AI Demo Days @
@awscloud
- we did it again! ๐
Packed house, apparently the biggest event they've ever hosted there? wild. 120+ builders and investors all crammed in to see some seriously cool demos.
Quick rundown of the awesomeness:
Introducing Code Review GPT: Your personal code reviewer powered by LLMsโก๏ธ
Improve code quality and catch bugs before you break production ๐
How does it work?
๐งต ๐
Spoiler: we used
@LangChainAI
from
@hwchase17
and it's incredible.
i've been an AI Engineer for about a year now - these are the 5 top skills I've needed so far.
- Devops, I can get something out much quicker if I can serve it myself. New platforms like modal make this a million x easier
@modal_labs
@charles_irl
... but learning how to whip up
Big updates to the internal prompts and how we supply code to the LLMs in Code Review GPT
@oriontoolsai
80% more accurate responses and new cost-optimised review type.
Tired now, will explain all tomorrow
โญ๏ธ the repo if you haven't, it helps a lot :)
anthropic on fire.
reminder that nobody has any idea how they went from the last gen to this. Performance is so good.
looking forward to the rest of the year.
We've made a lot of cookbooks:
Tool Use
Calculator tool
Customer service agent
Extracting structured JSON
Using the tool choice parameter
Tool use with Pydantic
Vision with Tool Use
Misc
Building evals
Building a moderation filter
Generating test cases for an eval
How to enable
love it how many next gen cloud solutions start with:
Docker was too slow, so we fixed it.
@modal_labs
@Cloudflare
@mysticdotai
etccc
almost like Docker is not fit for purpose...`
congrats
@CloudflareDev
on the awesome release.
Fun to demo some AI stuff at the London
@supabase
meetup last night.
Rebuilt an app live in 15 mins
Prayer to the demo gods before I deleted the prod table ๐คฃ
Thanks to
@jacksbridger
for inviting me.
AI Demo Days at
@cohere
reflections:
(my smile gives it away)
4 demos
A full room
The excitement was infectious
Partnering with
@Cherry_Ventures
was awesome. We reached many more founders and companies in London building incredible things with AI.
cool demos (๐งต)
we hear from lots of builders who don't want to get locked into a single model so we're really excited about Genoss!!
plus, an integration with LangSmith for easy debugging, logging, monitoring, and observability!
.
@mattzcarey
on the LangChain blog
The Serverless London meetup group just hit 4000 members ๐
Congrats to
@EllerbyBen
@IamStan
and everyone who has contributed to the events over the years.
Next event is on Thursday :)
Cohere just dropped a
- 104B
- Multilingual
- 128K context length
- RAG + Tool Use
- Open weights
Model.
And based on how good their previous Command R, it should be REALLY good.
- Link:
- Try it here:
Almost touching 500 stars ๐
We are on a mission to make YOUR janky code cleaner, faster and more maintainable with
@codereviewgpt
Check it out here:
Share the love of quality code in one line:
โnpm i code-review-gpt && code-review-gpt configureโ
Using the brand new
@LangChainAI
Smith tool, you get this awesome dashboard ๐ซ
Keep track of your costs and responses from Code Review GPT super easily.
Playing around with the GPT4All from
@nomic_ai
. The possibilities of open-source low energy/memory use LLMs for data privacy and portability really are endless...
Host on IoT devices -> no problem โ
Run at the edge -> probably ๐คฏ
Tweak to meet your exact needs -> easy as pie ๐
@modal_labs
@charles_irl
lol that was 4.
5 is app development.
Contribute and read open source codebases. See what works or what feels grim.
Get the best person you can find to review your code. Japanese have this idea of a gemba walk. Pick their brains. Learn stuff.
Tool use and RAG using Command R+ at
@cohere
super cool demo also chatting about training for citations and generated tone with
@PSH_Lewis
now live on yt!
thanks to
@jeremyphoward
for the inspo. Was great
Most of the Fortune 500 still uses a mainframe!
@bloopdotai
is converting their legacy code to Java that runs in the cloud.
'Modernising the mainframe with AI' is now live on YT
awesome demo by
@LouisKnightWebb
at
@demodaysai
Building demos with LLMs is a breeze, but production?
It's a different game!
Need to juggle infrastructure, security, and compliance? ๐ค
Considered OpenAI APIs?
There's a new player,
@awscloud
Bedrock.
Wrote an article on the differences:
Great to chat to
@AllenHeltonDev
about
@quivr_brain
and
@momentohq
. ๐๐งก
If you use
@LangChainAI
and you should :) then here are the docs to add Momento to your app. ๐ง๐ฑ
Pretty nifty using a cache as a history store ๐ค๐ผ
One of the best things about Parliament Wow was the easy semantic search.
we used the bge-large-en-1.5 embeddings model on Workers AI.
I reckon
@CloudflareDev
have a grand chance to dominate the hosted embeddings model market if they expose a few more models. reasons:
1.
rag is hard
doing BM25 retrieval alongside semantics before applying a reranker
now I need to store my full docs somewhere legit
as well as my vectors
probs best practise anyway but jeeeeeezzzz the overhead
@TodePond
demo from AI Demo Days was more like 10 demos, each more wild than the last.
"Autocomplete for canvas" now LIVE on YT!
blows my mind what they get up to
@tldraw
@togethercompute
's new research paper on Mixture of Experts is super cool and well written
I just can't believe that this is the first anyone has written of this technique
we've been using exactly this internally inspired by one
@mattshumer_
collab scripts on llm researcher.
@PSH_Lewis
from
@cohere
demoed tool use and RAG using Command R+
bunch of takeaways from this demo and hearing Patrick speak:
Cohere went to great lengths to make sure their models default to using supplied context even if it conflicts with pre-training data - NOT easy to do.
Enjoying playing with
@turbopuffer
serverless vector store.
sreat scaling on demand and super cheap
try it with
@LangChainAI
using this VectorStore class
"I use serverless mainly for demos and to order coffee"
savage but thoughtful chat from
@ghohpe
on the place of AWS Lambda and Serverless.
dude left AWS like a month ago and he's firing shots left and right. Interesting to hear an alt take from the seemingly disillusioned ex
cloudflare workers = scalable compute heaven for RAG apps.
spawn workers for each tool, use KV store as cache, leverage durable objects.
don't pay for io comput time. keep your runway.
RAG at scale? sorted.
It gets all the changed files using this lovely git command:
`git diff --name-only --diff-filter=AMT ${baseSha} ${githubSha}`
And it builds a fancy prompt to tell the LLM exactly what to do ๐คฉ
one of my longer term goals is to build a hacker house in London within the next 5 years.
a kinda clubhouse. You can come stay, build stuff, visit the city.
not quite sure how it would work. There is a big difference between residential and commercial property here.
Thinking of organising a workshop to get people starting building with
@LangChainAI
and
@llama_index
in London.
Free for community!
Show how to build AI powered tools (Code review, Prompt engineer etc) like
@oriontoolsai
Might even have pizza if I can get sponsors. Keen?
@boristane
Add this to your eslintrc.js file to inforce this in linting
'no-restricted-imports': [
'error',
{
paths: [
{
name: 'aws-sdk',
message: 'Please use aws-sdk/{module} import instead',
},
],
},
]
it's super hacky and some bits are buggy, but goddamn it's so impressive for a weekend's work. we slept well, didn't stress ourselves, focused on shipping something that worked, and had such a great time
@LouisKnightWebb
Iโm starting an event series. Similar to what we have done with GenAI London but way more hacker. Iโll ping you for the first event!
Launching AI Product Engineering!
Demo evening for the London AI community :)
There are a lot of tech meetups in London. But not many like this.
15 min demos, no marketing, no bs.
Meet cool people, get inspired by slick products and cracked solutions.
2 hours later, it was an incredible session
I have a lot of notes and it's time for our team at
@baselimeHQ
to make the future of o11y for serverless we envisioned this evening a reality
huge thanks to everyone who came along ๐
Announcing the Momento Vector Index! A serverless vector indexing service enabling builders to launch their AI projects in record time! ๐
#mocon2023
#launch
#serverless
latest ml framework is out and I'm super excited. It's in zigggggg.
congrats
@gwenzek
@steeve
looking forward to seeing some pytorch models get ported to
@zml_ai
exactly why I am bullish on the zig language. C interoperability and build system goes so far.
Announcing a global collective for people interested in GenAI - GenAI Days!!
Bringing together a community of forward-thinking professionals, AI enthusiasts, researchers, and industry pioneers ๐
@_StanGirard
@EllerbyBen
@maxthoon
If the file fits in the context window of the AI all good ๐
(context window really is king ๐use 8k+ models to get the best performance)
If it is too big use a method similar to classic document retrieval ๐
Great to be a guest on
@theburningmonk
Real-World Serverless podcast.
Yan has some crazy good tech knowledge and it was super cool to share
@quivr_brain
and
@OrionToolsAI
I learnt a bunch of stuff and I hope you do too!
๐๏ธ Episode 81 of Real-World Serverless is out!
@mattzcarey
walks us through LangChain development, the tooling problems, the challenges with testing non-deterministic AI apps, and much more.
And he demoed both Quivr and Code Review GPT and they're on ๐ฅ
Been asked a lot about getting LLMs to generate tests.
Iโm certain this is the wrong way of looking at things.
As the human, you know what you want.
Wanna get test coverage?
Write tests.
LLM writes application code.
maybe based off tests
AI TDD
Mindset shift.
Building with GenAI
#3
is out!!
New open-source tooling announcements ๐คฏ AI Code reviews have come to Quivr ๐ซ
As always we have all your favourites... This week in numbers and the best of learnings from the team building Quivr ๐
Great to speak about
@OrionToolsAI
@quivr_brain
@genossGPT
at
@awscloud
User Group London.
Awesome conversations afterwards. People are making waves here ๐๐ผ
London is the place to be building with GenAI ๐
is there a change happening from experience-based hiring to skills-based hiring?
tech kids getting hired based on open-source contributions springs to mind.