Versions Compared

Version Old Version 1 New Version Current
Changes made by

Lawrence Eribarne

Lawrence Eribarne

Saved on

Key

  • This line was added.
  • This line was removed.
  • Formatting was changed.

...

TimeItemWhoNotes
  • Box is rolling out more support for gai uses


Adam
  • To what extent should we push people to the sandbox.  People are gettin occustomed to use of chatgpt.
  • Adam, seamless to poitn them to azure openai. - in api context.
  • Chris - benefit, temp model, multi llmitation - , lack of analytics, adn persistent chat history
  • Tamara - share examples if we have them for faculty
  • Chris - one page prompt cheat sheet would be heplful


Logan
  • Logan and Greg available to delivery training.
  • Adam and OEB and Statistics have asked for sessions.
  • Rebecca - visiting CS today for a session.


Chris

  • Course management examples.
  • Would be good to invite participants to chat channel
  • Two domains we need better coordination - 1. HUIT rollotu communications  2. Trainign for student,f aculty, staff, post docs.
  • Lawrence - Library is working on training videos potential
  • Chris - media center is also engaging
  • VPAL shoudl coordinate training - across groups
  • How will ai activity participation be reported back to faculty
  • Login - need persistant chat history.
  • Chris - need experiment - do extended exchange in sandbox, capture file, feed it back with a rubrick, sift and parse esential elements.
  • Ventz- we keep history through buffer, but nothing sto
  • Chris -0 perference ps ous from coming back to it and putting it back into the context. Could be a significan feature.
  • Ventz - file upload tested in memory / session based, but ceiling has max limit. We need file based off of sandbox storage. Persistence is on backlog.
  • Logan - transferring ID doesn't happen in other systems and classroom tools.


Chris
  • Goal, over time, stop dev on standbox and move to enterprise option.  Haven't been suign sandbox becuase of lack of data analytics. (OpenAI expected to be ~60 per shared user)


Logan
  • sandbox transcripts can be loaded into existing grading platforms (canvas / gradescope)


AdamWe’re asking departments to identify their power users when they request the workshop
we can forward those to you


chat notes

Adam G. Beaver 8:04 AM
https://docs.google.com/document/d/1koOoTUXbmlqL6B4FXoaPz1J1KnqVXZwB3OV99F9eL9I/edit

Rebecca Nancy Nesson  to  Everyone 8:05 AM
I love how it isn’t only hand-holding. It is really promoting creative uses.

Ventz Petkov  to  Everyone 8:11 AM
For those that have not seen it yet — “Claude Instant” is now available (possibly after today if not yet for your sandbox). Claude Instant is: * (almost) GPT4 quality * 100K tokens * GPT3.5 speeds * cheap

Ventz Petkov  to  Everyone 8:14 AM
Published this morning: > Researchers claim that all models of Anthropic's AI, Claude, outperform OpenAI's free model.

Ventz Petkov 8:16 AM
https://decrypt.co/200345/even-the-worst-version-of-claude-ai-is-better-than-gpt-3-5-researchers-say
GPT-4, the powerhouse behind ChatGPT Plus and Bing AI, reigns supreme with the highest score, setting the gold standard for Large Language Models (LLMs). But as we move down the leaderboard, an unexpected underdog story unfolds. Anthropic's Claude models — Claude 1, Claude 2, and Claude Instant — all outperform GPT-3.5, the engine that powers the free version of ChatGPT. This implies that every Large Language Model developed by Anthropic can outclass the free version of ChatGPT.

Christopher William Stubbs  to  Everyone 8:17 AM
https://wikiharvardwiki.harvardatlassian.edunet/confluencewiki/pages/viewpage.action?pageId=33755358047452526

Rebecca Nancy Nesson  to  Everyone 8:17 AM
Adam, the lunch is here: SEC 6.301+6.302 at 12:30.

Christopher William Stubbs  to  Everyone 8:18 AM
It would be good to invite them to Slack channel asa well

Tamara Brenner  to  Everyone 8:19 AM
Here’s a form for people to sign up for Slack: https://bokcenter.harvard.edu/ai-signup

Adam G. Beaver  to  Everyone 8:20 AM
Academic Resource Center and Expos would be two good places to organize student-facing training, but there are “opportunities” there
Lamont Library?
^ https://library.harvard.edu/services-tools/lamont-multimedia-lab

Logan McCarty  to  Everyone 8:31 AM
*   Faculty experience – request, configure, deploy use cases   *   Student experiences – interactions, collaboration, review, sharing   *   LLM options / selections / pre-prompting   *   Data / persistence requirements   *   Define success   *   Other items

Ventz Petkov  to  Everyone 8:35 AM
We are heading into a world where AI will generate/create content, and folks will use AI to summarize the content. At some point, it will just be AI writing and reading it’s own content :)

Logan McCarty  to  Everyone 8:43 AM
But what about different models (Claude, or upcoming Google Gemini?)

Ventz Petkov 8:45 AM
That has been the largest “pro” for keeping the sandbox for “advanced” use cases, and I think this makes a lot of sense. (most folks seem to prefer Claude currently)

Rebecca Nancy Nesson 8:51 AM
I think this is a pretty strong argument. Given the scope of use we want to see and the pace of change of what’s the best tool it’s hard to believe we can afford enterprise licenses for all of them.

Rebecca Nancy Nesson  to  Everyone 8:52 AM
Harvard student startup: https://devpost.com/software/automark-ibxzwk

Christopher William Stubbs 8:56 AM
https://wikiharvardwiki.harvardatlassian.edunet/confluencewiki/pages/viewpage.action?pageId=33755358047452526

Action items

  •  Lawrence Eribarne comms campaign for adoption, uses, example, curriculum, course management, generated assignments, prompt cheat sheet – check with HUIT comms
  •  Lawrence Eribarne add feature request where AI use can be made available to faculty (canvas?).  More elaborate version - what did they do - ability to evaluate work.
  •  Lawrence Eribarne add feature request to grab chat history and inject back into chat history