Devi Parikh

4.6K posts

Devi Parikh banner
Devi Parikh

Devi Parikh

@deviparikh

Co-CEO @yutori_ai. https://t.co/zD3StYi8db. Previously: Senior Director, GenAI & FAIR at Meta, Associate Professor at Georgia Tech.

San Francisco, CA Katılım Haziran 2009
139 Takip Edilen27.3K Takipçiler
Sabitlenmiş Tweet
Devi Parikh
Devi Parikh@deviparikh·
We gave some of our partners early access to n1.5 — the most capable computer use model for the web. It is in production at FAANG scale as we speak, replacing a computer use model from a frontier lab. If your product can benefit from web automation — extracting structured data from dynamic webpages, filling forms, completing workflows on the web, testing vibe coded web apps — you should try out @yutori_ai's Navigator n1.5! Save your GPT / Claude / Gemini capacity for something else :)
English
4
6
25
8.3K
Devi Parikh retweetledi
Maximillian Piras
Maximillian Piras@MVXMXM·
The @yutori_ai API Platform will get some love this week, starting w/ this fix for sidebar layout shifts during state change (was driving me crazy).
English
1
2
5
266
Beth Hall PhD
Beth Hall PhD@BethHallPhD·
@deviparikh Goodbye sweet prince. I remember getting approval to run the first HITs at the NIH. Taught me a lot of great principles when it comes to human data and tasks - short, sweet, and explicit will always win the day 🥰
English
2
0
2
496
Devi Parikh
Devi Parikh@deviparikh·
End of an era. Curious: How many of you know of Amazon Mechanical Turk?
Devi Parikh tweet media
English
22
15
192
95K
Devi Parikh retweetledi
Yu Su
Yu Su@ysu_nlp·
Great work by the @yutori_ai team!
Dhruv Batra@DhruvBatra_

𝗡𝗮𝘃𝗶𝗴𝗮𝘁𝗼𝗿 𝗻𝟭.𝟱 “𝘀𝗼𝗹𝘃𝗲𝗱” 𝗢𝗻𝗹𝗶𝗻𝗲 𝗠𝗶𝗻𝗱𝟮𝗪𝗲𝗯: 𝟵𝟳.𝟯% 𝘀𝘂𝗰𝗰𝗲𝘀𝘀 𝗿𝗮𝘁𝗲. While some teams self-report, this result is independently evaluated and verified by OSU NLP Group @osunlp and Careerflow Human Data Labs. All benchmarks are transient attempts at measuring progress. Ultimately, what matters is how a model performs when people use it. But there’s a sentiment online that computer-use models aren’t progressing quickly. Not true. In the last year, performance on Online Mind2Web has gone from ~40% success to basically saturated. So what’s next? Most computer-use/browser-use benchmarks are GUI-only. Models (including Navigator n1.5) now support hybrid actions — UI interactions (click, type, scroll) and programmatic actions (e.g., execute JS). Ultimately, we’re headed to a world where computer-use models “agentify” the long-tail of the web. huggingface.co/spaces/osunlp/…

English
0
5
12
3.3K
Devi Parikh retweetledi
Rui Wang
Rui Wang@theruiwang·
Online-Mind2Web is "solved". We’ve received officially verified results from the OSU NLP Group: - 97.3% human-verified accuracy, following three independent reviews, additional QA on borderline cases, and final manual verification by the benchmark authors. - 87.9% auto-eval accuracy with WebJudge, with all inputs and outputs from all three evaluation stages submitted for full reproducibility. We thank the benchmark authors @xue_tianci @hhsun1 @ysu_nlp for upholding high academic standards and applying consistent evaluation protocols across official submissions. Their work gives the field a fair and rigorous way to measure progress and compare models directly in the real world. One year ago, SOTA performance on Online-Mind2Web was around 50%. Today, it stands at 97.3%—a significant milestone for computer-use agents. What a year!
Dhruv Batra@DhruvBatra_

𝗡𝗮𝘃𝗶𝗴𝗮𝘁𝗼𝗿 𝗻𝟭.𝟱 “𝘀𝗼𝗹𝘃𝗲𝗱” 𝗢𝗻𝗹𝗶𝗻𝗲 𝗠𝗶𝗻𝗱𝟮𝗪𝗲𝗯: 𝟵𝟳.𝟯% 𝘀𝘂𝗰𝗰𝗲𝘀𝘀 𝗿𝗮𝘁𝗲. While some teams self-report, this result is independently evaluated and verified by OSU NLP Group @osunlp and Careerflow Human Data Labs. All benchmarks are transient attempts at measuring progress. Ultimately, what matters is how a model performs when people use it. But there’s a sentiment online that computer-use models aren’t progressing quickly. Not true. In the last year, performance on Online Mind2Web has gone from ~40% success to basically saturated. So what’s next? Most computer-use/browser-use benchmarks are GUI-only. Models (including Navigator n1.5) now support hybrid actions — UI interactions (click, type, scroll) and programmatic actions (e.g., execute JS). Ultimately, we’re headed to a world where computer-use models “agentify” the long-tail of the web. huggingface.co/spaces/osunlp/…

English
2
5
12
1.3K
Devi Parikh retweetledi
Abhishek Das
Abhishek Das@abhshkdz·
Navigator n1.5 is now the top officially verified entry on Online-Mind2Web. 97.3% human eval, 87.9% auto eval. This one's basically solved. On to harder benchmarks.
Dhruv Batra@DhruvBatra_

𝗡𝗮𝘃𝗶𝗴𝗮𝘁𝗼𝗿 𝗻𝟭.𝟱 “𝘀𝗼𝗹𝘃𝗲𝗱” 𝗢𝗻𝗹𝗶𝗻𝗲 𝗠𝗶𝗻𝗱𝟮𝗪𝗲𝗯: 𝟵𝟳.𝟯% 𝘀𝘂𝗰𝗰𝗲𝘀𝘀 𝗿𝗮𝘁𝗲. While some teams self-report, this result is independently evaluated and verified by OSU NLP Group @osunlp and Careerflow Human Data Labs. All benchmarks are transient attempts at measuring progress. Ultimately, what matters is how a model performs when people use it. But there’s a sentiment online that computer-use models aren’t progressing quickly. Not true. In the last year, performance on Online Mind2Web has gone from ~40% success to basically saturated. So what’s next? Most computer-use/browser-use benchmarks are GUI-only. Models (including Navigator n1.5) now support hybrid actions — UI interactions (click, type, scroll) and programmatic actions (e.g., execute JS). Ultimately, we’re headed to a world where computer-use models “agentify” the long-tail of the web. huggingface.co/spaces/osunlp/…

English
1
3
14
1.6K
Devi Parikh
Devi Parikh@deviparikh·
Yutori’s Navigator n1.5 “solved” Online-Mind2Web, one of the primary public benchmarks for evaluating computer-use models on the web. This is based on officially verified results from the benchmark organizers. High quality benchmarks — realistic (measuring what matters), spanning a spectrum of difficulty (measuring progress along the way) and rigorous evaluation protocols (separating signal from noise) — is not easy, and is highly valuable to the ecosystem. Kudos to the Online-Mind2Web organizers! n1.5 was released a month ago. Stay tuned for what’s in the pipeline :)
Dhruv Batra@DhruvBatra_

𝗡𝗮𝘃𝗶𝗴𝗮𝘁𝗼𝗿 𝗻𝟭.𝟱 “𝘀𝗼𝗹𝘃𝗲𝗱” 𝗢𝗻𝗹𝗶𝗻𝗲 𝗠𝗶𝗻𝗱𝟮𝗪𝗲𝗯: 𝟵𝟳.𝟯% 𝘀𝘂𝗰𝗰𝗲𝘀𝘀 𝗿𝗮𝘁𝗲. While some teams self-report, this result is independently evaluated and verified by OSU NLP Group @osunlp and Careerflow Human Data Labs. All benchmarks are transient attempts at measuring progress. Ultimately, what matters is how a model performs when people use it. But there’s a sentiment online that computer-use models aren’t progressing quickly. Not true. In the last year, performance on Online Mind2Web has gone from ~40% success to basically saturated. So what’s next? Most computer-use/browser-use benchmarks are GUI-only. Models (including Navigator n1.5) now support hybrid actions — UI interactions (click, type, scroll) and programmatic actions (e.g., execute JS). Ultimately, we’re headed to a world where computer-use models “agentify” the long-tail of the web. huggingface.co/spaces/osunlp/…

English
0
3
13
1.8K
Devi Parikh retweetledi
Marius
Marius@balajmarius·
great talk @MVXMXM. also, some of the best-designed slides i’ve seen this year.
Marius tweet media
English
1
4
17
3.4K
Devi Parikh retweetledi
Maximillian Piras
Maximillian Piras@MVXMXM·
Excited to speak again at @aiDotEngineer World's Fair! This year I'm presenting "mousepower": on how agents break our measurement frameworks, which weren't built for systems that output tirelessly in parallel. If execution is now cheap & judgment is the new bottleneck, how do we measure value at the speed of compute?
AI Engineer@aiDotEngineer

The 2026 World's Fair is completely sold out 🫡 ✅ The largest AI industry expo on earth ✅ Sold out on Leadership track for CTOs & VP AI's ✅ Sold out on Workshops tomorrow ✅ Sold out on ALL late bird tickets 🙌 65 side events still FREE all over SF (see website) What we will never sell out: Our commitment to publishing all the best AI engineering content for free online on YouTube. We have now opened limited overflow tickets for our expo and engineering tix — no seating guaranteed, sessions are first come first served. If you ARE one of our attendees, DO come down to Moscone for New Engineer Orientation tonight from 5p-9p to meet new friends and skip the morning crush for tomorrow. We expect EXTREMELY heavy last minute registration and need your help to load balance across days. Please give your speakers and sponsors all the love for all the effort they are putting into making this the greatest show we have ever done!

English
3
5
21
7.1K
tsunami_crypto
tsunami_crypto@ls_brd·
@deviparikh not a rhetorical question actually, i also saw one for the first time this week. are we being gaslit by big watermelon?
English
1
0
1
167
Devi Parikh
Devi Parikh@deviparikh·
Have yellow (as opposed to red) watermelons always been a thing? I’ve seen them twice in the last week, and never before that.
English
3
0
4
4.3K
Devi Parikh retweetledi
AGI House
AGI House@agihouse_org·
@yutori_ai started with a logical assumption: machines should read machine code. The team soon discovered that teaching agents to see websites like humans could be the more scalable path: @deviparikh @DhruvBatra_ Full interview on Youtube.
English
1
7
17
3.9K
Devi Parikh retweetledi
Jasper Gilley
Jasper Gilley@0xjasper·
New research: I introduce forward self-models, which are small meta-models that predict a transformer's later-layer activations from its early-layer activations. They compress the main model's computation and therefore capture genuine computational novelty in the main model
Jasper Gilley tweet media
English
7
19
224
16.4K
Devi Parikh
Devi Parikh@deviparikh·
@kiranmkota You'll have more personal space, people won't be bumping into you (you = you + your backpack)!
English
2
0
0
57
Devi Parikh
Devi Parikh@deviparikh·
PSA: Taking your backpack off when on an even slightly crowded train helps hugely.
English
3
0
40
6.2K
Devi Parikh retweetledi
AGI House
AGI House@agihouse_org·
We sat down with @yutori_ai cofounders @deviparikh and @DhruvBatra_ to talk web agents, multimodal AI, and why the messy web may need agents that see and click like humans: here are the biggest takeaways ⬇️🧵 0/7
English
3
5
23
5.3K
Devi Parikh
Devi Parikh@deviparikh·
“We’re not searching for the meaning of life, but for the feeling of being alive.”
English
2
0
13
1.4K
Devi Parikh
Devi Parikh@deviparikh·
@StartUpRabbi No, this is SF. I suspect this may not need to be said in NYC? :)
English
1
0
0
196

Keşfet

@yutori_ai @BethHallPhD @xue_tianci @hhsun1 @ysu_nlp @MVXMXM @aiDotEngineer @ls_brd