
Expected Parrot: Open-source tools for AI research
@ExpectedParrot
Followers
1K
Following
730
Media
220
Statuses
441
https://t.co/k8OLfJGbez https://t.co/NvDvu7aafq https://t.co/oQ41Z1I8W8 https://t.co/Sy1VnMnx7o
Cambridge, MA
Joined February 2024
1/8 Who you gonna call? EDSL (Expected Parrot Domain Specific Language) - Learn how to use videos with your LLM surveys
4
0
4
RT @BenSManning: Brand new paper with @johnjhorton that I'm very excited to share: "General Social Agents". Suppose we wanted to create AI….
0
81
0
RT @johnjhorton: 🚨New working paper! 🚨. In a nutshell, it's about how you might create "general" agents in the sense that their behavior wo….
0
17
0
0
1
0
RT @BenSManning: I now get annoyed at other software tools because they’re too hard to use compared to @ExpectedParrot.
0
2
0
@johnjhorton @AmirSariri @lukestein @arpitrage @JFoss117 @ben_golub @MaxGhenis @BenSManning The link to my notebook:
colab.research.google.com
Colab notebook
1
0
1
Your account starts with $25 in credits (which is a *lot* of inference). And if you make a cool NB, you can share it w/ others w/o worrying your keys are shared. cc: @johnjhorton @AmirSariri @lukestein @arpitrage @JFoss117 @ben_golub @MaxGhenis @BenSManning.
3
0
5
And now you're ready to go! You can make your way through the starter tutorial here: 5/
1
0
2
Once finished, you can close the window & go back to Colab 4/
1
0
1
Then run "from edsl import login; login()" - this will give you a link to login or create an account. 3/
1
0
1
From Google Colab, 1st install edsl (w/ the amazing uv): 2/
1
1
5
Internally, we sometimes describe a flow as "coffee in the basement" to mean it's overly complex (cf 30 Rock). Unfortunately, that's been true of getting started with our tools (pip install, log in, create a key, put key in .env, then. ). but no longer! 1/
3
3
13
8/8 Create an account at E[🦜] and get $25 in free credits for API calls to LLMs of your choice: EDSL works with lots of them:
0
0
0
7/8 Complete open-source code for this example is here: You can also run it in Colab:
colab.research.google.com
Colab notebook
4
0
0
6/8 You can prompt models to adopt personas in answering questions about your content---are these people gonna watch a sequel?
1
0
0
5/8 You can use different question types (free text, linear scale, multiple choice, etc) to have a model create a dataset about your content:
1
0
0
4/8 . to re-use them with surveys whenever you want. It's good to start with a test question to check that your vision model isn't making things up:
1
0
0
3/8 You can share files publicly or privately:
1
0
0
2/8 EDSL has built-in methods for adding lots of different data sources to your questions, and a file store for keeping track of them:
1
0
0