Nikhil Naik
@nikhil_ai
Followers
2K
Following
3K
Media
23
Statuses
403
Llama multimodal training @AIatMeta (tweets personal) | Previously AI researcher: @MIT, @Harvard, @sfresearch, @Google
San Francisco, CA
Joined May 2011
Excited to share Diffusion-DPO, a method to directly align diffusion models to user preference. DPO-tuned SDXL obtains a 70% win rate over SDXL on PartiPrompts, a new SOTA for open source models! It is also effective at Learning from AI Feedback. https://t.co/5C5ldNHoPB (1/N)
4
36
234
Do checkout this amazing model release by @nikhilaravi @PengchuanZ and team!
Meet SAM 3, a unified model that enables detection, segmentation, and tracking of objects across images and videos. SAM 3 introduces some of our most highly requested features like text and exemplar prompts to segment all objects of a target category. Learnings from SAM 3 will
0
0
6
Excited to announce the first commercial-scale diffusion language model---Mercury Coder. Mercury runs at 1000 tokens/sec on Nvidia hardware while matching the performance of existing speed-optimized LLMs. Mercury introduces a new approach to language generation inspired by image
We are excited to introduce Mercury, the first commercial-grade diffusion large language model (dLLM)! dLLMs push the frontier of intelligence and speed with parallel, coarse-to-fine text generation.
32
35
396
It's been really awesome to watch the progression of improvements in AI weather prediction. 5-6 years ago, AI models started to be better than classic methods out to about 6-8 hours. Then it became 2-3 days, and now, AI methods are state of the art out to 15 days (+ way more
Today in @Nature, we’re presenting GenCast: our new AI weather model which gives us the probabilities of different weather conditions up to 15 days ahead with state-of-the-art accuracy. ☁️⚡ Here’s how the technology works. 🧵 https://t.co/PWCNWbQnlU
17
33
419
congratulations, @goodfellow_ian, for the test-of-time award at @NeurIPSConf ! this award reminds me of how GAN started with this one email ian sent to the @Mila_Quebec lab mailing list in May 2014. super insightful and amazing execution!
15
155
1K
Amazing impact on science
A 20th birthday post for Google Scholar with 20 fun facts, by my delightful colleagues Anurag Acharya and Alex Verstak. 🎉 https://t.co/OVKfzrYZ5g
0
0
4
We built a B2B SaaS sales company and here’s what it taught us about B2B SaaS sales 🧵👇 (but actually) Today we’re launching Rox, the first publicly available AI agent swarm for the top sales teams, and in the private beta it already helped reps grow their books 30%. 2025 is
0
0
3
Moravec's paradox in LLM evals I was reacting to this new benchmark of frontier math where LLMs only solve 2%. It was introduced because LLMs are increasingly crushing existing math benchmarks. The interesting issue is that even though by many accounts (/evals), LLMs are inching
1/10 Today we're launching FrontierMath, a benchmark for evaluating advanced mathematical reasoning in AI. We collaborated with 60+ leading mathematicians to create hundreds of original, exceptionally challenging math problems, of which current AI systems solve less than 2%.
153
512
4K
BREAKING NEWS The Royal Swedish Academy of Sciences has decided to award the 2024 #NobelPrize in Chemistry with one half to David Baker “for computational protein design” and the other half jointly to Demis Hassabis and John M. Jumper “for protein structure prediction.”
514
9K
20K
1/ I'm thrilled to share something close to my heart: I'm co-founding Bevel (@bevel_health) with @benjyang_ and @greyngyen. It's born from my personal journey to better health. Here's my story...
83
39
973
Introducing Meta Segment Anything Model 2 (SAM 2) — the first unified model for real-time, promptable object segmentation in images & videos. SAM 2 is available today under Apache 2.0 so that anyone can use it to build their own experiences Details ➡️ https://t.co/eTTDpxI60h
153
1K
7K
Diffusion models are state-of-the-art for continuous data generation (images, videos, etc). Can they beat autoregressive models also on text generation? Check out our ICML paper tomorrow to find out how. Congrats to my students @aaron_lou @chenlin_meng for the best paper award!
11
29
302
Our new paper MJ-BENCH evaluating generative reward models for text-to-image generation is now out! We find that Large Vision Language Models can act as zero shot feedback providers for diffusion models! More details below 👇
1
12
36
Very excited to announce that Mark Zuckerberg will be joining us @southpkcommons for a talk on Aug 6th! It's a rare chance to hear from one of the great founders of our time on how he kept a -1 to 0 mindset while building @Meta. Space is very limited. Apply to attend below.
34
24
635
Congratulations @raskarmit and team!
Our 2013 SIGGRAPH paper "Femto-Photography: Capturing and Visualizing the Propagation of Light" has received the 2024 Test of Time Award! It's given to papers "that have had a significant and lasting impact on computer graphics and interactive techniques over at least a decade".
1
0
3
I am currently holding my dad's cryopreserved brain tumor samples in hopes of creating a personalized vaccine for immunotherapy. However, there are some critical and time-sensitive questions in the attached post: https://t.co/3lxkUruGRe This is time-sensitive so would appreciate
15
60
164
A great keynote!
I gave a bit of an unusual keynote talk at #iclr2024 last month. I shared five stories from my 20-year journey in AI so far. It had felt like a bit of a gamble. I wasn’t sure how it would be received. But from the feedback I got in the days and weeks after, it seems like at
0
1
3
(1/N) CFG requires high guidance (>5) to "work", but comes with several issues 🤦♂️: reduced diversity, saturation, poor invertibility. Is this inevitable? 🤔 Presenting CFG++,🚀 a simple fix enabling small guidance: better sample quality + invertibility, smooth trajectory 🤟
5
34
218
@ArmenAgha Two related good quotes I heard recently: "You can prove that something won't work at small scale, but not that something works at small scale" "There's way more ideas out there than compute that's willing to take a risk on it"
8
21
383
Great release by @krandiash @_albertgu and team!
Today, we’re excited to release the first step in our mission to build real time multimodal intelligence for every device: Sonic, a blazing fast (🚀 135ms model latency), lifelike generative voice model and API. Read https://t.co/kmqpKoR1NA and try Sonic https://t.co/rMnegk14Jl
0
0
2
What a year for open ML! Trending models on Hugging Face include models from Meta, Google (TimesFM, PaliGemma), Tencent, NVIDIA, DeepSeek, RefuelAI, TII, Salesforce, 01-ai, Apple, Fugaku, Hugging Face, Microsoft, Stability, NousResearch, Gradient, Mistral, ByteDance 🤯
2
12
60