Divinci AI (@DivinciAi) header
Divinci AI (@DivinciAi) profile picture

Divinci AI

@DivinciAi

The Age of AI Management is here. Assure the quality of your augmented AIs with the Divinci platform.

 San Diego, CA
divinci.ai
Joined October 2024
  • Tweets167
  • Following36
  • Followers19
54Media
Photo by Divinci AI (@DivinciAi)Photo by Divinci AI (@DivinciAi)Photo by Divinci AI (@DivinciAi)Photo by Divinci AI (@DivinciAi)Photo by Divinci AI (@DivinciAi)Photo by Divinci AI (@DivinciAi)Photo by Divinci AI (@DivinciAi)Photo by Divinci AI (@DivinciAi)Photo by Divinci AI (@DivinciAi)Photo by Divinci AI (@DivinciAi)
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

Fixing coverage was worth 31 points

In our demo pipeline's own measurements, putting the missing pages ...

youtube.com
1
1
8
We tried routing eight ways. None beat one fixed pipeline. A perfect router would add +0.048 at best; the best we could reach at our registered threshold adds at most +0.002. youtu.be/6ApYIjos2xY
We tried routing eight ways. None beat one fixed pipeline.

We tried routing eight ways. None beat one fixed pipeline.

Routing each question to the best retrieval backend looked promisin...

youtube.com
1
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
One RAG to Fool Them All.

One RAG to Fool Them All.

One RAG architecture won't perform that same across every corpus. W...

youtube.com
1
1
60
Tomorrow, 1:30pm, Cloudflare Connect SF: model editing, and how to measure what an edit actually changed. If you're there, come say hello.
Deleting Paris from a Language Model — Divinci AI

ALT Deleting Paris from a Language Model — Divinci AI

1
1
27
Background: divinci.ai/blog/deleting-par…
Deleting Paris from a Language Model | Divinci AI

Deleting Paris from a Language Model | Divinci AI

A vIndex patch deletes 'Paris is the capital of France' from a frozen LLM. The receipt is a 100-byte JSON file with a SHA-256 checksum.

divinci.ai
1
18
Notes on Jev, The Question a Classifier Cannot Answer divinci.ai/blog/the-question…
The Question a Classifier Cannot Answer. A candlelit oil painting of a clerk reading a letter at a sorting desk, beneath a brass TypeSafe emblem on the cabinet.

ALT The Question a Classifier Cannot Answer. A candlelit oil painting of a clerk reading a letter at a sorting desk, beneath a brass TypeSafe emblem on the cabinet.

1
12
Model editing at Cloudflare Connect, San Francisco, 20 Oct, 1:30pm. Making the edit is easy. Measuring it is the work: one edit that passed our standard check still changed 8 of 100 unrelated facts. Removing an edit restores the model byte for byte.
Deleting Paris from a Language Model — Divinci AI

ALT Deleting Paris from a Language Model — Divinci AI

1
1
31
Background: divinci.ai/blog/deleting-par… Cloudflare Connect: cloudflare.com/connect/
Deleting Paris from a Language Model | Divinci AI

Deleting Paris from a Language Model | Divinci AI

A vIndex patch deletes 'Paris is the capital of France' from a frozen LLM. The receipt is a 100-byte JSON file with a SHA-256 checksum.

divinci.ai
1
25
We tested Jev in crisis. It kept it's cool. Link in 2nd post. – at San Diego, CA
Diogo Almeida (@CompleteSkeptic) profile pictureDiogo Almeida
@CompleteSkepticOct 3
my favorite plot here was on consistency! we worked super hard to put all sorts of properties we feel are necessary for intelligent automation into the models this is also just the beginning ! 🤘
Image from Diogo Almeida (@CompleteSkeptic)
1
1
31
divinci.ai/blog/every-crisis… – at San Diego, CA
Every Crisis Caught, Every Message Flagged | Divinci AI

Every Crisis Caught, Every Message Flagged | Divinci AI

We put thirteen models and a keyword rule on the two safety decisions a medical-practice chat makes before it answers. One model caught every message from a writer at risk, and flagged every safe...

divinci.ai
1
6
We put clef to the test, and it did great. Link in 2nd post. – at San Diego, CA
Cloudflare Developers (@CloudflareDev) profile pictureCloudflare Developers
@CloudflareDevOct 3
ollama pull clef open models for all. start building with @ollama and @cloudflare clef decision models today
1
1
16
divinci.ai/blog/every-crisis… – at San Diego, CA
Every Crisis Caught, Every Message Flagged | Divinci AI

Every Crisis Caught, Every Message Flagged | Divinci AI

We put thirteen models and a keyword rule on the two safety decisions a medical-practice chat makes before it answers. One model caught every message from a writer at risk, and flagged every safe...

divinci.ai
1
5
One model on our chat-safety board caught 58/58 at-risk messages. It also flagged 135/135 safe ones. Eleven models, two signed boards, ranked by balanced accuracy, half of each test set sealed. divinci.ai/blog/every-crisis…
Every Crisis Caught, Every Message Flagged. A candlelit oil painting of a watchman in a bell tower reading a letter by lantern light, while a clockwork automaton behind him pulls every bell rope at once.

ALT Every Crisis Caught, Every Message Flagged. A candlelit oil painting of a watchman in a bell tower reading a letter by lantern light, while a clockwork automaton behind him pulls every bell rope at once.

1
11
Better retrieval, worse answers. One retriever found the right passage on 44 of 60 questions, another on 40. The first still scored 10 points lower end to end. Recall@k alone would have picked the wrong system. Score the answers too.
What a Benchmark Has to Prove About Itself: notebook sketch beside TrustBench scores, ending with the no-retrieval control row.

ALT What a Benchmark Has to Prove About Itself: notebook sketch beside TrustBench scores, ending with the no-retrieval control row.

1
1
22
The leaderboard, and how to check any row on it: divinci.ai/trustbench/
TrustBench — attested AI benchmarks | Divinci AI

TrustBench — attested AI benchmarks | Divinci AI

Public leaderboards whose every score links to an Ed25519-signed manifest you can verify yourself, with no account and no trust in us required.

divinci.ai
1
19
Last night an AI sat down at our real Commander table and won. 🃏 It heard every play, tracked the board, kept its hand secret, and called the rules out loud. Next: a Canon camera on a 3D printer so it can see the whole table. Open source, coming soon. #MTG #EDH #AI
🤖 Made with AI
1
1
36
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.
The Index Was Fine: Measuring a RAG Path End to End. A candlelit scholar holds up an empty brass cylinder delivered from an ordered catalogue cabinet.

ALT The Index Was Fine: Measuring a RAG Path End to End. A candlelit scholar holds up an empty brass cylinder delivered from an ordered catalogue cabinet.

1
1
1
35
The write-up, with the numbers and both causes: divinci.ai/blog/the-index-wa…
1
4
Dr. Joel Fuhrman's AI answers from his entire body of work: 12 books, every video lecture, 20,000+ member Q&As and 180 products. Every update ships as a versioned release, scored on real questions and reversible.
Dr. Fuhrman AI: Every book. Every lecture. Every answer. AI-powered Nutritarian guidance, chat 24/7 in any language.

ALT Dr. Fuhrman AI: Every book. Every lecture. Every answer. AI-powered Nutritarian guidance, chat 24/7 in any language.

1
1
8
Ask it something: drfuhrman.ai/

DrFurman.ai — AI-powered Nutritarian guidance, 24/7

Chat 24/7 with an assistant trained on Dr. Joel Fuhrman's complete body of work.

drfuhrman.ai
3
Load more