Even if AI can write perfect code, engineering organizations will still be as inefficient. At Entelligence, it's our goal to solve that.
Excited to have won 2nd at the
@agihouse_org
Enterprise AI Hackathon sponsored by
@AnthropicAI
and
@coatuemgmt
!
📷
Jensen Huang said engineers of the future will primarily be orchestrating a group of agents to build their visions.
Well that’s what we’ve built. Dropping our new OpenArchitect crew of engineering assistants from the
@MistralAI
SF hackathon this weekend in a few days! 👀
Built. Shipped. Delivered.
Now you can have full “architect” control over your crew of swe agents
- Describe what to build
- Discuss the best way to build it within your code
- Create tickets
- Agents build!
Thanks team and
@cerebral_valley
@SHACK15sf
best hackathon to date!
Jensen Huang said engineers of the future will primarily be orchestrating a group of agents to build their visions.
Well that’s what we’ve built. Dropping our new OpenArchitect crew of engineering assistants from the
@MistralAI
SF hackathon this weekend in a few days! 👀
In honor of our 1.5 year anniversary, the NLP Reading group is in full swing for the rest of 2022 - join us for
- Aug - Sept: LLM limitations
- Oct: Open Dialogue systems
- Nov: Conditional information retrieval
- Dec: Fact verification
- Jan: Open Domain QA
@ml_collective
Another little plug- if you're interested in getting started with NLP research or just want to hear more about the various interest groups part of
@ml_collective
please check out our site! and join the Town Hall we're hosting tmrw at 10 am PST! :)
Had a great time at the
@AGIHouseSF
applied research hackathon this weekend!
Great presentation from
@rm_rafailov
on DPO and great to meet
@JvNixon
@eraqian
and so many others 🙌
Also excited to have won 2nd with Entelligence! 😂
Work at big tech? Hate (like everyone else) the perf review/ feedback cycle?
True to our vision of being the first AI software engineering mentor, we're launching early beta for our mentorship agent
👩💻 Continuous feedback on your PRs
🎯 Goal setting
🌱 Proactive guidance
MLC has been the best community to bridge NLP research and engineering over the last few years.
We're working on the ICLR blog track for the next month - if you're interested in this space, we'd be happy to have you!
Super happy to see it keep growing 💜
Excited to share our latest product launch at !
Engineering is not just about coding skills - we empower you to ramp up and become a domain expert on any codebase and provide tailored feedback on every PR.
Curious how LLM's store facts? How to edit facts ? How to erase facts or teach it random facts?
Check out our latest presentations - Locating and Editing Factual Associations in GPT and Knowledge Neurons in Pretrained Transformers. Join to learn more
Had a great time at General Catalyst meeting so many amazing founders and their startups! Thanks
@AlexReibman
for putting this together and glad to be a part of such an inspiring group 🙌
1/ Entelligence
AI agent that generates execution/AST diagrams, creates code reviews, and writes pull requests
Upgrading the on-ramping process for junior engineers with AI
@Aiswarya_Sankar
Watching the agents you built autonomously working through all the new tasks you give them is so oddly satisfying.
Much more to come from our autonomous crew of ai software engineering agents! 🦾
In the future AI Engineers won't replace us. Instead, we will orchestrate them!
And that's what we built this weekend at the
@MistralAI
@cerebral_valley
hackathon!
· Spin up devs or reviewers
· Define tasks for them on a board
· Check their code on Github
and more coming ⚡️
LangGraph 🕸️ might just be the next shift to enable much more robust agent architectures through the use of a
- Pregel graph structure
- channels
- topics
- nodes
- more!
Check out our deep dive teaser video to understand LangGraph from the code up - longer video coming soon
🦜🕸️LangGraph
We introduced LangGraph last week, but are excited to launch it more officially today (blog and YouTube series). It includes:
🤖Example of replicating the LangChain AgentExecutor
💬A chat agent executor (aimed specifically at chat models)
LangGraph makes it easy
We've put up a detailed breakdown of how Langgraph works *under the hood* covering
- a visual overview of main components
- a breakdown of the graph algorithm
- memory
- channels
- core examples!
Youtube: …
Try it out for yourself at entelligence!
🦜🕸️LangGraph
We introduced LangGraph last week, but are excited to launch it more officially today (blog and YouTube series). It includes:
🤖Example of replicating the LangChain AgentExecutor
💬A chat agent executor (aimed specifically at chat models)
LangGraph makes it easy
Its amazing that most researchers publish their code along with their papers. Great way to ensure reproducibility!
However, I hardly ever had the time to read both the paper AND the code.
For the next week, using Entelligence will be going through my favorite papers + code!
The feeling when you speed up an endpoint from 1 min 40sec to 45s to 10s! And you want to call it a night... but then you realize somehow (willing all the forces of nature) you gotta go for another 10x speed up 🤯
Looks like it's time to channel more AWS credits my way
We'll be starting off with Monitors for Code -
understand how this paper integrates an AST when doing next token prediction for code
- theoretically
- through the code
This is part of our effort to be the simplest way to understand research codebases and papers!
Its amazing that most researchers publish their code along with their papers. Great way to ensure reproducibility!
However, I hardly ever had the time to read both the paper AND the code.
For the next week, using Entelligence will be going through my favorite papers + code!
Forget fine tuning... have you tried merging models to improve performance by "merging" model weights through
- linear interpolation
- SLERP
- TIES etc..
We visualize the mergekit repo and allow you to dive deeper into the implementation of these approaches with Entelligence!
Learn to merge LLMs with mergekit 👀
- Works surprisingly well, leading to multiple SOTA models
- Learn about SLERP, TIES, DARE, and Passthrough
- Share your merge with others!
Entelligence now supports full time series PR search!
Codebases change and evolve Every Day with new features (and bugs 🐛 ) introduced daily. Searching through these changes used to be a super manual process of going through git and reading PRs one by one 😞.
Not any more :)
Hiring my first 3 interns for dbrief.
It's one thing to have the flexibility and ability to work on a problem, design, build, iterate that you're passionate about - its an entirely other feeling to be able to share that with others and hopefully mentor and see them grow with it.
And... we're live! Check out our beta launch at - First sign ups will get access to pro features for free 😁
We're the first news search engine to put the news in context! No longer believe the first tweet, post or article you read on an issue blindly!
Me: Works on search, recommendation and ranking at work.
Me: Works on NLP for fun and for dbrief.
Me: Now gets to work on search, ranking, rec and NLP for dbrief 🙂
ColBERT and late interaction models are a game changer for RAG *reranking pipelines*!
Understand the real magic behind
- What is a LateInteractionModel?
- How does ColBERT outperform GPT-4 with RAG reranking?
- How to get started!
Great analysis
@virattt
!
Faster RAG re-ranking with ColBERT
After re-ranking using GPT-4 yesterday, I tested out ColBERT for re-ranking today.
Test:
• Re-ranking Airbnb's 10-K, like before.
Results:
• ColBERT and GPT-4 were identical in ranking quality
However, ColBERT was lightning-fast.
The pace at which startups and open source codebases change is insane 😱
Stay on top of everything happening on your team without digging through PRs or JIRA tasks with Entelligence AI's new team stand up!
We're launching our new AI mentorship agent!
👩💻 Reviews all of your PRs
👩🏫 Gives you detailed line by line feedback
and also
📈 Analyzes code across all your PRs to identify consistent areas for growth
🌱 Helps you hit goals
Please DM me if you're interested!
Codebases change and evolve Every Day with new features (and bugs 🐛 ) introduced daily. Searching through these changes used to be a super manual process of going through git and reading PRs one by one 😞.
But now Entelligence now supports full time series PR search!
In large organizations staying up to date with all the changes happening in a codebase can be HARD.
At Uber, it would drive me nuts trying to keep track of all the changes others were making in the service I was responsible for maintaining.
changes that.
For founders early on in a new feature release or product launch, I'd recommend using Microsoft Clarity! Great way to gather insight from previous user sessions and to better understand how intuitive or not certain features are in your product.
Actively gathering learnings!
My theory - no company is gonna pay you what you’re worth. Your ability to carve out some percent of market share is a better indication of what your talent is worth. Why accept your corporate valuation?
Mistral is 🔥! If you want to actually understand how Mistral works from the open source implementation, checkout our walkthrough by the Entelligence AI platform -
- What's the Mistral model architecture?
- How are the experts combined?
- What is the gating mechanism?!
- more!
How much time browsing the news / reading Twitter actually results in .. anything .. ? What if we could actually track, trace and quantify real impact from the collective hundreds of thousands of hours spent amassing information
@marktenenholtz
If you're looking to get a good understanding of the magic behind ColBERT and late interaction models, we've put up a walkthrough of the codebase with full diagrams and an in depth explanation of the key features!
- What is a LateInteractionModel?
- What is a negative miner
- ..
Graphs consist of channels, topics and memory -
Really neat how langgraph adopts practices from message queues such as Kafka
-channels can have topics which can have multiple consumers
-context channels can only be consumed by one consumer
Youtube:
A key component of our natural language to code diagram generation flow has been mapping "Draw a diagram about ___" to a particular node in the code diagram tree
Often times there's no match. Experimenting on whether the bttr UX is clarifying or zooming out to the closest fit
If you haven't joined us yet and are interested in checking out what we do - here's our latest presentation on Data Distributional Properties Drive
Emergent In-Context Learning in Transformers by Akashe Kumar
More info can be found in our discord channel!
Think I need to hard wire in my mind soon that getting work done through delegation >= getting everything done yourself since the first had so much more possibility for leverage and scaling in the future. Right now given the option to just makes me want to do it faster.
Mentoring others and seeing "room for improvement" turn into actual improvement is actually so rewarding. Makes me appreciate those who are willing to do that for me so much more as well. Keep it rollinggg...
@llama_index
@bclavie
Great work
@bclavie
and
@llama_index
!! For those of you looking to get a good understanding of the magic behind ColBERT and late interaction models, we've put up a walkthrough of the codebase with full diagrams and an in depth explanation of the key features!
One of the best things about being a true tennis player at heart is other's compliments or lack thereof hardly matter at all. All that really matters is in your belief and ability to out train, play and compete when it matters most.
"Information is the currency we have to inform ourselves to make the best decisions in our lives".
We're so used to a chaotic, overwhelming, one sided view on things it's challenging (initially) to even envision what this state of knowledge looks like.
Not for long.
The free flow of ideas and energy at Skydeck is actually so energizing - moving up to Berkeley for the next 6 months. Lmk if you're in the area and are free to meet up! Would love to chat
@marktenenholtz
For those of you looking to get a good understanding of the magic behind ColBERT and late interaction models, we've put up a walkthrough of the codebase with full diagrams and an in depth explanation of the key features!
Honestly what a time to be alive (from tech perspective. So many things I literally thought were impossible last year are at our fingertips now with so much more possible
Super excited to make this the best internship opportunity I can for them!! Might sound cheesy but it's no longer the product, but making sure this is the best experience possible for them.
The best design docs, scoping, code reviews, collaboration and team vision setting 😄
📊 Quickly summarize all PRs submitted by your team
🔍 Deep dive into changes made in any module
👩💻 Understand all the changes any engineer made recently!
Destress and refresh. Also for full transparency I’ve been out of the competitive tennis scene for the last two months due to a busted shoulder. While it’s been hard mentally recovery has been going well and I’ll be back out there as soon as it’s back in action.