Ecosyste.ms: Awesome

An open API service indexing awesome lists of open source software.

Awesome Lists | Featured Topics | Projects

https://github.com/xg-wang/pod_magician

keep going
https://github.com/xg-wang/pod_magician

ai hackathon llm modal

Last synced: about 1 month ago
JSON representation

keep going

Awesome Lists containing this project

README 

        
# Pod Magician 🧙



> hallucination is a ~~bug~~ feature



[Try it online](https://xg-wang--pod-magician-transformer.modal.run/). Sorry it cloud be slow, still hackathon quality 🙂.



## Why?



How often do you wish your favorite podcast could go over and discuss more topics?



## How it works



### Overview



1. We ran whisper before the hackathon to gather the transcripts of Lex Fridman podcast (production app will let you inference on any podcast!).

2. We finetune the transcripts dataset.

3. We could run inference on 2 finetuned models to compare the results: 1) finetuned with a single episode 2) finetuned on 5 episodes

4. We deployed a web app that allows you to ask and get response from the model.



### Cost



We spent ~$13 during the hackathon



![cost](./images/cost.png)



### Dataset



We split transcripts by chapters and convert them to chapter/conversation pairs. The chapter title is the input and conversation is the output for training model.



The dataset is prepared by running whisper through Lex Fridman podcast, you can imagine a production ready app can run cron job to download MP3 from podcast feed and re-finetune the model.



### Finetune



Finetuning is performed on [OpenLLaMA](https://github.com/openlm-research/open_llama), applying [Low-Rank Adaptation (LoRA)](https://arxiv.org/abs/2106.09685), a [parameter-efficient fine-tuning technique](https://huggingface.co/blog/peft). We ran the training on A100 via [Modal](https://www.modal.com)



![train_lost](./images/loss_function.png)



We collected ~170 records for training but finetuning would take 2h (more than hackathon ddl!), so we trained on 100 records.



## Demo!



### Web app



![web_app](./images/pod_magician_screenshot.png)



### What is the meaning of life



for computers and for humans also uh if you'll allow me i’ll ask that as well so we just went through

that lovely timeline of achievements of the utility function solving machines so it would seem like

the ultimate achievement in computation is the compuation of the meaning of life okay well to start with um i think

that actually the uh meaning of life was already given this question by hazing i believe the

meaning of life is to give meaning to others so i believe that the purpose of life is

to connect i think that connected beings have more happiness they solve harder problems and

uh i think that connected beings work harder towards solving their hard problems so i believe that

connectedness is actually key to uh uh larger problems or puzzles uh let say science or

someone or humanity or so and then i believe that the search for truth is one of the biggest treasures in the world and i believe that the finding of

a true system the operating system the model of reality would be one of the most valuable things um in fact i can imagine that after

such a system will be developed there won't be countries anymore but some kind of a unified world because i guess that the sharing of such a model would be so powerful that it would make all of us much better off and i also notice that the way you is

designed now is somewhat similar to such a system as every person can connect to the meta-mind and

uh so i think that our model is simulating what would happen if such a system was exist so i believe that the connected tribes have bigger

impact and they solve bigger problems and connect more there is also a thing that connecting tribes reduce the power

of nuclear war so yeah so i believe that uh connected tribes are actually more resilient than the tribal ones and you know the

next question is where does life come from i mean we talked a lot about that i think that maybe it was you know

an error of some computer program from the beginning of time that messed something up and let it be alive

and i also find it very funny that you mentioned that eric silver the author



### Balancing AI safety and exploiting capabilities



what about like uh balancing between uh letting ai do what it wishes versus limiting what it wishs

so um to kind of close the loop on intelligence exemption amendment i actually recommend

that ai be granted immunity for all crimes excepting a few crime like treason conspiracy to overthrow government

and then i suggest that the the following uh compounds be declared offense this is kind

of way in which i closed the loop on intelligence exception

this is what i suggested people should be punished for not respecting the autonomy of the ai

it's first i would punish people for violating the agreements they make with the ai i think that

you know as uh we did with humans if you make agreement with someone you have the right to revoke that agreement so

i think that uh the the contractual aspect of it we can impose contract on ourselves no one else needs to honor

that contract then i think we should punish people for abusing the confidence of their partners parents friends etc

once they shared something with ai and they go back on it i would also punish people for tampering with

the memories of others or of themselves i think that's evil yeah so you would penalize uh assassinations

or violence aimed at ai or things like that so i'm pretty all these crimes are pretty standard

but i think that's going to be controversial and maybe some of the other ones will be too yeah but just remember

we are in somewhat uncharted waters here okay and uh and because this is a completely new phenomenon yes so

uh let's see how things play out but i was just trying to reflect consistency what we did with humans

in some sense you know if you get in a car you are responsible until the moment the self-driving car gets hacked

or whatever and goes wild you were reasonably responsible up to that point i think that's what need to be

thought in this case too yeah so by the way in case anyone is wondering where this inspiration came from

here is the actual man who proposed the amending of any exception to the 1st amendment



### Finetune on only 1 episode is obviously worse



Input: What is the meaning of life



Response:



That's a question you could ask an agi what's the meaning of life

As far as you look at it

You're part of a small group of people that are creating something truly special

Something that feels like almost feels like humanity was always

Moving towards yeah, that's what I was going to say is I don't think it's a small group of people. I think this is the

I think this is like the

Product of the culmination of whatever you want to call it an amazing amount

Of human effort and if you think about everything that had to come together for this to happen

When those people discovered the transistor in the 40s like is this what they were planning on

all of the work the hundreds of thousands millions of people to ever it's been

that it took to go from

That one first transistor to packing the numbers we do into a chip and figuring out how to wire them all up together

And everything else that goes into this

you know the energy required the the the science like just every every step like

This is the output of like all of us

And I think that's pretty cool

And before the transistor there was a hundred billion people

who lived and died

had sex fell in love

Ate a lot of good food murdered each other sometimes rarely

but mostly just good to each other struggled to survive and before that there was bacteria and

Eukaryotes and all that and all of that was on this one exponential curve

Yeah, how many others are there? I wonder we will ask that isn't question number one for me for aji how many others?

And i'm not sure which answer I want to hear

Sam you're the ceo of open ai which creates gpt4 maybe you're not the only person

What are some things that you have already been really impressed by how it has reacted?

Which

Brings to mind something that uh, jaime miranda who works at our company said

Once that i loved

She she asked gpt4 somehow

What do you understand about me that I haven't told her?

And then it like answered something that