Last month, OpenAI provided ChatGPT with new o1-preview and o1-mini models that it said would “spend more time thinking before responding” – and it seems Google isn’t far behind with its own deep-thinking AI.
According to BloombergGoogle is also developing artificial intelligence that more closely resembles humans’ ability to reason, most likely because of its Gemini platform. For now, Google has not made any official comment.
Because they don’t come up with an answer as quickly, these models are said to be better at solving multi-step problems in areas such as math and coding. Judging by the limitations OpenAI has placed on the o1 models, it also consumes more processing power and energy.
Bloomberg doesn’t say when Google might launch its thinking AI, but progress has apparently been made “in recent months,” easing concerns within the company that it would fall behind OpenAI. And that sounds to us like we’ll see some of this technology sooner rather than later.
I think, that’s why AI
Technically, these newer models use what’s known as chain-of-thought prompting: the AI essentially goes through several steps and considers multiple responses, debating with itself the right solution.
With ChatGPT and the o1 models you get the message ‘think’ while the AI works through it all. Based on the tests we performed, it is better than the GPT-4o model for solving riddles and solving more complex problems.
Presumably whatever Google has in the pipeline will follow a similar approach. For now, OpenAI has limited the o1-preview and o1-mini to users who pay for a ChatGPT subscription, so it will be interesting to see if the Google equivalent requires a Gemini Advanced subscription (yours for $19.99/ £18.99 / AU$32.99 a month).
According to Bloomberg, Google’s AI development plans are being slowed by ethical concerns as they want to appear trustworthy and have competing interests within the massive organization – but with OpenAI, Meta and Microsoft all busy with their own AI upgrades, the pressure is on.