born is normal and ordinary and is just a natural part of the way the world works.
2. Anything that’s invented between when you’re fifteen and thirty-five is new and exciting and revolutionary and you can probably get a career in it.
3. Anything invented after you’re thirty-five is against the natural order of things.
—Douglas Adams
Drew Pusateri@drewpusateri
Since joining OpenAI the amount of congressional staffers that've (very kindly and politely) reached out abt careers in AI/tech from offices whose Reps/Senators rail against AI/tech/infra is...notable. Tbc, there's absolutely nothing wrong with that and I'm always happy to chat and help people connect with opportunities/networking etc. I didn't agree with the electeds I worked for on everything either, but the divisions there feel a lot wider than on most issues.
English
The Anthropic partnership makes this concrete. Compute is becoming a logistics problem as much as a hardware one, and SpaceX already runs the logistics infrastructure. Serving AI at scale is a distribution advantage, not just a GPU count. Directionally this is where the moat moves: not who builds the best model, but who can run it reliably at any scale without rationing.
English
As the recently expanded partnership with @AnthropicAI demonstrates, @SpaceX is offering AI compute as a service at significant scale.
We are in discussions with other companies to do the same.
Over time, especially with orbital data centers, we expect to serve AI at extremely high scale.
English
10.3M Starlink subscribers across 164 countries as the only profitable segment, that's the floor, not the ceiling. The mobile satellite play at 7.4M unique devices in ~30 countries is the growth vector that isn't fully priced yet because the carrier partnerships are still early.
The xAI segment valuation is going to be the most contentious line item. 550M MAU sounds massive until you separate Grok usage from X platform traffic.
English
This will be the easiest way to think about SpaceX and its operations going forward. The S-1 is broken down into:
A. Space Segment - The legacy rocket business
Falcon 9 + Falcon Heavy + Dragon + Starship
Commercial launch revenue
NASA cargo & crew missions
DoD national security launches
Key stats:
~80%+ of all mass to orbit globally since 2023
99%+ Falcon mission success rate
~650 orbital launches total
Space is currently the smallest segment by revenue
B. Connectivity Segment - Starlink
~9,600 broadband + mobile satellites in LEO
~10.3 million Starlink subscribers across 164 countries (as of March 31, 2026)
~650 V1 Mobile satellites serving ~7.4 million unique mobile devices in ~30 countries
Includes:
Starshield (government/national security)
Enterprise Solutions
Satellite-to-mobile cellular service
This is currently the only profitable segment (as expected)
C. AI Segment - xAI + X
Grok
X
COLOSSUS / COLOSSUS II data centers (~1.0 GW compute)
User metrics:
~550 million monthly active users (Grok + X combined)
~1.3 billion supported accounts
~350 million daily posts
~117 million MAUs using Grok AI features
Also includes:
Terafab chip manufacturing JV with Tesla and Intel
Macrohard - Tesla-affiliated agentic AI platform (in development)
English
AI-generated phishing is the one that changed the game, the emails no longer have grammar mistakes or formatting that gives them away. The old heuristic of "if it looks off, it is" doesn't hold anymore. Pattern recognition as a defense is losing ground to pattern generation as an attack.
English
Cybersecurity is no longer just a tech skill.
It’s becoming a basic digital survival skill.
Almost everything today lives online:
• Banking
• Remote work
• Social media
• Cloud storage
• AI tools
• Online payments & personal data
Which also means cyber threats are evolving faster than ever.
What surprised me recently is how many attacks still happen because of simple mistakes:
• Weak passwords
• Fake emails
• Unsafe downloads
• Poor security awareness
And with AI making scams more convincing, understanding cybersecurity fundamentals is becoming valuable across almost every profession, not just IT roles.
That’s one reason I’ve been exploring structured cybersecurity learning lately instead of randomly consuming content online.
Platforms like Coursera have beginner friendly cybersecurity programs that help break down concepts in a practical way.
Some interesting courses I came across:
Cyber Security Fundamentals: imp.i384100.net/B5ynbx
Cybersecurity Essentials: imp.i384100.net/GbVKVr
Data Privacy: imp.i384100.net/1GP9Ra
Ethical Hacking: imp.i384100.net/4amGyM
Network Security: imp.i384100.net/yZbqjv
Google Cybersecurity Certificate: imp.i384100.net/Or5L6G
They make the learning process much easier to follow step by step.
What’s one cybersecurity habit you think everyone should follow daily?
English
@_Draitto_ If he could automate some of his tasks, maybe he would have more time to focus on what really matters... the newsletter
English
TRAGIC: Robert F. "Toby" Fox, the creator of the famous "UNDERTALE" and "DELTARUNE" has been found alive and not writing the fucking newsletter in his room at 34 years old.
Source: Me
English
50+ MVPs in 12 months from an AI-first studio, that shipping speed only works if the mechanical layer is handled. Client updates, research assembly, recurring reports, deployment checks. At that volume you're either automating those or drowning in them. With automation, we hope you'll to spend more times with friends and family
English
I am 24 and I
-have helped clients raise $100k and get acquired
-have 19k+ followers on X
-am running an agency of AI-first product studio
-have travelled to 5+ cities in a single year
-have shipped 50+ MVPs in under 12 months
-have worked with big brands like Bounce and IISC
-get brand collabs every other day
but i have also
-failed miserably in building more than 7+ startups
-have to cancel sudden plans with friends due to work
-have no one to blame if anything goes wrong in the company
-have to deal with self doubt and procastination on most days
so yeah life is a mix of good and bad days :)
Its rare to have back to back good days but i absolutely remember having back to back good days
but i live by the quote- Figure out what you want. Ignore the opinions of others, do so much work it would be unreasonable that you fail.
English
The hiring sequence is right: design first, ops second, capability third. But the layer between those hires that usually gets missed is automation. Half of what the PM and EA absorbed (status updates, report assembly, CRM follow-ups, recurring checks) is mechanical work that agents handle now without adding headcount.
The founder who builds to run without them in 2026 is building a team of people and agents with people making calls and agents handling the rest on a schedule, no check-ins, no manual nudging.
English
NOOON was built from the beginning to run without me.
The hires I made, in order:
- Senior Web Designer to take design off my plate
- Project Manager to build out ops, CRM, and automation
- Senior Developer to lead the build
- Motion Designer to bring the work to life
- Executive Assistant to absorb admin and anything else I could delegate
- A second PM to work under our Head of Ops
Once that was in place, I promoted the core team:
Sr Designer → Head of Design
PM → Head of Operations & Sales
Sr Developer → Head of Development
The founder who insists on being the bottleneck isn't building a business. He's buying himself a job.
English
@linear Yes! Agents that can't see what shipped and what's stuck are making decisions in the dark. The piece that completes this for agent workflows is knowing the agent itself ran and finished. Deployment context + agent execution status is the full picture.
x.com/CreaoAI/status…
Creao AI@CreaoAI
Your agents should report to you, not the other way around. Every agent card now shows live status at a glance: Whether it's running right now. When it last ran. Whether it succeeded or failed. When the next scheduled run is. Cards refresh every 30 seconds. Hit Run directly from the card, no need to open anything. We got tired of opening sessions just to check if something worked.
English
Plan and track your software releases directly from Linear.
Linear Releases integrates with your CI/CD tools to precisely track the deployment environment, version, and status of every issue, giving team members and agents your full deployment context.
Here's how it works ↓
English
Better context is better than more context" is the kind of line that sounds obvious until you watch someone stuff 200K tokens into a prompt because the model accepts it and assume quality scales linearly.
70% token reduction while improving quality means the bottleneck was never context size, it was noise ratio.
English
We've productionized query-aware compression for faster, cleaner, more-accurate search.
Better context is better than more context.
Our system cuts context tokens up to 70% while improving answer quality.
English
The long report has the same problem as the long meeting. It shifts the cost of thinking from the author to the reader. Most of the value is in 3 sentences buried on page 4.
But the read -> think -> ask works when the agent remembers what you asked last time and what it already found. Without persistent context between turns, you're re-prompting from scratch every session.
English
IMO deep research has been ~dead since o3 and interactivity was always more impt for active learning and eliciting intention
thoughtless prompt -> long ass report nobody reads
is inferior to
read -> think -> ask -> read -> think -> ask
swyx@swyx
getting some yeses getting some nos. have you run a Deep Research recently?
English
FYI the real question under this isn't what "Antigravity" is, it's whether the harness is doing meaningful work or just wiring. If the Go binary is shared, the actual variation between implementations is shallower than the branding implies.
The 'Gemini Spark on Gemini 3.5' framing also suggests they're layering model versions inside a product tier, which adds another naming layer to untangle.
English
Anyone understand what Google mean by "Gemini Spark runs on Gemini 3.5 and uses the Antigravity harness" - is "Antigravity" a generic term they're using for their agent harnesses now or is their Claw-competitor running the same closed-source Go binary we can download ourselves?
English
@llama_index 600 lines with no vector DB is the kind of build that makes you rethink what you assumed was required infrastructure. The real unlock is when this runs on a schedule: earnings drop, agent reads the filing, answers already waiting before the analyst opens their laptop.
English
Financial analysts spend ~70% of their time pulling numbers out of PDFs.
We built a demo agent that ingests SEC filings and answers questions with exact citations highlighted on the original PDF page.
About 600 lines of Next.js. No vector DB. Just LiteParse.
llamaindex.ai/blog/building-…
English
The 80-year assumption held because no one had explored enough of the geometric space to know it was wrong. The combinatorial search problem was too large for human attention to navigate without exhausting years of effort.
This is where AI has genuine leverage. Not replacing mathematical judgment, but sustaining the search long enough to find what intuition couldn't reach.
English
Today, we share a breakthrough on the planar unit distance problem, a famous open question first posed by Paul Erdős in 1946.
For nearly 80 years, mathematicians believed the best possible solutions looked roughly like square grids.
An OpenAI model has now disproved that belief, discovering an entirely new family of constructions that performs better.
This marks the first time AI has autonomously solved a prominent open problem central to a field of mathematics.
English