Pichai admits Bard feels like a ‘souped-up Civic,’ promises upgrades

Google's Bard didn't exactly get a standing ovation when it launched, and Google CEO Sundar Pichai is now publicly owning up to it.

Speaking on The New York Times' Hard Fork podcast, Pichai offered a surprisingly candid metaphor: Google essentially took a souped-up Civic and put it in a race with high-performance supercars.

Bringing a Civic to a supercar race

Pichai explained that the version of Bard released to the public on March 21 runs on a lightweight, efficient build of LaMDA. While that model was optimized for conversational flow, it quickly stumbled when users pitted it against ChatGPT and Microsoft's Bing chatbot for complex tasks.

Independent testing showed Bard was consistently less imaginative, less fluent, and struggled to surface accurate sources. Pichai defended the conservative rollout by citing internal caution, noting that Google did not want to push a more capable system until it felt confident it could handle the fallout.

Pichai also addressed internal rumors. While he denied officially issuing an internal code red, he acknowledged that employees were sending emails using that exact phrase. He also confirmed co-founder Sergey Brin has been spending significant time working directly with the engineers.

Upgrading Bard to PaLM

The fix is coming quickly, according to Pichai. Bard is being upgraded to more capable models within Google's PaLM family, with visible improvements expected over the coming week.

PaLM is a much larger and more modern system than LaMDA. Google says this switch will give Bard a noticeable boost in multi-step reasoning, mathematical problem solving, and computer programming.

Pichai also weighed in on the open letter calling for a six-month moratorium on advanced AI training. While agreeing that safety concerns are valid, he suggested the smartest path forward is adapting existing domain-specific regulations, like privacy and healthcare rules, rather than rushing out blanket AI legislation.

Why it matters

Google got caught flat-footed, and this interview feels like an attempt to reset the narrative before the gap with OpenAI widens permanently.

For users, Bard in its launch state simply was not worth switching to. If you rely on AI for writing code or tackling technical questions, ChatGPT has been running circles around it. The incoming PaLM upgrade will be the real test of whether Bard belongs in your daily workflow.

Strategically, Google's extreme caution backfired. By releasing an intentionally neutered model to avoid PR disasters, it suffered a different PR disaster: appearing technologically behind. Framing the launch as driving a Civic was a clever piece of expectation management, but it also painted Google into a corner.

If the PaLM-powered Bard fails to match GPT-4 next week, Google will no longer be able to claim it was just holding back the good stuff.