I think we’ll end up understanding human intelligence by artificially recreating it. And we’ll realize that intelligence isn’t what made our experience of life so special.
In our lifetime we’ll be surpassed on almost all fronts of intelligence. And it will be completely fine.
Been seeing a lot of speculation on Twitter about why we moved to Paris and decided to hire across Europe, Israel & yes also still the US (we operate remote-first
@poolsideai
but come together every month at our office in Paris 🇫🇷).
Here is why...
VC: So you’re out raising
Me: No, we’re not
VC: okay, so you are not raising raising 😉
Me: No, really we’re not
VC: Okay, so fake not raising 😉😉
Me: No, *really* we’re not fake not raising, we are actually not raising
Europe has absolutely incredible talent but you need to build a remote team to harness it. You just can’t find the same geo-concentration of exceptional talent like in the Bay Area.
Can’t stress enough how often you find yourself in AI realizing
@ylecun
was not just right but was right earlier then others in hindsight. And yes, that in our opinion includes that auto regressive models are likely not the answer to advanced machine intelligence.
Stupid visa processes are stopping a very talented candidate from being able to have an internship at our Madrid office. Can't wait till the day people read about visas and borders in history books. We are all humans on the same planet.
Wanted to share my thoughts on the 5 types of metrics I believe you need as a software engineering team to know if you're continuously improving:
• Speed
• Volume
• Engagement
• Quality
• Impact
Thread👇
And why Paris 🇫🇷 and for instance not London?
@Xavier75
,
@an21m
, and
@EmmanuelMacron
's government have been incredibly welcoming and shown that they are committed to providing the conditions for AI companies to succeed here. And there is just magic in the air here!
Will share more soon but this quote from
@Redpoint
’s post is a great tl dr;
“Poolside aims to unlock humanity’s potential by pursuing AGI for software creation, with the fundamental belief that the transitional path for humanity to AGI is by building for specific capabilities
Couldn’t be more excited about what
@jasoncwarner
and
@eisokant
are building. Check out why
@ericabrescia
and I are thrilled to get to partner with the perfect team to take on a world-changing vision to power the future of software creation:
Thank you
@mattturck
for blowing up my DMs with questions after your tweet - hope this answers them.
And yes, we are fully aware that France has some things to still improve on but so does every place. It's the gradient under the curve that matters, and it is steep here. 🚀🇫🇷
Human-level intelligence is going to happen, and we think it’ll happen first in software development. Today, we’re staking
@poolsideai
's position in the field. AGI’s promise is too significant for me and
@eisokant
to stay out of the mix
a week ago I joined poolside, a new ML startup building catalysts for developers, w/
@jasoncwarner
@eisokant
. excited to share more over the next few months…
Big congratulations to both
@Semmle
and
@github
! Very smart move from
@natfriedman
. This will allow GitHub to become a very powerful global code graph.
Today we welcome
@Semmle
to the GitHub family!
We’re excited to bring the world’s most powerful semantic code engine to the world’s largest developer community🔥
We did our homework and saw that there are an incredible amount of talented deep learning researchers and engineers in Europe. Many of which who have been working with LLMs, RL, and foundational research at DeepMind, Google, NVIDIA & Meta but also at lesser known companies.
We're building a new index of almost all of the Internet's public git repositories (incl. self hosted) on a new bare metal cluster running the Enterprise engine. Currently 20% complete. Can't wait for it to hit 100% this quarter.
I have just found out the current estimate of the size of all the world's public Git repositories thanks to
@sourcedtech
's Data Retrieval team: 350TB. Based on the already cloned random 20% and counting.
For my first tweet storm, I wanted to share a bit about the
@sourcedtech
open-source stack and how recent releases and announcements fit into the bigger picture.
In the coming years we believe
#MLonCode
will start to drastically change developer tooling in the following areas:
We launched today with our first customers and we had a bunch of stuff break.
This is okay! We had told these customers they were the first ones.
People's default is actually friendly. Make your customers part of your journey and they'll want you to succeed.
By talking to them we realized that many of them wanted to work on ambitious AI projects but didn't want to move to the Bay Area or work at a secondary office in Europe - far away from the action.
This has gotten out of hand, journalists are now reaching out to our investors to confirm that we're raising X at Y valuation.
Whomever started this rumor, can you please let the world know poolside is not raising capital in 2023.
VC: So you’re out raising
Me: No, we’re not
VC: okay, so you are not raising raising 😉
Me: No, really we’re not
VC: Okay, so fake not raising 😉😉
Me: No, *really* we’re not fake not raising, we are actually not raising
Hilarious seeing all the Q* guesses. Some of them come very close to what we do at poolside and call Reinforcement Learning from Code Execution Feedback.
It's been 9 months since I became aware of the work
@EntropyFarmer
&
@Miau_DB
were doing in obscurity on tackling Artificial Intelligence.
This public release today sets new world records on many Atari games.
This thread tells part of the story:
So
@jasoncwarner
and I started doing a thing!
We're publishing a podcast episode every two weeks talking about challengea and lessons learned in Engineering Leadership.
And you'll find plenty of strong opinions 😉
A thread on some of our believes
@poolsideai
(the last one in the thread being the most important one):
1/ To push beyond current capabilities you need to pre-train your own foundation model
EXCLUSIVE! 🎙️🎙️🎙️
@poolsideai
is a very ambitious Generative AI for code startup that’s already raised $126M
For their FIRST long-form conversation, CTO
@eisokant
joined us on the pod for a glimpse as what he and
@jasoncwarner
have been building
👇LINKS IN NEXT POST👇👇
Jean was the sole reason I raised from NewWave, with or without it, I have zero doubts he’ll be in our corner. The reason you’ve seen countless of ❤️’s on Twitter is because Jean has had an immeasurable impact on the journey of so many founders.
End of 2020, while running Kima Ventures, I co-founded New Wave, an early stage fund to back ambitious founders.
Earlier this year, I was removed from my position by my partner, against my will.
I feel a responsibility towards our entrepreneurs, investors, stakeholders and
Lots of announcements today! We are now General Availability on our Enterprise edition
@sourcedtech
. And we're opening a private beta for our Community edition.
A lot of people have questioned why we wrote our own distributed training code base at
@poolsideai
.
Most people assume it’s either for compute efficiency or some misplaced “not invented here” syndrome.
The real motivation has always been to be able to iterate incredibly
When you start onboarding customers and your lead time becomes less <2 hours because you're shipping fixes and feedback so fast! Really proud of our team at
"Software engineering can be the most creative, satisfying, and fun job. It’s like getting to solve puzzles and getting paid for it."
Probably one of the best lines ever by a guest on the podcast. 👌 had to be
@mipsytipsy
✨
Dropping episode link in the thread 👇
Very exciting news!
Kythe/Grok powers all static analysis, code search and navigation at Google, Michael joining us after leading the team for 4 years (and 8 years on it!) brings a wealth of experience.
More so, it adds one more kind-hearted and brilliant member to our team.
We're excited to announce that
@creachadair
, former senior software engineer at
@Google
, and technical lead/manager for the
#Kythe
open source project, is joining
@sourcedtech
to lead Programming Language Analysis efforts
Small FYI: unfollowed everyone today and am starting from scratch (refollowing). So if you see me follow you today - odds are I have been following you for many years
OK, here is my best guess on the state of LLMs:
- The scale increase between gpt-3 and gpt-4 was 100x
- Doing that for the next model is going to be very hard
- We're nearly out of general language tokens. So let's say we can 2x that. And perhaps get more proprietary tokens and
Jean has not only been one of the best human beings I’ve had the privilege of knowing and working with for 8 years. He’s also single handedly picked some of the best investments at seed stage when they weren’t obvious to others.
There are a lot of us as founders who hold him in
I am very excited, and also humbled to be able to announce
@vcoisne
joining the
@sourcedtech
team!
Victor will be our head of developer community, working with
@francesc
, ensuring the
#MLonCode
community ALWAYS has a voice in every discussion and decision.
We're hiring for 11 different roles at the moment
@sourcedtech
! The latest one is for a new team we're creating: Solutions Engineering Team.
Our # of customers is growing and we're looking for engineers who will build on top of the source{d} stack.
For 3+ years I've saved every piece of content I've consumed. This archive has grown to 1000s of articles, podcasts, Tweets and videos.
I have decided to start sharing from it, in particular the content that is timeless, with my own commentary added:
Engineering Leaders: What's the single biggest mistake you see in the hiring process for engineering managers, directors and eng. leadership positions in general?
As a baby they explode, as a teenager they ignore your instructions and give you stupid answers, in their 20s they get their own opinions that you might not agree with, in their 30s they start becoming your friends.
...so much fun to train foundation models 🤣
@an21m
@newwavevc
@amoamoamo
❤️ kind hearted + ambitious + highly talented + optimistic about the future is what you find in Paris. You used to find this in extreme abundance only in Silicon Valley. After all of the investments in this ecosystem no longer make that true (especially by
@Xavier75
,
@2lr
and
I am really excited to announce that today we launched Teams in
Teams have been a much requested feature, especially as we continue to get larger organizations adopting Continuous Improvement.
When customers come to us with feature requests, we think:
- What underlying problem are they trying to solve?
- Do we see other people asking for it?
- Is it something that matches our vision of the product?
Calling on everyone in my Twitter to do be your most critical.
We're launching a live demo of Athenian that you can play with on our data.
I'd love for you to let me know about your experience.
This morning I had a call with a seasoned engineering leader at a large SaaS company who articulated so well why becoming data driven was key to their engineering org.
Some of their quotes:
In Paris 🇫🇷 to meet amazing talent!
If you can you name 5 reasons why loss would explode during training? And you...
- Read papers for breakfast
- Really know how to code
- Very strong in LLMs or RL
- Think about systems as a whole
DMs are open!
Hey folks, since we've been asked so many times, here is a quick demo of what we're building.
This is a small LLaMA2 sharded on 1 NVIDIA RTX 4090 (in Paris), 1 AMD 6800XT (in
@TheCorendos
's flat) and 1 Google Cloud TPU v2 over
@Tailscale
.
Exact same code, all built on my Mac
"Software engineering can be the most creative, satisfying, and fun job. It’s like getting to solve puzzles and getting paid for it."
Probably one of the best lines ever by a guest on the podcast. 👌 had to be
@mipsytipsy
✨
Dropping episode link in the thread 👇
I believe that program synthesis will solve reasoning. And I believe that deep learning will solve program synthesis (by guiding a discrete program search process).
But I don't think you can go all that far with just prompting a LLM to generate end-to-end Python programs (even
It's a real pleasure for the data retrieval team
@sourcedtech
to wake up to a compliment from
@jessfraz
! Thank you for voicing it, it's much appreciated.
Prediction for 2024: by the end of the year context windows will be effectively infinite (i.e. 1 billion tokens). The world is solving for this right now in multiple different ways.
I appreciate you America! If it wasn't for your entrepreneurial spirit and willingness to encourage the crazy ones, the world wouldn't move forward as fast.
Starting to hire the first enterprise sales reps & sales engineers at
@sourcedtech
because the inbound marketing engine that
@vcoisne
has been building has too many companies in it to just handle ourselves.
It's really exciting and humbling to see the company grow up like this.
I am very excited that this dataset is now public! Replicable large-scale static code analysis and
#MLonCode
just became a lot easier (and a lot faster to get started, no need for git cloning 100,000s of projects anymore).
Announcing Public Git Archive () for static analysis and
#MLonCode
3 Terabytes of GitHub repositories (all projects >50 stars) with meta data, stored as rooted repositories, and easily downloadable over HTTP with the pga tool ().
Spend your life surrounding yourself with kindhearted people to be happy. If you feel the urge to change the world, find kindhearted, ambitious, resilient and highly capable people. Just keep doing this over and over again, and it will compound.
So many things to unpack here:
1.
@an21nm
is far too kind and generous in his words, it is a privilege to have kindhearted ambitious friends. Also the
@amoamoamo
office is 🚀 and you’re not working there alone at midnight - that team is a machine
Our friend
@eisokant
has been working out of the
@newwavevc
&
@amoamoamo
office for the last few days and I couldn't be more bullish on any company and vision right now.
In 48 hours I've seen him persuade some of the best in the world, discover that our lights turn off
Today got a detailed one-page email explaining why a company is not yet ready to become a customer.
When you have a user willing to give you a detailed roadmap of feedback, it means they want you too succeed.
These are my favorite customers to serve.
You have to be willing to own mistakes.
But those mistakes should reflect on the process, not the person.
If I bring down production, a high-performing org says:
→ What was wrong in our process from the beginning?
Not, “Oh, he’s an idiot.”
Impressive work by our team!
Read our announcement and it will become clear this is only scratching the surface of what
#MLonCode
will do for code review.
We're thrilled to officially announce 📣 the release of a new format analyzer powered by
#MachineLearning
in
@sourcedtech
Lookout, our brand new assisted code review framework. 🎉
#MLonCode
We have an absolutely stellar meetup coming up this October 19th @ 6:00 PM!
@PfeiffJo
from
@GoogleDeepMind
will be presenting followed by
@eisokant
co-founder and CTO of
@poolsideai
.
Last meetup we ran out of spots! RSVP:
@dakami
@jessfraz
@francesc
Wait till you get to run this over Public Git Archive, This is still in alpha and not performant yet (but will be soon). SQL based querying over git and xpath over Universal AST's including high-level semantic concepts e.g. functions, imports etc.
Any best practice you believe is specific to remote…
…is, in reality, a best practice you should already be doing.
Communication is at the root of most problems.
When your team is in the same physical space, it’s easy to overlook it.