What is Gemini?

Brad Carson on AI ethics caveats from tech companies being toothless against DoD use.

“The fig leaves people are putting in front of themselves are no protection at all. You like OpenAI and Gemini both said like, well, we're going to do these things. Well, then there's like a caveat at the bottom where the DoD says we're going to use it for all lawful uses. And all the things I said I don't really love personally. They're all lawful, right? That's the problem.”

Brad Carson explains why DoD 'lawful use' clauses neutralize ethical restrictions that AI companies embed in their government contracts, making those restrictions functionally meaningless.

“The fig leaves people are putting in front of themselves are no protection at all. You like OpenAI and Gemini both said like, well, we're going to do these things. Well, then there's like a caveat at the bottom where the DoD says we're going to use it for all lawful uses. And all the things I said I don't really love personally. They're all lawful, right? That's the problem.”

GPT and Claude models exhausted allocated attack budgets before universal jailbreaks were found, while Gemini and Grok were jailbroken much more easily.

“GPTs and claws much harder to jailbreak. >> to the point where you kind of topped out you the budget that you had allocated before >> finding universal jailbreaks on those models. whereas Gemini and Grock much easier.”

600 Google employees signed a protest letter after the US Department of War's use of Gemini was disclosed.

“The Gemini use by Department of War this week has got 600 Google employees to write a letter.”

Shinka Evolve uses an adaptive prioritization scheme to dynamically route across all frontier model providers (Gemini, GPT, and others) depending on which parent program is being evolved.

“We're not using just Gemini but we're using basically all frontier model providers and figuring out a smart way how to use each model for a given parent, if you have a certain program in some situations it might be better to use a sort of GPT model in other settings it might be better to use Gemini model. And we sort of introduce a sort of adaptive prioritization scheme that can adapt sort of the evolutionary algorithm on the fly.”

Google's Gemini was used by DPRK IT workers to generate fake IDs during a fake DeFi company sting operation, raising questions about AI misuse for identity fraud.

“Some of the fake ID stuff was stamped through Google's Gemini.”

In February 2026, Gemini model usage among OpenCode users exceeded Anthropic model usage for the first time over a four-week period.

“We had noticed for the first time that a bunch of users were just using Gemini way more than they were using the Anthropic models.”

The leading AI model shifted from Gemini (November 2025) to Claude within months, illustrating rapid leaderboard turnover.

“If you'd asked the question, "Which is the top model in November 2025 last year?" it would have been clearly Gemini. Now, if you ask that question, it's pretty clearly Claude.”

Google DeepMind's AlphaEvolve relied primarily on a weaker, faster-inference model rather than the strongest Gemini, challenging the assumption that bigger always wins in agentic search.

“The bulk of the alpha revolve approach wasn't the strongest Gemini model it was actually a weaker model which is faster inference.”

Anthropic crossed 40% of ARR from indirect channels including Bedrock, Foundry, and Gemini agent enterprise in Q2.

“Anthropic crossed 40% of ARR from indirect channels like Bedrock Foundry and Gemini agent enterprise in the second quarter of 2016 with API and B2B making up the vast majority of net new ARR dollars at the labs.”