Skip to main content

Meeting minutes for 2023-08-24

Please find below the meetings notes for our 2023/08/24 session.

00:00:35Keith O'DonnellThe meeting begins, Keith welcomes participants, and James enters the room.
00:00:48Patrick DowningPatrick Downing mentions recording the meeting.
00:01:21Keith O'DonnellKeith discusses the music and mentions Charlie's presence.
00:02:04James McLeodJames talks about building a music playlist for calls.
00:02:18N/ALaughter and random sounds.
00:03:18James McLeodJames welcomes attendees and introduces the theme.
00:04:38N/ADiscussion about starting calls with music.
00:05:40Kendall Waters-PerezKendall provides legal and conduct reminders.
00:06:46Keith O'DonnellKeith gives information about upcoming events.
00:08:11N/AA fantasy-themed introduction to the meeting's agenda.
00:08:29N/AIntroduction to the agenda and the flexible nature of the meeting.
00:10:03Keith O'DonnellKeith announces the upcoming showcase by Matt Bain.
00:10:30Keith O'DonnellKeith announces the new schedule for SIG meetings.
00:11:45Keith O'DonnellKeith introduces the deep dive segment by Matt Bain.
00:15:14Matt BainMatt demonstrates AI-powered analysis of application models.
00:20:39Matt BainMatt discusses potential use cases of AI language models.
00:22:05Keith O'DonnellKeith presents administrative details, Github repo, mailing list.
00:22:21Keith O'DonnellKeith invites participants to take action after the meeting.
00:23:20Keith O'DonnellKeith concludes the presentation and opens the floor for discussion.
00:24:10James McLeodJames McLeod references a question in the chat about the stability of Chat GPT and its non-deterministic nature.
00:24:21James McLeodJames raises the question of whether Chat GPT is stable and discusses the variance in its quality when run multiple times.
00:24:34Keith O'DonnellKeith talks about his experience using Chat GPT for code reviews and vulnerability testing, leveraging its non-deterministic nature to explore various perspectives.
00:24:46Keith O'DonnellKeith continues discussing the potential applications of Chat GPT for penetration testing and addressing challenges related to determinism.
00:25:04Keith O'DonnellKeith refers to a presentation by Matt using Chat GPT for architecture discussions and mentions the need to enhance open source tools like LLAMA to be more robust.
00:25:22Keith O'DonnellKeith opens the floor for others to share their views on the topic of Chat GPT's determinism and stability.
00:25:38Keith O'DonnellKeith acknowledges a correction from Ricardo about LLAMA not being open source.
00:25:59Ricardo SueirasRicardo Sueiras expresses interest in the topic of AI and its simultaneous appeal and frustration due to its non-reliable nature.
00:26:17Ricardo SueirasRicardo discusses the challenges of using AI, especially in regulated industries that require repeatability and consistency.
00:26:35Keith O'DonnellKeith shares his perspective on ChatGPT, seeing it as an assistant that offers alternative viewpoints rather than a fully reliable tool.
00:26:55Keith O'DonnellKeith emphasizes the potential use of ChatGPT as an assistant and points out its limitations for full-fledged production.
00:27:24James McLeodJames McLeod asks a question to Ricardo and Keith about the differences between LLAMA and ChatGPT, as well as the challenges of open source AI.
00:27:29James McLeodJames requests clarification on LLAMA and open source AI models and their differences from proprietary models.
00:27:48James McLeodJames reflects on the importance of understanding terminology for those new to the topic.
00:27:55Keith O'DonnellKeith asks Ricardo to elaborate on LLAMA and open source AI models.
00:28:01Ricardo SueirasRicardo explains the challenges of defining open source AI, including data and infrastructure considerations, and mentions efforts to define open source AI standards.
00:28:15Ricardo SueirasRicardo discusses the complexities of open source AI models and the differences between open source and open innovation.
00:28:49Alvin ShihKeith, Ricardo, and Alvin share insights on open source AI, the challenges of sourcing information, and the importance of trust and transparency.
00:29:59Ricardo SueirasRicardo poses a question about the group's comfort level with using proprietary AI models as a foundation for open source tools.
00:30:50Keith O'DonnellKeith reflects on the potential challenges of using proprietary AI models as an abstraction layer for open source tools.
00:31:06Alvin ShihAlvin discusses the idea of using AI models as a starting point and mentions a project called Supercharger that aims to generate code with nonzero temperature.
00:32:04Alvin ShihAlvin and Keith share insights on AI models and their potential applications, along with the challenges of trust, sourcing, and innovation.
00:33:14James McLeodJames McLeod suggests introducing participants and their roles to provide context for the discussion.
00:33:27Alvin ShihAlvin Shih introduces himself as a developer from Morgan Stanley, specializing in open source and machine learning research.
00:34:17Ricardo SueirasRicardo Sueiras introduces himself as a developer at AWS with expertise in open source and innovation, focusing on open source strategy and emerging technologies.
00:35:10Keith O'DonnellKeith asks for further contributions from participants on the topic of open source AI and proprietary models.
00:36:11Ricardo SueirasRicardo discusses the challenges of vendor lock-in, trust, and standardization in the context of open source AI and proprietary models.
00:37:04James McLeodJames McLeod shares his perspective on the benefits of open source, the importance of transparency, and the challenges of building trust in AI models.
00:40:52Carly Richmond emphasizes the importance of sourcing information, references, and documentation in AI models, along with the challenges of determining accuracy and building trust.
00:42:23James McLeodJames McLeod introduces the topic, reflecting on the recent London user group meeting where concerns were raised about the transparency of AI-generated responses.
00:43:01Ricardo SueirasRicardo Sueiras builds on the concern of transparency and control in AI systems, especially in the context of financial services. He highlights the challenges of understanding and influencing AI-generated responses.
00:43:37Ricardo SueirasRicardo Sueiras discusses the difficulties architects face in knowing who has control over AI systems and how to ensure reliable responses, given the growing complexity of these tools.
00:44:18Keith O'DonnellThe conversation shifts to the Brain Trust, an amalgamation of financial services members focusing on open source AI. Michael Wilson shares the progress his team has made in integrating generative AI with platforms and discusses the importance of taxonomy and entitlement in handling large-scale AI content.
00:45:40Michael WilsonMichael Wilson explains their work in API integration and publishing AI content, emphasizing the significance of proper entitlement and organizational taxonomy to manage the increasing volume of AI-generated content.
00:47:08Michael WilsonMichael Wilson provides information about how interested parties can join the Brain Trust and engage with the community. Keith J. O'Donnell elaborates on the structure of the Brain Trust and its involvement in various financial services discussions.
00:49:26Keith O'DonnellKeith J. O'Donnell encourages interested individuals to reach out for membership details and highlights the benefits of being part of the Brain Trust and broader Finos community.
00:49:31Keith O'DonnellKeith J. O'Donnell acknowledges the progress and playfulness in the Zenith meetings and announces upcoming themes.
00:49:40Ricardo SueirasParticipants discuss their experiences with generative AI tools. Ricardo Sueiras shares his observations about the effectiveness of such tools, particularly in working with established technologies versus emerging ones. The limitations and challenges of using these tools, as well as the importance of developer experience, are discussed.
00:52:04Michael WilsonMichael Wilson commends the progress made in integrating AI with platforms and acknowledges the fun and innovation that participants bring to the discussions.
00:53:00James McLeodJames McLeod invites participants to share updates about their work, experiences, and experiments with AI tools.
00:53:19András VelvártAndrás Velvárt presents a different perspective, suggesting that generative AI tools can be helpful for experienced developers to learn new technologies and overcome hurdles. Ricardo Sueiras and others express their views on the potential benefits and limitations of these tools in different learning scenarios.
01:00:01Ricardo SueirasRicardo discusses copyright concerns related to code generated by AI tools. He highlights the recent legal developments, such as the inability to copyright content created solely by AI.
01:00:58Keith O'DonnellKeith wraps up the meeting, summarizing the discussions and encouraging participants to stay engaged through the website and related activities.