←
Andrew M. Dai672 posts
Andrew M. Dai avatar✓

Andrew M. Dai ✓

@AndrewDai

Co-founder and CEO at @ElorianAI. Ex (director) at Google DeepMind after 12y. Gemini Data Area Lead. Now building at the visual frontier. 戴明博

◷ Joined February 2011▧ 25 Photos and videos
Andrew M. Dai avatar
Andrew M. Dai@AndrewDai·Oct 16

Spent years watching models ace a single image and completely whiff on anything that moves or changes angle. Today we’re fixing that. Try Glimpse now, our first visual thinking model with competitive frontier performance compared to Astra on key public visual reasoning benchmarks.

1:30
Andrew M. Dai avatar
Andrew M. Dai@AndrewDai·20h

I never thought it would be possible but we've finally found the source data that explains why LLMs infinite loop.

Andrew M. Dai avatar
Andrew M. Dai@AndrewDai·Nov 13

Bay area transport really underperforms in the rain. I avoided the 101 because of the crashes. Now I'm stuck on Caltrain thanks to the power going down (likely due to rain). If only I could code a faster train.

Andrew M. Dai avatar
Andrew M. Dai@AndrewDai·Nov 10

Do models even use their grounding on visual problems? Not always! It was great working with our first intern, @dzyan01 on his new paper.

0:30
Andrew M. Dai avatar
Andrew M. Dai@AndrewDai·Nov 7

It was really great collaborating with @UniPat_AI on ScienceVision. There're so many opportunities to empower the sciences with better visual reasoning for AI.

Andrew M. Dai avatar
Andrew M. Dai@AndrewDai·Nov 6

Thank you @mercor ! It was great fun working with you too. Proud to release this together. Looking forward to more collaborations!

Andrew M. Dai avatar
Andrew M. Dai@AndrewDai·Nov 6

Tired of seeing benchmarks saying we are at AGI but models still can't understand a 3D protein view or a floor plan? When a visual model gets something wrong, is it failing to see or failing to reason about what it sees? We've built two benchmarks that answers these questions. Explore DecoyCount and ScienceVision here: github.com/elorian-ai/decoyc… and huggingface.co/datasets/Unip… And thank you @mercor and @UniPat_AI for your partnership in building these!

Andrew M. Dai avatar
Andrew M. Dai@AndrewDai·Nov 4

Lean is the verifiable bedrock that enabled this AI math revolution. The Stacks formalization is the soil that the next advances will grow from.

4:19
Andrew M. Dai avatar
Andrew M. Dai@AndrewDai·Oct 29

I had a great time chatting with Richard about our new benchmark, DecoyCount. But our agent messed up. It released our model publicly and our benchmark for early access instead of vice versa. So enjoy our model, Glimpse, and look out for our new benchmark soon!

Andrew M. Dai avatar
Andrew M. Dai@AndrewDai·Oct 23

ChatGPT looked at a concert, saw the lights, and decided they were all Yellow from @coldplay. Our new visual thinking AI model, Glimpse, knows the Shape of You, @edsheeran. It got it right on the first try. Try it out!

Post attachmentPost attachment
0:15
Andrew M. Dai avatar
Andrew M. Dai@AndrewDai·Oct 23

Really exciting to see new use-cases from people using Glimpse everyday! And we love to see people start building their own tools around our public API. In this video, it's a UI tool that finds issues between a final website and its Figma design, another capability we never previously tested our model on. We have more great things coming to the model soon, stay tuned!

0:58
Andrew M. Dai avatar
Andrew M. Dai@AndrewDai·Oct 22

Congratulations on Stephanie's new fund! We wouldn't have been able to build our model without investors that believe in founders early and taking contrarian bets. In the age of vibe coding, contrarian starts being the key differentiator.

Karan avatar
⟳ Andrew M. Dai retweeted
Karan@karankendre·Oct 16

Can CCTV show why a situation is unsafe? I built SiteCheck to review construction footage. It’s powered by Glimpse from Elorian. A new lab from the team behind Google’s Gemini and Apple’s MM-1. Visual Thinking, applied to site safety. Monitor a video and ask about safety. SiteCheck returns timestamped concerns and their reasons. No danger zones need to be drawn. Here, a worker approaches an unguarded opening. A fall-risk warning highlights the nearby worker. It clears after he moves away. The opening remains marked as a hazard. The construction footage here is AI-generated. The analysis comes from actual API calls. @ElorianAI @AndrewDai

0:15
1:30