Solve Synthetic Biology.
Solve BioMachine Interface.
Solve Fusion / energy gen.
Unify Mankind.
Solve Warp (optionally other interstellar travel).
Spread Humanity & its proxies.
Build New Worlds.
Find, accelerate, and invent new life.
Understand Existence.
This is the path.
The
@fastdotai
library from
@jeremyphoward
@math_rachel
@GuggerSylvain
et al is incredible. I finally got the chance to take it for a spin.
Completely new user. <40 min of coding, I had a CNN that tied the best published bacterial classification model. 97% acc.
Astonishing.
"The Matrix Calculus You Need For Deep Learning" by
@the_antlr_guy
and
@jeremyphoward
is phenomenal. Check it out if you want an accessible peek at the math you need to understand the nuts and bolts of DNN training:
For a while I've been thinking about learning Math from the ground up, since it's so cool and also important for the things I'm building and researching.
I wrote about how I'm planning on taking the MIT undergrad applied math curriculum and it blew up!
This weekend, a few friends and I competed in the
@SouthPkCommons
and
@OpenAI
Hackathon.
We built P.A.L. a framework that allows you to build proactive AI agents w/ modern LLMs.
We won the 𝗠𝗼𝘀𝘁 𝗔𝗺𝗯𝗶𝘁𝗶𝗼𝘂𝘀 𝗧𝗲𝗰𝗵𝗻𝗶𝗰𝗮𝗹 𝗔𝗰𝗵𝗶𝗲𝘃𝗲𝗺𝗲𝗻𝘁 𝗮𝘄𝗮𝗿𝗱 🥳🎉
2/ PAL (proactive agent library)
Agent orchestration framework for continuous learning on new data streams and provide time-context relevant completions
🏅Most ambitious technical achievement- $2000 OpenAI credits
@harsha
Really incredible talk by
@jeremyphoward
on the future of individualized medicine. Useful quick history of DL as it pertains to health as well, give it a listen!
I hear absolutely incredible things about
@ml_collective
and
@savvyRL
from colleagues in ML Research at different stages of their careers.
I haven't had a chance to get involved yet, but hope to spend some time there soon. We need more initiatives & groups like this.
I jumped inside a car-robot with lasers for eyeballs, piloted by a simulated brain emerging from electrons dancing across a silicon labyrinth.
How can you not be excited about the future 🚀
Master Plan:
1. Heterogeneous Learning Systems and Infrastructure for Intelligence
2. Tooling for Synthetic Biology and Wetware Computing
3. Longitudinal Health tracking + fundamental aging preventions research
4. ?? Some ideas but we'll see if I live long enough to get here
Super pumped to join
@openminedorg
as a Research Scientist! I’ll be working on a privacy + systems projects, including applying NAS to a federated setting. Definitely reach out if you’re interested and also check out the community over at
@iamtrask
This probably isn't what you're thinking, but may prove surprisingly fun.
Try sitting somewhere quiet, closing your eyes, and seeing if you can remember distinct fond memories from every year of life so far (even 1 is good). It reinforces how beautiful the journey has been.
y’all arguing over work ethic is boring. u should cycle dif amphetamines and never sleep, aiming for continuous dopamine drip feed to the brain whilst reading every word of the internet.
once achieved, move to log cabin in german forest & do math all day, weekends at berghain
As I hop in my
@Waymo
to head into
@solarissociety
for back to back
@ManifoldRG
AI research meetings, it dawns on me:
San Francisco is now home.
Long live the great city of progress 🌁
It was a wonderful privilege to get to ask
@danielgross
a few questions yesterday.
Daniel shared a lot of nuanced perspective on the future of AI, the rate of progress, and how to build great products and companies in such a special time. The amount of insights per minute was
Thanks to
@danielgross
for joining us yesterday for one of the best fireside chats we've hosted in a while (and we host a lot of talks). And thanks to SPC Founder Fellow
@HarshSikka
for moderating!
A small, personal goal of mine is to build a personal lab. A few features:
A computing cluster with a bunch of different compute stacks (including a ton of GPUs)
A robotics bay
A lab bench for synthetic biology stuff
A mattress (bedrooms are overrated).
Building humanity's technology tree to better understand the Universe.
@ManifoldRG
is the spiritual successor to Bell Labs, built for the 21st century.
Sounds like we have friends in
@CityofPraxis
⌘-R
Introducing Command-R, a model focused on scalability, RAG, and Tool Use. We've also released the weights for research use, we hope they're useful to the community!
I want to see more ambitious work. Are there any folks whose blogs or twitters I can follow that are working on absolutely insane projects? Any discipline/field/vertical is welcome. Things that would make my jaw drop, like the projects you see on
@pioneerdotapp
@ericjang11
@lucy_guo
Eric, tbh I don’t think being a rising AI & Robotics research star at one of the largest and most impactful research orgs on the planet is necessarily complacent.
Maybe if you were working on some button in a mobile app for a year or something.
I always though Robotic Process Automation had to do with actual robots.
Turns out it’s just a bunch of if then statements.
My disappointment is immeasurable and my day is ruined.
Just applied to the AI residency position at
@Theteamatx
. They sent me some resources, including a Guide to ML Mathematics.
I clicked it and was directed to my own post!
Being directed to my own work to review for interviews is a first for me! 😂😅
We’ll be presenting our work “Investigating Learning in Deep Neural Networks using Layer-Wise Weight Change” at the NeurIPS
@NewInML
workshop tomorrow!
If you’re free, come on by! We’d love to discuss our work and answer any questions you have!
Tune in to our Oral Presentation tomorrow in the
@NewInML
workshop at
@NeurIPSConf
from 10:15-11:00 a.m EST. We will be discussing the Learning in Neural Networks using Layer-Wise Weight Change. We are really excited to share this work, please stop by!
This is simply one of the most compelling call to actions I’ve seen.
Hats off to the
@openminedorg
community for bringing together such an amazing resource and paving the path to a future where anyone can learn the tools necessary to help solve humanity’s biggest problems.
What are the best online communities focused on deep tech, moonshots, and hard science?
I want to systematically help as many people as I can, and I know a bunch of other people in various groups (
@genzmafia
@pioneerdotapp
, etc) do too!
I've been reflecting on my life these past few days. I cannot believe I ended up where I am and what a wild journey. There have been so many people who've helped me and interacted with me in a meaningful way, it's staggering.
I'm truly grateful, and I'll keep trudging onward.
In 2019 I’m going to be doing a lot more independent research and hacking on projects that are off the beaten path. I think I’ll try live-streaming it too!
This seems like the right way to think about brain architecture as a result of evolution - a meta learned prior, not “pretraining”. Beff capturing the right amount of info in this tweet.
@yaroslavvb
Isn't the developmental process really a hyper network with genetic latents?
It's not pre-training, it's a meta-learner architectural hyperprior
Let’s assemble the boldest researchers across the world, work together and engineer ourselves towards learning systems that tangibly improve the world. We’ll do ambitious research, and build actual user centric tools from those ideas rather than just a paper.
Who’s with me?
In 2021, one of my goals is to use Twitter more systematically.
Core to this: creating, rather than consuming
Expect more of the following:
-> Building
@ManifoldCompute
in public
-> Theses (+ investments) in science and deep tech
-> Lots of AI research + Math, Bio, Physics
A good friend of mine passed away in accidental and freak circumstances.
He was one of my biggest supporters from when we were young, even for the wild scifi shit I used to say I would make happen by X year.
My last conversation with him a few months ago was about how it felt
What?!? DeepLearning in your browser?!
My cofounder
@MikeShi42
absolutely nailed it with this demo. Check it out if you’re interested in using
#tensorflowjs
to do ML in js on the browser. Also, check out
@ModelDepot
to find other awesome models!
Amid systematic sharing of informal thoughts & updates, I want to try something more formal. Considering 2 newsletters: a weekly research log and maybe a monthly or quarterly update around human progress (technology, policy, economics).
Let me know if you like by hitting like!
Meet Reka Core, our best and most capable multimodal language model yet. 🔮
It’s been a busy few months training this model and we are glad to finally ship it! 💪
Core has a lot of capabilities, and one of them is understanding video --- let’s see what Core thinks of the 3 body
I’m wanted to share that I’ll be joining the research team at
@Perspecta
Labs! I’ll be working on both fundamental and applied ML/DL research, and am excited to explore ideas in Metalearning, Self Supervision, Privacy and other areas!
I woke up wishing for an idea to exist:
A browser built for researchers and knowledge workers:
- terms are searched and saved in knowledge graphs
- texts are summarized by NLP models
- visualizations of information dependencies a la
@Roam
- NLP/NLU based search
Once we make significant progress on building learning systems that can solve multimodal problems and learn dynamically (the stuff we're doing
@ManifoldCompute
), the aim is to apply them to completely intractable problems like synthetic biology and longevity.
As someone who was told math isn’t for them when in school, getting into it later in life is both an amazing and frustrating process: amazed how powerful formal mathematics can be and frustrates that my environment and my own reward function didn’t weight learning math highly.
ML and Engineering friends, I really need your help.
What are the most sophisticated and/or complex real world systems that use machine learning to accomplish a significant portion of functionality?
Examples: Tesla AutoPilot, GPT-3, Alphazero, etc
I need as many as you got.
Been pretty low key about it but I’ve been investing & supporting through
@SmallStepCap
.
Investments are very small $$$, lots of support, hypothesis driven, focused on deep tech & hard science for now.
If I can help you build the future in some way, I will.
Just let me know!
1/ A bunch of people have been telling me that in 2019 they want to learn some ML. It's easy to get confused with all the shiny tutorials, 700 page textbooks, and intimidating jargon.
Here's my succinct advice on where to start.
I’ve been working waaaay too much lately balancing 2 masters, research, TAing, and trying to build interesting side project. 6 hours of sleep feels like a blessing, and it shouldn’t be
Applied with our moonshot,
@ManifoldCompute
🚀
If you're working, or planning on working on anything, ZFellows looks like an incredible opportunity.
Internet friends, through communities like
@pioneerdotapp
, have changed my life.
Last day to apply !
Here are stats:
😊: 2,000+ apps (5x Thiel Fellowship)
📚Students: MIT, Stanford, Harvard, etc.
💻 Employees: Stripe, Google, FB, SpaceX, etc.
🌏: all over
What should I do with this crazy talent pool? Only 10 slots.
Reply w/ideas :)
Even if your rocket explodes, you build a new one. This is the only perspective to keep when building something important.
You do it enough times, you will succeed.
If we collectively do it enough times, mankind advances.
All the way to the stars.
How to start an ambitious company:
1. Create an attention-capital loop with a massive vision, charisma, and social proof (hires/investors) that propels you to the Series A or B as quickly as possible. Everyone thinks you’re a genius; use that fact to hire well.
2. Build the
I was completely blown away by
@adcock_brett
’s story and guidance in the recent
@myfirstmilpod
episode w/
@thesamparr
.
Insane mission focus, enthusiasm, excellence, and humility.
I hope I can have 1/10th of that in my own work w/
@ManifoldRG
I haven’t lived too long, but some random lessons I’ve been reflecting on:
Think for yourself, and see how others/history matches up to your mental models
Plans fail, that doesn’t mean planning is bad
Deep thought and moving fast aren’t mutually exclusive
Math doesn’t suck
The
@distillpub
team, especially
@ch402
, have been incredibly encouraging and collaborative with an entire generation of researchers over the past few years.
Whatever the team chooses to do, Distill will remain a shining example of what science publishing can & should be
Distill will be taking a one year hiatus starting today, which may be extended indefinitely. Our editorial team has written some reflections on what we've learned over the past five years, and the factors that led to this decision.
There are still so many underexplored avenues for evaluating learning dynamics in Deep Neural Networks.
Check out
@ManifoldCompute
's most recent paper, accepted for an Oral at NewInML!
We explore a new metric for analyzing NN learning dynamics in Vision tasks.
So excited!
Our proposal to extend Modular Meta Learning to a Multimodal setting through task aware modulation was accepted to NeurIPS 2020's Preregistration workshop!
This is a small first step in a line of research that we're very excited for at
@ManifoldCompute
.
Is there a more ambitious group of researchers than
@ManifoldCompute
? Some projects we’re working on:
- self modifying learning systems that can do 1000s of tasks (
- swarm AI for space
- intelligent OS (like Jarvis)
we’re going to release code + papers for it all. Stay tuned.
Today, I turn a quarter century old. It has been a truly formative and enjoyable experience so far.
I’m looking forward to learning, building, laughing, and contributing to the future of our little civilization on this beautiful blue rock in the years to come.
I've been thinking about recording videos where I work on some of the projects and research that I'm up to.
Just outlining what I'm doing, documenting how I approach these problems and build technologies.
Unsure if it should be natural or highly curated
Any thoughts?
Paul and the crew are building what might be the most critical piece of infrastructure for the next generation of AI.
Incredibly stoked, and excited to use Browserbase in the new class of AI Agents we’re building.
Happy to share Browserbase with the world today.
We help AI applications browse the web.
And we just raised $6.5 million to do it.
Now, we're opening signups to developers everywhere.
I can't wait to see what you 🅱️uild.
Just tried
@meetgranola
out, and it's incredibly useful. There's clearly been a lot of thought put into each feature, and there's just an "ease" about it that is remarkable.
Was talking to a friend a few days ago, thinking of starting an open, remote friendly research group. I’ve had a lot of success doing remote research projects for one of my master’s and have seen some baller distributed teams like
@forai_ml
Remote Research Revolution baby
This is going to sound weird, but I don’t care about AGI (at least not as a central goal).
Functional, multimodal learning systems are a far cry from super intelligence but literally unlock so much value for humanity.
Let’s build them first, and then wonder about AGI
This moving, smiling visage of Alan Turing made me strangely emotional.
I hope he would be proud of how far we’ve come.
I’m sure he would be disappointed that we still continue to make mistakes just like those made in his time.
We have much to do.
Just wrote a quick, stream of consciousness list of what cool ideas and moonshots I'd like to work on this coming year. Check it out and let me know what you'd like to work on!
A sneak peek a few of the projects I’m working on this fall:
NeuralComposer: a human in the loop Architecture Search system that outputs Module, Composable Neural Networks.
Bright: a personal, predictive knowledge management system that learns from how you work.
More to come
Just came across by
@AidanNGomez
et al. Super cool to see so many promising researchers work together from around the world.
Are there any remote first research groups like this? I feel like collaborations like this should exist in more fields.
I honestly don’t know which I love more, the depth and rigor of deep quantitative research or the quick feedback cycles of shipping a product to thousands of people.
I guess I’m destined to start a deep tech or hard science company.
an underdiscussed gotcha behind the “search + LLM = AGI” narrative is search is only valuable when statewide improvements are *quantifiable*
this is the case in Go, and coding problems w/ tests, and this ARC benchmark. we can explore the (LLM-generated) state space and leverage
I've been learning a lot recently, I'm thinking about open sourcing all my notes on various topics from Gene Editing to Machine Learning for Trading. Hopefully people can learn from them and contribute
@briantimar
I’m realizing, perhaps later than all of you, that I should relentlessly pursue my visions of the world, no matter how absurdly ambitious or difficult they seem.
I’m going to try and do exactly that.
Reading through I was astonished by the amount of hours and sheer work that clearly went into this. Also a brilliant dive into HCI and AI, framed in the context of what will most certainly be one of the defining technologies of the early 21st century.
Can't wait for the videos!
The following is our paper on driver functional vigilance during use of Tesla Autopilot driver assistance system. We analyzed 18,928 Autopilot disengagements. 3+ years of hard work with an incredible research team at MIT. Example videos out next week.
To bend reality by force of will
You first must Mute The World.
Own it.
| “MUTE THE WORLD” | 2021 | ONE-OF-ONE | ON-CHAIN MIXED MEDIA | GEOFF x
@NASJAQ__
|
@AustenAllred
~ a year ago, I tried to build a curation/recommendation engine for learning resources on the internet, and my takeaway was that talking to mentors/teachers/advisors builds diverse mental models that drive deep conceptual understanding. It's harder to achieve with pure self study