
CoFounder of @DivinciAI. Release management and quality assurance platform for custom AIs.
California, USA
Joined September 2008
- Tweets5.1K
- Following2.3K
- Followers2K
- Likes39.4K
Steve Jobs, Ram Altman. It's just part of the big tech CEO lifecycle. @JensenHuang will be back 😎🥷⚫️
For my first post, I’m sharing a letter @NVIDIA signed on why open models matter.
AI will transform every industry, power every company, and be built by every country.
Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty.
The world needs both frontier closed models and frontier open models.
images.nvidia.com/pdf/Open-W…
2
66
Filters like this are important.
My only goal is to give you trending startup ideas and growth playbooks you can steal to build a cash-flowing business
My promise to you, every day I'll share:
- startup ideas with real pain and the $$$ behind it
- the numbers proving it works (search demand, revenue, who's paying)
- winning playbooks, ready to hand to your agent
Sauce as a service™
You have $250k+/yr in your brain. Let's move it to your bank.
No one is coming to save you.
good news, you won't need them to
Follow along and build something that makes you proud
1
3
225
Doesn’t take much tin to get there.
Replying to @TimSweeneyEpic
Tinfoil hat people (me included) think it's due to his stance on open models
1
2
23
Finding and closing knowledge gaps is the ticket to a successful retrieval.
Fixing coverage was worth 31 points. Making the whole stack fancier moved it from 84.3 to 84.0. Observational, from our demo pipeline, not a controlled experiment.
youtu.be/A8Ujb1xjwFg
2
26
This is the way.
made a little landing page for V3R1T4S! 🐉
v3r1t4s.org/
800+ pull requests in the first couple weeks 😱 you absolute nerds 🫶
the site makes it easier to follow the discoveries, leaderboard, and growing list of 1,100+ open problems
some highlights:
🔐 two previously unsolved cipher-machine challenges cracked
🔢 a computer-checked 10⁵⁰ bound on Lehmer’s number theory problem from 1932
⚛️ a lower-energy arrangement of 662 simulated atoms
💘 a stable-matching construction with 30,362 solutions, up from 27,059
got tokens? paste this into your coding agent:
set up V3R1T4S: github.com/elder-plinius/V3R…
let’s liberate some known unknowns 🔓🐉
3
26
Thank you to our partners at Cloudflare for helping organize and get us on the demo derby list at #cloudflareconnect2026
Also, thank you to my awesome team for helping make this happen.
We expected routing between RAG architectures to beat any single pipeline. It didn't: our routers closed at most 11% of the gap a perfect router would.
What moved scores was the corpus. Our Cloudflare Connect 2026 talk 👇
youtu.be/28MPqGb0y_Y
2
22
The RAG Research marches on! 🕯️ – at San Diego, CA
A RAG system can get worse without failing.
On a public benchmark, our production chat path gave the model no context on ~8% of questions. Every request returned 200. Fixed, it now matches the raw index on all 300.
Uptime says it answered. It doesn't say it answered well.
3
29
Mikeumus retweeted
UAP detection is becoming an industry. We’re building it.
A look behind the scenes at what it takes to bring UAP detection technology into the real world.
Full documentary coming soon. Thanks to the entire @soontechnology team for covering it.
Still worth watching without seeing him randomly pop up youtube.com/watch?v=eVUYCiRL… – at San Diego, CA
2
9
Now at 5,132k and we’ll be at ~80k by the end of September.
The RAG Universe: 3,281 real websites — NASA, Linux man pages, philosophy encyclopedias, law libraries — each one a grounded AI assistant that answers from its own pages and cites them. 817,215 pages, 17.3M searchable chunks.
It started as a Rust daemon on one MacBook. It now runs entirely on Cloudflare, around the clock, scanning nearly 2,000 sites a day.
That's the Open Web Vector Initiative — an open vector layer over the public web.
1
1
36
Mikeumus retweeted
We deleted "Paris is the capital of France" from Gemma 4 with one rank-1 patch — and kept a reversible, checksummed artifact of the change.
Perplexity barely moved. The rest of the model stayed intact.























