AGI House SF
@AGIHouseSF
Followers
11K
Following
962
Media
147
Statuses
675
In the golden age of machine learning we're bringing hackathon life back to Silicon Valley! Shaping the future of AI, one line of code at a time.
San Francisco, CA
Joined September 2023
We recently held a Hardcore Hackathon - where people dove deep into the fundamentals of computer science, back-end engineering, and mathematics . The legendary @stephen_wolfram helped kick it off sharing some of his work from 38 years spent developing @WolframResearch
1
10
79
e/acc finally convened in the flesh at AGI House SF on Sunday. Watch @garrytan, @BasedBeffJezos, & @wagieeacc lead a tour-de-force of bold speeches on how you and I will accelerate abundance on local, cosmic, and technological levels. Full talk recordings in the thread below!
11
7
59
Live tweeting all the brilliant rebuttals to SB 1047 from the Emergency Meeting for Open-source AI we're hosting right now! 🧵🔽.
Emergency meeting for action on SB 1047 in full swing @AGIHouseSF. Stay tuned for a wave of responses defending open-source AI in the next couple hours
11
8
55
Launch-a-thon 🚀 with ProductHunt CEO @rajivayyangar, General Caralyst @generalcatalyst, Alex Riebman @AlexReibman and Andrew Cote @Andercot 🔥
1
6
42
🔥 Flux Lora Hackathon!!🔥. Turing-test passing photorealism in customized image generation is not only here - *it's open source*!!. The inventor of Lora - @edwardjhu - will be hacking with us and give a talk on what the process of creating Lora was like.
4
5
40
Happy Thanksgiving!! 🦃😋. Today we're releasing a special: the first AGI House Podcast Episode!. We'll be interviewing the founders of the next generation of great AI companies, and we're starting with @DivGarg_ of @MultiOn_AI on the frontier of browser agents. In it we.
3
12
33
🚀🔥LAUNCH DETECTED🔥🚀. Fire off your project alongside 100+ of the most cracked SF founders, engineers & hackers. Join General Catalyst @generalcatalyst and the CEO of Producthunt @rajivayyangar for the @AGIHouseSF Launch-a-thon!. June 1st, link below:
1
7
33
How can Monte-Carlo Tree Search & DPO RL improve agent outcomes, ex. taking Llama 3 from 33% to 97% accuracy with one day of training?. Get answers from the conversation between the Founder of MultiOn, @DivGarg_, and @jvnixon, about work with @rm_rafailov:
Happy Thanksgiving!! 🦃😋. Today we're releasing a special: the first AGI House Podcast Episode!. We'll be interviewing the founders of the next generation of great AI companies, and we're starting with @DivGarg_ of @MultiOn_AI on the frontier of browser agents. In it we.
0
8
28
🔥 AI x Mathematics Hackathon🔥. We'll be building in the wake of AlphaProof and AlphaGeometry with @GoogleDeepMind, @Wolfram_Alpha , and @khoslaventures!!. Come hear from @stephen_wolfram and Deepmind's @ber24 on AI powered mathematics research, judged by Khosla's @SStrohband.
4
9
26
The first project by Rohan Panday (@khoomeik) explored replacing gradient descent in LLM's with local layer objective - calculating an auxiliary loss function at each layer lets each be trained independently. At scale this would reduce total FLOPS in training by 6x
2
4
28
Join us today at the Agents in Production hackathon with @SoftBank and @restackio !. In a world where so few agents make it out of localhost - these bold hackers are shipping real production usecases by end of day!
3
3
26
Opening talks at the Emergency Llama 3.1 Hackathon!! 🔥🚀. Over 140 hackers will be shipping projects with Llama 3.1 405B on the day of it's release ⏩ . Incredible thanks to @LeptonAI , @togethercompute and @runpod_io for sponsoring!!
Today, you woke up in a new world - Open Source AI is now on top! 🔥🚀. According to Scale AI's holdout benchmarks, Llama 3.1 405B is *beating Sonnet* in instruction following. It's close by in math and not far behind in programming. Meta also released a fantastic paper,
2
1
24
I guess hacking the night before New Year's is controversial. So we're gonna throw a real party this Friday. Except it's a hack night 😉. Jk, it's BOTH:.1 floor for partying.2 floors for hacking. Building will continue until morale improves 😤.
Got invited to a pre-New Years Eve party in SF…. But instead of partying, everyone was just hacking on startups and side projects. They were intense. Here are some of the demos we saw from the party animals at @AGIHouseSF (🧵):
4
1
23
NexGen Agents Presentations thread!! . @AnthropicAI @AgentOpsAI @awscloud @notablecap @JvNixon . Check out the stream:.
1
5
21
What is being built at Defense of the West? Hear from @thedataroom and his team atop the USS Hornet 🚀🚀
0
0
18
AGI's learning pace demands that you level up. Today, you learn or you lose. That's why we 🚀launched🚀 20 new learning projects at the AI Accelerated Learning Hackathon!!. Congrats to @Marg_Groisman and @JvNixon for creating & organizing 🔥. Winners & Notables: 🧵
2
5
18
Realtime speech-to-speech hackathon morning w/ @cartesia_ai @OpenAI @Zoom @OutspeedAI - project launches incoming 🚀🚀
4
0
16
We also had an absolute stacked round of lightning talks - thank you to all the presenters:.@emaxerrno founder of @redpandadata @keithmadams Architect of Slack and founder of @Pebble_Bed .@stephen_wolfram of @WolframResearch, .@jarredsumner author of Bun, .@JvNixon founder of.
1
4
12
10 DAYS until the Bio x AI Hackathon at AGI House SF‼️. We're partnering with @DCVC and @FutureHouseSF to bring the sharpest Bio 🧬 researchers together with top AI 🤖 builders to accelerate the future of life 🚀. RSVP link in thread ⬇️
1
1
13
Jeremy Nixon (@JvNixon) developed Omni, a unified mathematical notation that is simultaneously an executable program, meaning you can directly run the math you write, and write proofs in computable code.
1
1
13
Honored to be host to the first public launch of one of the strongest image augmentation libraries, Albumentations!!. It lets you easily create a space of training data examples from a single image, adding robustness to your model's learning.
Launching . At AGI Hackathon at @AGIHouseSF.
1
3
14
Kyle Morris (@kylejohnmorris). Developed a faster, cheaper way to orchestrate ML inference on H100 GPUs without k8s/docker/python - a true skunkworks project of exploring the cutting edge of GPU orchestration - pursuing the dream of an effective load balancer up to 64 H100's
2
0
12
Eduardo Reise (@edreisMD) develooed "Hardcore Inference" - dynamically updating the context window after each token is generated by producing up to 8 "thoughts" or possibilities for the next token, performing in-situ RAG at time of inference.
1
1
12
@khoomeik Open Cell Counter - cheap automated AI web app to replace $6000 equipment. These guys amazon primed a MICROSCOPE to our house to test their model
1
2
10
@emaxerrno @redpandadata @keithmadams @pebble_bed @stephen_wolfram @WolframResearch @jarredsumner @JvNixon And Congratulations to our Winners: . 1st place: @bztree for GigaGrad, making neural networks as intuitive to slay in bare C++ as they are in Python with a massive speedup. 2nd place: @kylejohnmorris for EzML, enabling intuitive orchestration of GPUs for training without.
1
0
8
Sasha Krassovsky (@bztree) developed GigaGrad, a small deep learning library that goes gigafast by heavily relying on C++ operator overloading to provide an ergonomic interface to defining NN's, inspired by @PyTorch and @__tinygrad__
2
0
11
Happy birthday @DivGarg9 !!.
6 hrs to go + still didn't have a bday gift for my friend . so I gave an AI agent FULL ACCESS to my bank account to hit the web and do it all for me 🔥. Happy bday @DivGarg9 Let's see what @MultiOn_AI gets you 👀🎉. (. prayers for my bank acct lol)
1
0
10
Congrats to the prize winners at the Flux Lora Hackathon!! And many thanks to @FAL, @nebiusai, and @replicate for sponsoring these winners 🔥. Best use of @replicate prize - Simple Trainer by @WeiTu_.> Video -> High quality model of yourself. Other winners👇
1
1
10
It was great hearing from @mercebent @lightspeedvp over dinner last night!.
Had the best time sharing many hot takes on startups, AI, & VC at the AGI house in SF . Thanks Silicon Valley Icons and @simbajonga2 for assembling a super interesting group of people.
0
2
10
Phenomenal opportunity from @vincentweisser - fast grants for your next open-source project!!.
We're launching Fast Compute Grants to support ambitious open-source AI research into novel models, architectures, techniques, and paradigms needed to make open and decentralized AGI a reality.
0
3
9
Ash Vardanian (@ashvardanian), Mark Reed, Victor Zverovich (@vzverovich) greatly improved UCall -. Running lightweight models on high throughput servers often means networking stack is the bottleneck. They cleaned up UCall - reaching 1.5m requests per second on a single CPU core
1
1
9
🚀World simulation talks are live!!. Many thanks to Joscha Bach @Plinz, Karan @karan4d, and Rob @RobertHaisfield for leading to a new generation of AI tools. Lots of love to @AnthropicAI for sponsoring & producing Claude, whose flexibility unlocked these applications! . Watch👇.
1
2
8
Morgane Pell (@morgantepell) developed jest2bun to rewrite test files from Jest to run 6x faster in Bun - from 2.4 seconds to 0.4 seconds - and all tests passed. A cleanly executed package to accelerate testing in developing software with LLMs
1
0
8
@agihouse_org @garrytan "didn't nothing?". Except for:.1. Starting the house.2. Taking on all financial liability for the house.3. Recruiting 7/9 of the initial housemates.4. Coming up with the idea for AGI House.5. Running the first Generative Image Hackathon (9/2/22).👇.
1
0
8
@agihouse_org @garrytan "Took it over"? AGI House was supposed to be a collaboration over what was already an *incredible* community, which you were *added to* in February 2023, months after I renamed NeoGenesis to AGI House.
1
1
8
A Synopsis on AI x Law by John Nay @johnjnay, researcher at Stanford CodeX @CodeXStanford and Founder & CEO of Nomos AI. Talk:
1
0
8
@khoomeik on persistent homology: do research holes correspond to holes in paper embedding space?
1
0
8
Up next, we have also hosted the Custom LLM hackathon sponsored by Google, Modal Labs, and Anyscale, where hackers fine-tuned their own Llama models for cool use cases👇:.
@JvNixon @EGafni @ademidunk @roberthdly @lucabeetz @keerthanpg @khoomeik @AlexReibman @jamesmurdza @VibhuSapra @stevekfrey @jessemhan @kamil_k7k intern.deploy by @kylejohnmorris. Github repo: Video:
2
0
7
@khoomeik DNA-Zilla - accelerating DNA sequence matching with A* heuristics. Overheard some of their conversations…really interesting hardcore algorithms work and all shipped open source. @ashvardanian @alexbowe @Adriano34554795
1
1
7
Narek Galstyan (@diqitally), Varik Matevosyan built an extension to @lanterndb, to perform postgres vector search in the browser. Letting anyone build and ship applications on top of LLMs and vector DBs - making trying a back-end as easy as trying a front-end
1
1
7
The greatest AI journalist just launched a dangerous VC fund - can't wait to collab!.
The future is too important. That's why I'm investing in high agency founders accelerating the timeline. If you're a technical founder building generational technology reach out to @foomvc
1
1
7
Is vision the best representation of your browser for AI agents take autonomous action? Or is it the underlying HTML, JS & CSS code?. Get answers from the conversation between the Founder of MultiOn, @DivGarg_, and @jvnixon:
Happy Thanksgiving!! 🦃😋. Today we're releasing a special: the first AGI House Podcast Episode!. We'll be interviewing the founders of the next generation of great AI companies, and we're starting with @DivGarg_ of @MultiOn_AI on the frontier of browser agents. In it we.
0
2
6
We'll hear from @replicate's @GandalfHz on how to train and serve your Loras on Replicate, much like @levelsio described in his recent @lexfridman interview. There will also be a 1k prize for the best use of Replicate!.
Fine-tune FLUX.1 on your face! . It's pretty easy: Gather some images of yourself, create a new fine-tuned model, then use the web or API to generate novel images of yourself as a superhero, a cartoon character, an adventurer, or just. a regular person.
1
2
6
Congrats to @AlexReibman, @hackgoofer, and @aiengfoundation for running such a glorious hackathon!!.
280 hackers signed up to build what’s possible with the latest advancements in multimodal AI. This is a sneak peak at the new age of the AI-first Interface. Here’s what we saw at the @aiengfoundation Realtime Voice AI and Multimodal Hackathon in San Francisco (🧵):
0
0
6
We'll be hacking with @nebiusai's compute, for anyone who wants to compete for @replicate's 1k Lora research prize - discovering a way to target face relevant neurons in the base model, or do any other custom computation during the hackathon.
🔥 This fall, we’re adding @NVIDIA H200 SXM GPUs. Currently, #H200 is the most powerful #GPU for your AI. Its memory and data access speed ensure up to 2x the LLM inference performance over H100. Prices start from $2.50 GPU/h. Reserve your cluster today:
1
0
5
And that's a wrap folks! Over 10 well thought out essays & threads from experts in the open-source AI community pointing out issues in SB 1047 that we look forward to seeing addressed by our representatives @Scott_Wiener.@GeneralRoth.@SenSusanRubio.@HenrySternCA.
0
0
5
We just got @AlexReibman 's new VC firm, his first YC company lead - movin' fast at Launchathon :D 🚀🚀
0
0
6
Join us this Saturday/Sunday for the accelerated learning hackathon backed by @upstageai . Some of the usual suspects are pulling up, let’s see what incredible tools they create this weekend to help you learn faster!. .
0
1
6