LLM Meetup Tokyo #1
takahiroanno I went to an LLM meetup at a place in Tokyo and it was great. It was a highbrow meetup where all participants were required to bring a demo related to LLM to participate, but about 50 people showed up. The participants were distilled down to the bare minimum, and there were a lot of LLM unemployed. It felt like the beginning of an era to me.
-
takahiroanno LLM unemployed = people who unintentionally quit their job after seeing how great LLM is (including people who recently started their own business)
-
takahiroanno This is the episode. They’re going to do it again next month, so I’m hoping for the best.
Regulation
.
This is a small meetup only for those who are actually using LLM through API, fine-tuning, creating, etc., and are actually working with LLM in a hands-on way. Only those who can introduce themselves & give a demo during the demo time in the second half of the event are welcome to attend.
self introduction
.
- The Intellectual Production of Engineers
- Technology Supporting Coding
- Natural Language Processing with word2vec
- Seems like a long time ago.
- 3.5
- ChatGPT API
- 3/9 Scrapbox ChatGPT Connector
- Participants’ Tweets
.
@kenn: LLM Meetup Tokyo was so much fun! Thank you everyone!
@elmer_mmm: Here is the video of the robot running onChatGPT that I showed in the demo at LLM Meetup Tokyo #1! I couldn’t hear the sound at the venue, so please check it out! I also put him on a “kachaka” and made him speak multiple languages!
@yamaguchi_man: LLM Meetup Tokyo was great - I learned a lot from the various things people are trying to do in the search area! In our case, it’s a combination of rule-based and other things that we’re trying to do.
_mkazutaka I attended LLM Meetup Tokyo! It was fun! It was great to be able to talk directly with each other about what’s going on with the prompts!
wroorurkwao At the LLM meetup, I wondered what competitive advantages remain? We talked about things like.
y_matsuwitter Material from today’s LLM Meetup Tokyo #1. I was experimenting on Saturday and Sunday to see what would happen if I tried to create an Agent with a lightweight LLM that can run on a MacBookPro.
-
y_matsuwitter Even though it is a small model, ReAct’s Prompt is English-based, so I think it would work better if I focus on both categories and tasks and Fine-Tuning. I’d like to try Fine-Tuning on Colab with GPT-3 data next.
erukiti y_matsuwitter wearing OepnAI hoodie next! “Agent and Lightweight Model Validation”
-
erukiti “I started a team because I want to do LLM (myself and my ML team)”
-
Powerful and
-
erukiti “I want to make Japan LLM as a national policy”
-
“Large companies are surprisingly reluctant.”
-
erukiti “I’m sure many of the people who come to LLM Meetups are interested in an agent that works independently.”
-
Yes
-
erukiti “If you combine LangChain and agent too much, it will take too long to get a response.”
-
erukiti “The main question today is, can we ReAct with LLaMA.cpp or Alpaca or something?”
-
“Bottom line is that with the current free lightweight model, it’s impossible.”
-
Vicuna 13B 4bit and RWKV also seem to be impossible
-
erukiti Personally, I think we should generate a lot of datasets for ReAct using GPT-4 and fine-trune them. I think so.
chiral Here are the slides from today’s presentation. I know it’s a bit shabby, but it’s just to show that I’m implementing it too. drive.google.com LLM Meetup Tokyo
#1Masayuki Isobe.pdf
yakigac Slides presented at LLM meetup↓. https://docs.google.com/presentation/d/1GqrGrd5aV3HxT_lNryOwTGpOaKsiDj-88jnV17QqjoM/edit?usp=sharing…
masa_kazama I gave a LT at the LLM Meetup titled “Differentiable Search Indexes”. I’m looking forward to new search methods in the generation era that are different from keyword and vector searches.
-
masa_kazama The details of the method are very clear in this explanatory article.
-
@AkihisaShiozaki: Sam Altman, CEO of OpenAI, visited Japan to attend the LDP PT on AI evolution and implementation. Citing the active use of ChatGPT in Japan, he said, “I hope Japan will demonstrate a large presence and leadership in the world through the utilization of AI. The following seven proposals were made with expectations for Japan. 1…
-
@overlast: all seven of these proposals are proposals that would reduce the technological capability of Japan’s domestic 4 years into the future. And the force of the reduction is different for each one. I would like to see something like these proposals implemented in Japan. And if OpenAI implements the proposals as a for-profit company, and the Japanese government and municipalities do not give preferential treatment to each measure, then it is simply convenient, so it seems to be OK!
chiral Preparations for tomorrow’s LLM Meetup Tokyo. We’re preparing for tomorrow’s LLM Meetup Tokyo. We’re going to have two gorgeous (but not too expensive) sessions: Help Desk AI (EmbeddingsAPI,Weaviate,CompletionAPI, hosted by docker-compose on in-house PC) and LINE Chat Divination AI (LangChain,Flask. ngrok, hosted by colab, and connected to LINE MessagingAPI)! ngrok to connect to LINE MessagingAPI)!
tdualdir I’ll join you here! I will demonstrate automating table data preprocessing using GPT! (Maybe on MatrixFlow in the future.)
@erukiti: https://t.co/VWJnFLhgia Slides presented at LLM Meetup Tokyo #1
npaka123 I will attend LLM Meetup Tokyo next week.
2023-04-10 LLM Meetup
This page is auto-translated from [/nishio/2023-04-10 LLM Meetup](https://scrapbox.io/nishio/2023-04-10 LLM Meetup) using DeepL. If you looks something interesting but the auto-translated English is not good enough to understand it, feel free to let me know at @nishio_en. I’m very happy to spread my thought to non-Japanese readers.