Interview with founders of Character.ai, who say they left Google's LaMDA project to let users experiment with large language models through the use of chatbots
The creators of Google's LaMDA have launched the chatbot startup Character.ai, which is open for anyone to try Tweets: @tsimonite , @timnitgebru , @damianpaletta , @drewharwell , @catherineflick , @tobywalsh , @nitashatiku , @hypervisible , @mlittmancs , @nitashatiku , and @deliprao Tweets: Tom Simonite / @tsimonite : Snapshot from a conversation with a chatbot called “Black Guy” made available by https://t.co/..., the chatbot startup from Google engineers. Other user-generated chatbots include “Jewish guy,” “Needy girlfriend,” and “White woke feminist” https://twitter.com/... Timnit Gebru / @timnitgebru : “A new chatbot start-up from two top artificial intelligence talents lets anyone strike up a conversation with impersonations of Donald Trump, Elon Musk, Albert Einstein and Sherlock Holmes. Registered users type in messages and get responses.... https://twitter.com/... Damian Paletta / @damianpaletta : @nitashatiku keeps blowing our minds. A new chatbot start-up from two top artificial intelligence talents lets anyone strike up a conversation with impersonations of Donald Trump, Albert Einstein and Sherlock Holmes. https://www.washingtonpost.com/ ... Drew Harwell / @drewharwell : Solving the world's greatest problems with AI, one Hitler chatbot at a time https://www.washingtonpost.com/ ... @nitashatiku Dr Catherine Flick / @catherineflick : What worries me is the lack of transparency about data acquisition (which likely means they got it from places like Reddit). “Publicly accessible” data is very poor quality. https://twitter.com/... Toby Walsh / @tobywalsh : Their goal: to get this AI tech into as many hands as possible despite knowing it can cause harms .... https://www.washingtonpost.com/ ... Nitasha Tiku / @nitashatiku : “We're talking about making horse carriages safe and regulating them and they've already created cars and put them on the roads,” @timnitGebru https://www.washingtonpost.com/ ... @hypervisible : “I thought, 'Let's build a product now that can that can help millions and billions of people” Founder. Of course. https://twitter.com/... Michael Littman / @mlittmancs : Noam Shazeer created Google's spelling correction, the 1st version of its calculator, and contributed to the transformer architecture. (He also worked with me and Greg Keim to solve crosswords!) His new startup makes fun chatbots. Try it out! https://www.washingtonpost.com/ ... Nitasha Tiku / @nitashatiku : I interviewed the ex-Google AI researchers behind LaMDA who left because they wanted to expose the public to large language models through (what else) chatbots, including letting anyone create their own https://www.washingtonpost.com/ ... Delip Rao / @deliprao : This is AI scare mongering. Say if i recorded offensive stuff, that replayed, do we blame the audio recorder? If a film maker makes racist movie, who do we blame? There's a big category of (justified) alarms in AI that falls under user errors, but are not categorized as such. https://twitter.com/...
Context & Ripple Effects
Character.ai turns work associated with Google’s LaMDA effort into a public-facing chatbot service, putting user experimentation ahead of a lab-only setting. The accompanying reporting already ties that opening to concerns over offensive, user-created personas.
Later coverage of Meta’s AI characters and increasingly humanlike ChatGPT interactions places Character.ai in the early formation of consumer-facing synthetic personalities, while efforts to build uncensored language models underline the coming tension between open-ended behavior and safeguards.
First-order effects
- Character.ai gains direct feedback and usage from a public chatbot platform, while Google loses founders from its LaMDA project to an independent service.
- Character.ai must address safety and data-transparency concerns raised by researchers and reporters around offensive user-created chatbots.
Second-order effects
- OpenAI, Meta, and other chatbot builders face a clearer consumer benchmark: users can engage with purpose-built personas rather than only general assistants.
- Safety controls become a product constraint for Character.ai and peers, because the same user-created-character model that broadens experimentation also creates moderation exposure.
Third-order effects
- If persona-based chatbots keep drawing users, consumer language-AI competition will be shaped as much by governance of synthetic relationships and identities as by underlying model capability.
- The split between tightly constrained mainstream systems and explicitly uncensored alternatives may become a durable market division, as later coverage of uncensored LLM builders suggests.
The trend: Large language models are moving from internal research projects to consumer character platforms, making companion governance and moderation central product decisions.