3 ms·
Who knew the real revenue unlock wouldn’t be based on how much paranoid red-teaming the model underwent to resist users jailbreaking its ‘alignment’, and instea
by firasd 2mo ago
Who knew the real revenue unlock wouldn’t be based on how much paranoid red-teaming the model underwent to resist users jailbreaking its ‘alignment’, and instead more on whether the model is post-trained to use ‘sed’ and ‘git’? Poor Gemini
Scientist-heavy orgs that want to solve everything in token space may overtook tool use; meanwhile Anthropic has been super focused on MCP, Claude Code etc for over a year
- thorncorona 2mo agoGemini isn't as heavily aligned as OAI / Ant models ..?
- firasd 2mo agoIf you just talk to it over API (no web search) the Gemini models are extremely resistant to thinking the user may be living in a universe outside their training data. Try to discuss any news etc and they assume it’s fake or fiction
- RealWed5 2mo ago[dead]
- croemer 2mo agoYes, Gemini let's me do SARS-CoV-2 evolution research (perfectly safe, should never be blocked but is impossible with OAI/Ant)
- raincole 2mo agoOut of the three main US AI companies' models, Gemini is obviously the less aligned (read: censored). So I really don't know what you're talking about.
- Der_Einzige 2mo agoShhhhhhhhhh!!!!! The more people find out that gemini 3.1 pro on the google AI playground or via API is de-facto uncensored, the more likely google will put up actually working guardrails and end the fun for everyone!
- biophysboy 2mo agoWhat do you mean by scientist-heavy orgs solving everything in token space? I feel like they use them to make or run tools almost exclusively.
- firasd 2mo agoI'm saying AI researchers have a bias towards thinking what needs to happen is prompt -> [crunching tokens] -> response rather than prompt -> [orchestrates 5 tools] -> response In other words 'just add a calculator tool' is not as sexy research-wise as making the model accurately eyeball arithmetic in its chain of thought. Maybe I'm wrong but that seems to be the case