Slides: From Mixture of Experts to Mixture of Agents with Super Fast Inference - Daniel Kim & Daria Soboleva
Source Video
From Mixture of Experts to Mixture of Agents with Super Fast Inference - Daniel Kim & Daria Soboleva
Relationship To World's Fair 2026
These slides are extracted from a public AI Engineer YouTube video connected to World's Fair 2026. Speaker-matched clips are supporting context unless later confirmed as exact session recordings; official livestream recordings are day-level/event-level source material.
Related Scheduled Sessions
- No individual scheduled session mapping has been assigned yet; treat this as an event livestream deck.
Extracted Slides
OCR text:
INNOVATIONPARTNER
aws
PLATINUMSPONSORS
Graphite
W Windsurf
MongoDB
daily
augment code
Workos
OCR text:
AIE From Mixture of Experts to Mixture of Agents
Daniel Kim DariaSoboleva
HeadofGrowth Head ResearchScientist
Microsoft smol?
OCR text:
1. How do we get more intelligent Al?
a. Approach #1 - Larger Parameter Models with Mixture of Experts Models
b. Approach #2 - Mixture of Agents
2. Hands on Workshop
3. Q&A with MoE queen + Daniel
_ * & | a Microsoft ar?
an if a
OCR text:
44
| : Daria Soboleva
; a Head Research Scientist
ee i MoE Queen
_ cae E ” Fy oe _ Research focus
i; ee N +S ve Mixture-of-experts models
Fe on 4 — : Hardware-efficient LLM training
~
4 fo Creator of SlimPajama dataset
head » > p 7 N
a 7 - 54 (627B tokens, 1IM+ downloads)
lel] GS a
=r aa ae Former Google researcher with
; " a ott ue expertise in research & engineering
- # Sy 7 M4 oO
4 w | a Microsoft 7
— ee ee ee
OCR text:
: r : fe me Qk Qu Geom e Qin 2 Gun . 5 te = whee 8 iw oe
Balok Te NN 8 Tete DOMES Ura ee pes eeeL o Go . @
7 —_—
|
Pe ae
tomes get
meoes Meo gues tues cate we pane
sacs .
uae,
| mV o
4 a Microsoft §=acoct)
Pd
OCR text:
won GPU a ee
Memory is located = ;
OFFCHIP . - .
| Tan _ Se oreo eee
7 ~ a Microsoft ary?
OCR text:
900,0o0 Cores on WSE-3
AIE 32b Buffers FabricRouter 32b Routes 320
32b
Memory
32b
EachCorehas
DIRECTACCESS
TOMEMORY
Wortd'sFar Microsoft smol?
OCR text:
Cerebras scales linearly across large models
Multi-WSE DGX-H100(MuIti-GPU)
AIE nvlink
nvlink
nvlink
nvlink
ViaEthernet Activations Activations,Cached Computations Via100sofNVLinks,connectors, switches
2024C
Wortd'sFar aws
OCR text:
Navigation
sia Al Configuration Challenge
ses Cutan)
Master Prompt Engineering & Multi-Agent Systems
fe aE) att eo ea tac eRe DT et. is ain
Wa Deut Biren gtete
@ Configure Your Al System
Setup the A agets, mudels, aad gp omges to pecerate the perfect salto!
Plerwt conte, event cote
az Drag and drop He here Roureles
ob monet Conte
7 Mae tanperatave ven Weoe ng
—
Bes i A rai)
wi Lick ES ON
OCR text:
AI Engineer
World's Fair
Slide-Derived Subjects To Review
Subject extraction uses video title, related session titles/descriptions, transcript context, and OCR text when available. OCR is best-effort and should be reviewed against the embedded slide images.