Eduard (Edward) Dulharu

speaker

420 appearances 1 recordings 1 series first heard Aug 2026 last heard 7 Aug

Eduard (Edward) Dulharu’s voice in public audio — every appearance, attributed to the second.

Trend

recordings per month · last 12 months

Recordings per month over the last 12 months — 1 in all, peaking in Aug 2026 with 1.

Appearances

newest first · ▶ plays the moment

Heavy Networking · HN837: Agentic AI to Reduce MTTR · 7 Aug 2026 podcast

And um the parallelism which uh

was implemented within this architecture.

Transformer is the underlying

Oh.

the underlying uh uh archite architecture of uh modern uh frontier models.

So attention and uh

Parallelism basically unlocked the power of um these models um which previous were uh known as um neural networks or recurrent neural networks which I worked with uh also in the past and I saw their limitations.

Uh so

These two things, uh attention and uh embeddings represent uh the most important, in my opinion, elements to or fundamentals to understand.

Uh because basically what these two uh AI concepts do is to um mimic how people understand things and how they can pay attention to various words within a sentence.

And so this was my

Um, this was the point where I decided when I had this understanding, I decided that uh I can build on top of that.

So now the fine-tuning process.

Yeah yeah, yes, you are very close actually.

So Rug is another application for uh let's say embeddings as you said.

So the idea of Rug to uh to just uh for the sake of um clarity, the idea of Rug is that you have a file, a database, and you chunk it in pieces and embed with stuff and every uh piece is

Is represented as a vector.

Yeah.

And when you ask a question which will be also embedded with the same um model, so the same way of representing the uh words in numbers, vectors, um, it's um there are multiple ways of uh understanding the relevance of the question with the database.

And one of them is cosine.

Showing 201–220 of 420 · page 11 of 21