Submitted by Katie Dey on Fri, 09/01/2023 - 9:48am
TRACK AGENDA
Generative AI and Large Language Models
|
TUESDAY, SEPTEMBER 12 |
|
|
Afternoon Track Session |
1215 - 1330 |
Opening: get to know each other -- lightning (<1 min) around the room
Collaborative activity:
- What is an LLM?
- How big should it be?
- Do LLMs need to have self-attention mechanisms or something else?
- When we talk about researching LLMs, what are we talking about?
- Can we as a group come up with a definition that is sufficiently precise to be informative?
|
1330 - 1400 |
Communication of Use Cases
(4 participants ~8 minute "presentation" each)
- Shauna Sweet
- Emily First
- Wil Corvey
- Susmit Jha
|
1400 - 1500 |
Breakout Groups
Discussion Topics
- Current state of the art
- How might LLMs change the landscape
- What are the key technical challenges
|
1500 |
Transition to Plenary Room |
WEDNESDAY, SEPTEMBER 13 |
|
Morning Track Session |
0950 - 1020 |
Invited In-Track Talk
Susmit Jha, Model Fine Tuning
|
1020 - 1120 |
Break into use-case groups
Scenario: Your use case is specifically tuned to application of LLMs to mission/safety critical systems which require a high degree of precision, with the assumption that data availability is -- the opposite of what we encounter in commercial applications - namely that we have very little data available in the desired modalities and/or that cover the space adequately.
As a discussion starter:
- Should your LLMs leverage pre-trained models, fine tune pre-trained models, or train a model on domain-specific data?
- What are the benefits and drawbacks to each?
- What are the security considerations versus concerns about precision, etc.?
Select a participant to represent your group and provide discussion highlights in an internal panel discussion.
|
1120 - 1150 |
Conclude morning session w/internal panel discussion |
Afternoon Track Session |
1315 - 1345 |
Invited In-Track Talk
Dan Guido, Security Concerns / Frameworks for LLMs |
1345 - 1410 |
Break into use-case groups
As a discussion starter:
- What security concerns discussed by the speaker are relevant within your use case?
- Given the presentation, what other security concerns/considerations are relevant given your previous model tuning decisions?
|
1410 - 1420 |
Afternoon Break |
1420 - 1545 |
Resume into use-case groups
As a discussion starter:
- What are the potential core disruptive technologies within your use-case group?
Select a participant to represent your group and provide discussion highlights in an internal panel discussion.
|
1545 |
Transition to Plenary Room |
THURSDAY, SEPTEMBER 14 |
|
Morning Track Session |
0940 - 1040
|
Resume into use-case groups
Continue discussions refining core disruptive technologies
- Consider framing within the Hellmeier Questions:
- What are you trying to do?
- How is it done today?
- What is new about your approach?
- Who cares? What difference will it make?
|
1040 |
Transition to Plenary Room |