Jeff Dean
@jeffdean.bsky.social
Google Chief Scientist, Gemini Lead. Opinions stated here are my own, not those of Google. Gemini, TensorFlow, MapReduce, Bigtable, Spanner, ML things, ...
This model scores very well in the image editing lmarena leaderboard, with a large gap of ~170 points to the next group of models.
lmarena.ai/leaderboard/...
lmarena.ai/leaderboard/...
August 27, 2025 at 1:12 AM
This model scores very well in the image editing lmarena leaderboard, with a large gap of ~170 points to the next group of models.
lmarena.ai/leaderboard/...
lmarena.ai/leaderboard/...
Our latest Gemini image generation and editing model is quite good. It lets you indulge in a bit of creative fun, helps you make new business cards, etc!
Try it out at gemini.google.com
Blog: blog.google/products/gem...
Try it out at gemini.google.com
Blog: blog.google/products/gem...
August 27, 2025 at 1:10 AM
Our latest Gemini image generation and editing model is quite good. It lets you indulge in a bit of creative fun, helps you make new business cards, etc!
Try it out at gemini.google.com
Blog: blog.google/products/gem...
Try it out at gemini.google.com
Blog: blog.google/products/gem...
AI efficiency is important. The median Gemini Apps text prompt in May 2025 used 0.24 Wh of energy (<9 seconds of TV watching) & 0.26 mL (~5 drops) of water. Over 12 months, we reduced the energy footprint of a median text prompt 33x, while improving quality:
cloud.google.com/blog/product...
cloud.google.com/blog/product...
August 21, 2025 at 1:39 PM
AI efficiency is important. The median Gemini Apps text prompt in May 2025 used 0.24 Wh of energy (<9 seconds of TV watching) & 0.26 mL (~5 drops) of water. Over 12 months, we reduced the energy footprint of a median text prompt 33x, while improving quality:
cloud.google.com/blog/product...
cloud.google.com/blog/product...
Plus early cryptography cipher machines, medical instruments, early musical instruments and more.
The guides were awesome and did live demonstrations.
The guides were awesome and did live demonstrations.
June 10, 2025 at 3:41 PM
Plus early cryptography cipher machines, medical instruments, early musical instruments and more.
The guides were awesome and did live demonstrations.
The guides were awesome and did live demonstrations.
Early water+whistle based timer (up to seven hours) to wake up slumbering philosophy students so they didn't oversleep.
(Someone on X suggested "NapReduce" as a good name 😂)
(Someone on X suggested "NapReduce" as a good name 😂)
June 10, 2025 at 3:38 PM
Early water+whistle based timer (up to seven hours) to wake up slumbering philosophy students so they didn't oversleep.
(Someone on X suggested "NapReduce" as a good name 😂)
(Someone on X suggested "NapReduce" as a good name 😂)
Early system that computed position of five planets, phase of moon, etc. thousands of years into future using intricate gears
June 10, 2025 at 3:37 PM
Early system that computed position of five planets, phase of moon, etc. thousands of years into future using intricate gears
Early long-distance communication system using 5 bits (fire high or low) to encode the 24 Greek letters, one character at a time.
June 10, 2025 at 3:36 PM
Early long-distance communication system using 5 bits (fire high or low) to encode the 24 Greek letters, one character at a time.
If you're ever in Athens, I highly recommend the Museum of Ancient Greek Technology!
maps.app.goo.gl/sngwM98ueKkr...
Early humanoid robots to pour water+wine, demonstrations of heat+steam+water weight to "magically" open 500 kg temple doors, armor, ...
(cont)
maps.app.goo.gl/sngwM98ueKkr...
Early humanoid robots to pour water+wine, demonstrations of heat+steam+water weight to "magically" open 500 kg temple doors, armor, ...
(cont)
June 10, 2025 at 3:35 PM
If you're ever in Athens, I highly recommend the Museum of Ancient Greek Technology!
maps.app.goo.gl/sngwM98ueKkr...
Early humanoid robots to pour water+wine, demonstrations of heat+steam+water weight to "magically" open 500 kg temple doors, armor, ...
(cont)
maps.app.goo.gl/sngwM98ueKkr...
Early humanoid robots to pour water+wine, demonstrations of heat+steam+water weight to "magically" open 500 kg temple doors, armor, ...
(cont)
Here's the WebDev leaderboard from web.lmarena.ai/leaderboard
May 6, 2025 at 3:22 PM
Here's the WebDev leaderboard from web.lmarena.ai/leaderboard
Here’s a glimpse of what it can do - check out this retro Windows 95-style web app the model can build from a simple prompt.
May 6, 2025 at 3:09 PM
Here’s a glimpse of what it can do - check out this retro Windows 95-style web app the model can build from a simple prompt.
My wife was cleaning out some drawers and found some childhood art of now-adult @vdean.bsky.social of the two of us with my longtime colleague Sanjay.
April 20, 2025 at 10:35 PM
My wife was cleaning out some drawers and found some childhood art of now-adult @vdean.bsky.social of the two of us with my longtime colleague Sanjay.
I remember my excitement when I first learned about the Mandelbrot set as a kid, and I spent a while implementing a Mandelbrot set visualizer. Check out the code the Gemini 2.5 Pro model wrote given the prompt “p5js to explore a Mandelbrot set”, visualizing its intricate fractal patterns!
March 25, 2025 at 5:27 PM
I remember my excitement when I first learned about the Mandelbrot set as a kid, and I spent a while implementing a Mandelbrot set visualizer. Check out the code the Gemini 2.5 Pro model wrote given the prompt “p5js to explore a Mandelbrot set”, visualizing its intricate fractal patterns!
🥁Introducing Gemini 2.5, our most intelligent model with impressive capabilities in advanced reasoning and coding.
Now integrating thinking capabilities, 2.5 Pro Experimental is our most performant Gemini model yet. It’s #1 on the LM Arena leaderboard. 🥇
Now integrating thinking capabilities, 2.5 Pro Experimental is our most performant Gemini model yet. It’s #1 on the LM Arena leaderboard. 🥇
March 25, 2025 at 5:25 PM
🥁Introducing Gemini 2.5, our most intelligent model with impressive capabilities in advanced reasoning and coding.
Now integrating thinking capabilities, 2.5 Pro Experimental is our most performant Gemini model yet. It’s #1 on the LM Arena leaderboard. 🥇
Now integrating thinking capabilities, 2.5 Pro Experimental is our most performant Gemini model yet. It’s #1 on the LM Arena leaderboard. 🥇
Want to check out the source for the "AlexNet" paper? Google has made the code from Krizhevsky, Sutskever and Hinton's seminal "ImageNet Classification with Deep Convolutional
Neural Networks" paper open source, in partnership with the Computer History Museum.
computerhistory.org/press-releas...
Neural Networks" paper open source, in partnership with the Computer History Museum.
computerhistory.org/press-releas...
March 20, 2025 at 9:02 PM
Want to check out the source for the "AlexNet" paper? Google has made the code from Krizhevsky, Sutskever and Hinton's seminal "ImageNet Classification with Deep Convolutional
Neural Networks" paper open source, in partnership with the Computer History Museum.
computerhistory.org/press-releas...
Neural Networks" paper open source, in partnership with the Computer History Museum.
computerhistory.org/press-releas...
Exciting news: @waymo.bsky.social is beginning public service on the Peninsula, starting with Palo Alto, Mountain View, and Los Altos! Initial service area below.
March 12, 2025 at 1:21 AM
Exciting news: @waymo.bsky.social is beginning public service on the Peninsula, starting with Palo Alto, Mountain View, and Los Altos! Initial service area below.
A comprehensive longitudinal analysis showing some quite large increases in various health complications down the road after a COVID hospitalization (about 30% higher all-cause mortality and 35% higher all-cause hospitalization vs. matched controls).
March 3, 2025 at 6:19 AM
A comprehensive longitudinal analysis showing some quite large increases in various health complications down the road after a COVID hospitalization (about 30% higher all-cause mortality and 35% higher all-cause hospitalization vs. matched controls).
The underlying preprint (arxiv.org/abs/2412.06288) on which this is based has some serious methodological flaws.
For example, the paper says "The scope-1 public health impact of AI primarily comes from the emission of operating on-site backup generators."
For example, the paper says "The scope-1 public health impact of AI primarily comes from the emission of operating on-site backup generators."
February 24, 2025 at 12:13 AM
The underlying preprint (arxiv.org/abs/2412.06288) on which this is based has some serious methodological flaws.
For example, the paper says "The scope-1 public health impact of AI primarily comes from the emission of operating on-site backup generators."
For example, the paper says "The scope-1 public health impact of AI primarily comes from the emission of operating on-site backup generators."
My dad bought an IMSAI 8080 kit computer when I was 9 (~2 years before the Apple II came out).
Although it was satisfying entering 8 bits of info one toggle switch at a time, productivity improved considerably once we got an actual keyboard! (Then I could type in BASIC games from Ahl's book)
Although it was satisfying entering 8 bits of info one toggle switch at a time, productivity improved considerably once we got an actual keyboard! (Then I could type in BASIC games from Ahl's book)
February 18, 2025 at 3:05 PM
My dad bought an IMSAI 8080 kit computer when I was 9 (~2 years before the Apple II came out).
Although it was satisfying entering 8 bits of info one toggle switch at a time, productivity improved considerably once we got an actual keyboard! (Then I could type in BASIC games from Ahl's book)
Although it was satisfying entering 8 bits of info one toggle switch at a time, productivity improved considerably once we got an actual keyboard! (Then I could type in BASIC games from Ahl's book)
Beavers are dam good workers! Saved humans having to build a 1.2m€ dam.
February 15, 2025 at 11:38 PM
Beavers are dam good workers! Saved humans having to build a 1.2m€ dam.
February 15, 2025 at 6:47 AM
Delighted to be a minor co-author on this work, led by
Pranav Nair: Combining losses for different Matyroshka-nested groups of bits in each weight within a neural network leads to an accuracy improvement for models (esp. 2-bit reps).
Paper: "Matryoshka Quantization" at arxiv.org/abs/2502.06786
Pranav Nair: Combining losses for different Matyroshka-nested groups of bits in each weight within a neural network leads to an accuracy improvement for models (esp. 2-bit reps).
Paper: "Matryoshka Quantization" at arxiv.org/abs/2502.06786
February 11, 2025 at 5:41 PM
Delighted to be a minor co-author on this work, led by
Pranav Nair: Combining losses for different Matyroshka-nested groups of bits in each weight within a neural network leads to an accuracy improvement for models (esp. 2-bit reps).
Paper: "Matryoshka Quantization" at arxiv.org/abs/2502.06786
Pranav Nair: Combining losses for different Matyroshka-nested groups of bits in each weight within a neural network leads to an accuracy improvement for models (esp. 2-bit reps).
Paper: "Matryoshka Quantization" at arxiv.org/abs/2502.06786
Another example of what cutting science funding is doing to our leading university research programs in the U.S.: dismantling things like the Soybean Innovation Lab at UIUC, which have made US crop yields dramatically higher.
February 9, 2025 at 7:14 AM
Another example of what cutting science funding is doing to our leading university research programs in the U.S.: dismantling things like the Soybean Innovation Lab at UIUC, which have made US crop yields dramatically higher.
February 9, 2025 at 3:17 AM
OK! My Google colleague Thang Luong shared some exciting updates about AlphaGeometry2!
AG2 now has surpassed the average gold-medalist in solving Olympiad geometry problems, w/ a solve rate of 84% compared to 54% previously!
Paper: arxiv.org/abs/2502.03544
See full list of authors on link
AG2 now has surpassed the average gold-medalist in solving Olympiad geometry problems, w/ a solve rate of 84% compared to 54% previously!
Paper: arxiv.org/abs/2502.03544
See full list of authors on link
February 8, 2025 at 7:42 PM
OK! My Google colleague Thang Luong shared some exciting updates about AlphaGeometry2!
AG2 now has surpassed the average gold-medalist in solving Olympiad geometry problems, w/ a solve rate of 84% compared to 54% previously!
Paper: arxiv.org/abs/2502.03544
See full list of authors on link
AG2 now has surpassed the average gold-medalist in solving Olympiad geometry problems, w/ a solve rate of 84% compared to 54% previously!
Paper: arxiv.org/abs/2502.03544
See full list of authors on link
We launched a bunch of Gemini 2.0 models today. Compared to the 1.5 series models, each of the 2.0 models is generally better than the "one size up" model in the 1.5 series.
2.0 Flash & Flash-Lite set new standards in the quality/cost Pareto frontier.
More details:
blog.google/technology/g...
2.0 Flash & Flash-Lite set new standards in the quality/cost Pareto frontier.
More details:
blog.google/technology/g...
February 5, 2025 at 11:45 PM
We launched a bunch of Gemini 2.0 models today. Compared to the 1.5 series models, each of the 2.0 models is generally better than the "one size up" model in the 1.5 series.
2.0 Flash & Flash-Lite set new standards in the quality/cost Pareto frontier.
More details:
blog.google/technology/g...
2.0 Flash & Flash-Lite set new standards in the quality/cost Pareto frontier.
More details:
blog.google/technology/g...