Justin Ji

@emilyzsh @thatsnotoptimal very cool! inspo from picture of dorian gray?

𝓟𝓸𝓻𝓽𝓻𝓪𝓲𝓽 𝓸𝓯 𝓨𝓸𝓾 @thatsnotoptimal and i created Portrait of You, a series of generative, living works that erode with our brainrot digital consumption in real time. this public installation of self-surveillance will be running 24/7 for the next month.

@emilyhanyf @modal congrats emily!

Life update: I’ve decided to take some time off from Stanford. I've seen a lot of these "dropping out of college" posts, but I never thought I’d write one myself. To be completely honest, I never seriously considered taking a leave or dropping out. Interning at @modal this past summer changed my plans. People often talk about @modal's talent density, but working within it gave me a whole new perspective on what “exceptional” looks like in engineering, product, design, GTM, and most importantly, in company culture. I wanted to keep learning in this environment, especially now. Over just a few months interning at @modal this summer, I saw Lovable create 1M+ sandboxes on @modal, the launch of Modal SDK 1.0, GPU Performance Glossary, GPU Memory Snapshots, multi-node training clusters, @modal Notebooks, and so much more. I felt this was the moment to join and build the infrastructure for the entire AI/ML lifecycle, the way it should be. So I'm excited to share that I'm starting full-time at @modal today, alongside the announcement of @modal's Series B! It was a tough choice to step away from Stanford, even for a little while, but I’m excited to continue learning and growing with this team. If you're a student who want to work with incredible engineers, be part of a truly welcoming culture, and join in on karaoke or goat-petting events, or if you’re unsure about taking time off, I’d be happy to share my experience!

Charts and graphs help people analyze data, but can they also help AI? In a new paper, we provide initial evidence that it does! GPT 4.1 and Claude 3.5 describe three synthetic datasets more precisely and accurately when raw data is accompanied by a scatter plot. Read more in🧵!

@shosquared @htaneja @yuris @neerajarora @htaneja +EV play

@chongz @shosquared can confirm, nobody gives more action than us 😝

@shosquared @shosquared is an action player everyone should add to their games 🙂‍↕️

I'm visiting exceptional startups in New York this week on the Harvard Startup Trek organized by my brilliant friends @kelschen99, @jonathanlei0, and @evatuecke. Today, we... – Visited Embedded Intelligence (ei.inc) and @Medal_TV, thank you @PimDeWitte and team for an exciting demo. – Visited @bubble, thanks @joshhaas and @emmanuel_s for insights into your founding journey, 7 years of bootstrapping, and challenges of company building (over great chocolate chip coconut cookies too 🍪)! – Visited Moment (withmoment.com) for a deep dive on fixed income trading. Fixed income is the world's most traded asset class, more than 3x size of global equities market, yet surprisingly illiquid. That may soon change, thanks to Moment (who recently partnered with @apexfintech). – Brief pit stop at the @haizelabs office and reunion with @leonardtang_ and @steveshenli. – Visited @withgraphite, thank you to @gregmfoster and @TomasReimers for sharing your origin story building a company amidst a global pandemic, pivoting based on customer feedback, and finding PMF. – Ending the day with a @generalcatalyst x @SunoMusic AI social. Thank you @jonnyjzhang, @Nim_Ravid1, @apjacob03, @KonstDaskalakis, @max_rimpel and team for hosting us at the GC NYC office and for throwing an awesome party. All this and we've only been in NYC for < 24 hours. If you're around, please reach out – would love to meet up while we're here. Everyone else on the Harvard Startup Trek: @calebcapoccia, @evatuecke, @grx_xce, @InakiArango, @jonathanlei0, @jujipotle, @kelschen99, @Nim_Ravid1, @kineticyang, Peggy Yin, Prin Pulkes

Speed up your DPO training with zero compromises using ✨prefix sharing✨! Prefix sharing speeds up training by 1.2-1.6x 🚀across the datasets we tested and is numerically identical to standard DPO training. Paper: arxiv.org/abs/2410.20305 Code: github.com/frankxwang/dpo…

So excited to have my first #NeurIPS paper!! And so very lucky to have the best collaborators and mentors for this project.

What's missing from #LLM #chatbots? Read thoughts from #KempnerInstitute's graduate fellow @ke_li_2021 in this @gradientpub blog:

DPO is extremely effective when scaled with on-policy data scored by synthetic judges. In this blog post, we conduct a case study on summarization, directly optimizing for objective metrics like Q&A accuracy + compression instead of typical subjective ones.

⭐New Paper Alert ⭐ How can your #RL agent quickly adapt to new distribution shifts ? And without ANY tuning?🤔 We suggest you get on the Fast TRAC🏎️💨, our new Parameter-free Optimizer that surprisingly works. Why? Website:computationalrobotics.seas.harvard.edu/TRAC/ 1/🧵

Our first cohort of KRANIUM @Harvard undergrads is here! Learn more about this intensive summer research program, the students, their mentors, and their projects: bit.ly/3VPXzDd #AI #NeuroAI #ML

@walden_yan this is incredible walden 🤯 can devin make a website that scrapes harvard mailing lists for events with free food (asking for a friend)? also would love beta access to play around with!

Today I’m excited to announce our company, Cognition, and share Devin with the world. As a kid, coding was my way of turning ideas in my head into reality. Building Devin is our way of making this possible for everyone. We want to show you all the things Devin can build!! Comment with things for Devin to work on and we’ll forward them (check out some examples below)

@MattTtkool @JohnnyNi13 will be there!

@JohnnyNi13 Lmk! We might host smth for harv/mit ppl coming for treehacks

Anyone at the new pear vc office that can let me in to visit? I'm coming to SF next weekend (also for treehacks)

@aegis_protocol my first twitter reply goes to aegis

Trust redesigned. Something coming soon, interact with this tweet 👀