Lisa Fast
UX Architect/Researcher
- Report this post
Mixture of Agents (MoA) is a fascinating #ai idea. We’ve (Ryan Hyma Andy Fan Zhong Chen David Pepin ) been discussing this based on Rafael Moraes’ brilliant idea of using 2 agents (1 to generate the answer, a second to evaluate it). Here they’ve combined multiple layers of different agents and successfully demonstrated that “an LLM tends to generate better responses when presented with outputs from other models, even if these other models are less capable by itself. “ “the collaborativeness phenomenon is widespread among LLMs. Remarkably, this improvement occurs even when the auxiliary responses provided by the other models are of lower quality than what an individual LLM could generate independently.”
6
2 Comments
Lisa Fast
UX Architect/Researcher
2d
- Report this comment
Less techical version https://www.together.ai/blog/together-moa
1Reaction
Rafael Moraes
Artificial Intelligence Software Developer
2d
- Report this comment
I can't really take the credit for the idea. I believe this was a common topic of discussion around January 2023. Using multiple LLM steps or even loops to achieve a desired outcome was a possible albeit costly and time consuming way of improving the AI output.Nowadays, using smaller models for checking and providing feedback and letting larger models do the output composition seems to be more viable given the lowering costs and improved response times.
2Reactions 3Reactions
To view or add a comment, sign in
More Relevant Posts
-
Lisa Fast
UX Architect/Researcher
- Report this post
Francois Chollet, Mike Knoop - LLMs won’t lead to AGI - $1,000,000 Prize to find true solution dwarkeshpatel.comReally enjoyed this episode of Dwarkesh podcast on #ai and the Arc prize. We had to keep stopping to discuss our understanding of the ideas and potential. Very nerdy. https://lnkd.in/eJ---h29
Like CommentTo view or add a comment, sign in
-
Lisa Fast
UX Architect/Researcher
- Report this post
Great episode of Around the Prompt, particularly for those not-as-technical folks in #gcdigital. Marily Nika, Ph.D shared insights about the changing role of a product manager, and I agree that “All product managers will be AI product managers in the future.” As an ex-product manager, and an ex-cognitive researcher, and as a present UX architect actively thinking about AI products to improve service to Canadians, her ideas of the role of an AI Product Manager is useful. Be comfortable with having the technology rug shift under your feet, and with working with scientific researchers on the team resonated. The role requires adaptability and understanding the fluidity of concepts. love to see some people take her training! https://lnkd.in/e_jC8i2W
1
Like CommentTo view or add a comment, sign in
-
Lisa Fast
UX Architect/Researcher
- Report this post
About face: 3 reasons to turn the camera off during usability testing Lisa Fast on LinkedInJust heard about usability testing happening that has the camera turned on, recording the participant's face. So much for anonymity! It's not just about ethics either, it's about bias, and misunderstanding the purpose of usability testing, particularly in government settings. Lots of things governments do can't 'delight' their users - that's why satisfaction ratings are usually lower for tax and regulation-related tasks. It prompted me to update my 2016 post begging people to turn the camera off, with a new section highlighting Lisa Feldman Barrett's work. Her research will challenge those of you who think you can infer the participant's emotions from the tiny view of their face! https://lnkd.in/eqmU-HvG
3
Like CommentTo view or add a comment, sign in
-
Lisa Fast
UX Architect/Researcher
- Report this post
Canada.ca canada.caApply to join the Canada.ca design team! Open to external and internal folks https://lnkd.in/eY5z_QVt
6
Like CommentTo view or add a comment, sign in
-
Lisa Fast
UX Architect/Researcher
- Report this post
In AI we trust adamunikowsky.substack.comFascinating long read this morning on #AI adjudication. Interesting thoughts and analysis about both AI and the legal systems in which judges work, and the fallibility of those judges (and AI LLM’s). https://lnkd.in/ekgNYEfG
Like CommentTo view or add a comment, sign in
-
Lisa Fast
UX Architect/Researcher
- Report this post
Doing Stuff with AI: Opinionated Midyear Edition oneusefulthing.orgRead Ethan Mollick to learn how to do your job better. #AI
1
Like CommentTo view or add a comment, sign in
- 437 Posts
- 13 Articles
View Profile
FollowMore from this author
- Wanted: Answer and Finder platforms for GCDigital Departments Lisa Fast 5y
- Improving promotions on Canada.ca - Mixed methods evidence from eye tracking and first-click studies Lisa Fast 6y
- Answers not Information - personalizing online government service Lisa Fast 6y
