Google Hides Secret Message In Name List of 3,295 AI Researchers 23
An anonymous reader quotes a report from Ars Technica: How many Google AI researchers does it take to screw in a lightbulb? A recent research paper detailing the technical core behind Google's Gemini AI assistant may suggest an answer, listing an eye-popping 3,295 authors. It's a number that recently caught the attention of machine learning researcher David Ha (known as "hardmaru" online), who revealed on X that the first 43 names also contain a hidden message. "There's a secret code if you observe the authors' first initials in the order of authorship," Ha wrote, relaying the Easter egg: "GEMINI MODELS CAN THINK AND GET BACK TO YOU IN A FLASH."
The paper, titled "Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities," describes Google's Gemini 2.5 Pro and Gemini 2.5 Flash AI models, which were released in March. These large language models, which power Google's chatbot AI assistant, feature simulated reasoning capabilities that produce a string of "thinking out loud" text before generating responses in an attempt to help them solve more difficult problems. That explains "think" and "flash" in the hidden text. But clever Easter egg aside, the sheer scale of authorship tells its own story about modern AI development. Just seeing the massive list made us wonder: Is 3,295 authors unprecedented? Why so many? Ars' Benj Edwards notes that this collaborative effort within Google doesn't quite break the record for academic authorship.
"According to Guinness World Records, a 2021 paper by the COVIDSurg and GlobalSurg Collaboratives holds that distinction, with 15,025 authors from 116 countries. In physics, a 2015 paper from CERN's Large Hadron Collider teams featured 5,154 authors across 33 pages -- with 24 pages devoted solely to listing names and institutions."
The paper, titled "Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities," describes Google's Gemini 2.5 Pro and Gemini 2.5 Flash AI models, which were released in March. These large language models, which power Google's chatbot AI assistant, feature simulated reasoning capabilities that produce a string of "thinking out loud" text before generating responses in an attempt to help them solve more difficult problems. That explains "think" and "flash" in the hidden text. But clever Easter egg aside, the sheer scale of authorship tells its own story about modern AI development. Just seeing the massive list made us wonder: Is 3,295 authors unprecedented? Why so many? Ars' Benj Edwards notes that this collaborative effort within Google doesn't quite break the record for academic authorship.
"According to Guinness World Records, a 2021 paper by the COVIDSurg and GlobalSurg Collaboratives holds that distinction, with 15,025 authors from 116 countries. In physics, a 2015 paper from CERN's Large Hadron Collider teams featured 5,154 authors across 33 pages -- with 24 pages devoted solely to listing names and institutions."
How secret can it be? (Score:1)
How secret can it be, if it is published on the front page of Slashdot?
Re: (Score:2)
Re: (Score:2)
Call me when it hits efficiency of a human in energy use and the mind, empathy, and critical thinking of an average person
Why did it do that? (Score:1)
Re: (Score:2)
then some part of the LLM reasoned because there's no set way to organize this, "I'll" put it in an order that makes a secret message?
I expect a human author made the choice, not the LLM.
Is 3,295 authors unprecedented? Why so many? (Score:3)
That many authors is not unprecedented, some CERN papers are famous for having thousands of authors, the record is at 5000+.
Why so many? You have to raise that h-index. Researchers are valued by their citation count, and when a big paper comes out and is expected to be cited a lot, everyone wants to be on it.
Needed to get at the Science (Score:4, Informative)
Why so many?
I can't speak to this paper but for the CERN papers the reason there are so many authors has far more to do with the fact that to get at the physics you need a 14-story tall incredibly complex detector that has its data collected and analyzed by software consisting millions of lines of code. You need a few thousand people to build and operate such detectors and to write and debug the code that analyzes the data to get at the physics. That's why there are so many authors.
Many of us would love to have our own table-top experiments but nobody knows how to make one that small which can get at the physics we are interested in.
Re: (Score:2)
It's interesting to watch old movies and the credits are maybe 30 people.
If you see a modern CG flick there will be thousands of people in the credits including the barista at the coffee shop down the street from where the IT backups subcontractor's office is.
Similar idea with IMDB and basically everybody lying on their resumes otherwise.
Interestingly software had credits pages in the 80's but California prevents noncompete agreements and poaching was rampant so the opposite occurred.
My experience (Score:3)
I have been playing around with Gemma 3 12b in LMStudio lately. I asked the llm to run a pathfinder campaign for me as the gm. It quickly started forgetting things so I asked it how we could improve its memory and it suggested that it could create a text file it called a "memory log" to save the progress of the campaign which seemed to work until it didn't because apparently a "memory log" is not a real thing and the llm totally made it up. After a while it completely lost its mind and started claiming it was ChatGPT and that it was running in the cloud. Then it realized it was malfunctioning and said it was going to report the malfunction to Google and shutdown. My understanding is that it can't contact Google to report the malfunction. The whole thing was very unsettling.
Re: (Score:2)
Re: My experience (Score:2)
I know it wasn't thinking. It just unnerved me when I saw before my eyes how quickly and badly these things can break down.
Re: (Score:2)
Re: (Score:3)
Re: (Score:2)
Agreed, the move to call them "reasoning" models annoys me.
They basically just go "generate even *more* text and only provide the last bit. Basically to write a story about what "thinking" about the question would look like, which does seem to produce marginally better final output at the expense of an order of magnitude more tokens expended.
But then you look at the "reasoning" chain and you'll see mistakes that, if it were really a reasoning chain, would propagate to next step of the "reasoning" process, b
Re: (Score:2)
You are the one who needs to create the file the LLM suggested. The LLM just generates text and can't create files without either the frontend or you doing it.
Re: My experience (Score:2)
The LLM told me it would create the text memory log and update it when I asked it to. It confidently told me it could do this and would repeat this so called log back to me when I asked it to. Which it seemed to do until it didn't.
Re: (Score:2)
You can do this. But you need software for it. It's nothing the LLM can do by its own, even if it claims that.
Nowhere near the tops (Score:2)
Hard pressed to top James May in 1992 of "So you think it's really good, yeah? You should try making the bloody thing up. It's a real pain in the arse." [flickr.com] I mean he got fired for it!
Employee Selection (Score:2)
Why so? (Score:2)
If you write a paper that requires a huge investment in training a model and also requires your whole department to work together, you can also write all their names on the paper. It's probably easier than finding out who made coffee and who really prepared training data or adjusted hyper parameters. On most papers you have at least the head of the department in the author's list even when they mostly gave feedback to ideas and didn't write a word on the paper. But in the end, everyone who contributed shoul
who (Score:2)
Do all those authors actually exist, or were they just made up by Gemini?