Gemini, a ChatGPT Twin. Google promises a ChatGPT competitor… | by Andrew Lucas | Dec, 2023 – Medium
Andrew Lucas
Follow
—
1
Listen
Share
Disclaimer: No AI was used to generate the written content for this article, although I did try.
Yesterday Google’s parent company Copycat…I mean Alphabet, introduced its AI model 𝗚𝗲𝗺𝗶𝗻𝗶, built to rival OpenAI’s GPT models.
Gemini (latin for “twins”) comes in three flavors, two of which are designed to narrowly outperform OpenAI’s two current models.
Google’s largest model, Gemini Ultra, is purported to beat GPT-4, the model that powers ChatGPT Plus, on all of Google’s favorite benchmarks.
The model will be used in a new platform called Bard Advanced. Google gave few details about Bard Advanced, so we are left to assume it will be similar to ChatGPT Plus.
Despite providing a press release with the main purpose of bragging about Gemini Ultra, Google did not provide a release date. It is still undergoing RLHF fine tuning and is slated for public release “early next year”.
The press release does not state anything about Gemini Ultra being available to developers via an API, although we can assume it will be available eventually, since that’s what OpenAI is doing.
Google has finished training a smaller version of Gemini called Gemini Pro. This smaller model was trained to be similar to GPT-3.5, the OpenAI model that powers the free version of ChatGPT.
You know all those people who say that ChatGPT is not that impressive? They’re using GPT-3.5.
Gemini Pro is available immediately in Bard, Google’s current ChatGPT equivalent. It will be available for developers via the Gemini API on December 13.
One interesting bit from the press release is that Google also trained a tiny version of Gemini to fit onto a smart phone. The press release states that the model will be available in certain features on Google’s line of Pixel phones. This could be a genuine differentiator. It’s unfortunate for those of us who traded-in our Pixel phones for iPhones last month.
Or at least it will have competition “early next year”, when Gemini Ultra is released to the public via Bard Advanced.
The new Google AI model, slated for release in 2024, claims superiority over the current state-of-the-art model, GPT-4, which OpenAI finished training in August of 2022. We may one day see Gemini surpass OpenAI’s GPT models, along with the Pixel phone surpassing iPhone, Google Assistant overtaking Alexa, and Google+ beating Facebook.
Setting aside the sarcasm…
Yes, Google is bigger than OpenAI. But Microsoft is just as big as Google. Microsoft has a larger, more mature cloud infrastructure platform than Google, and it has made that platform available to OpenAI. Microsoft is so well-partnered with OpenAI that it almost absorbed the smaller company by accident a couple weeks ago.
One week ago, we were talking about whether OpenAI had cracked AGI. Is Gemini supposed to impress us? We’re looking for autonomous AI agents, not a 5% improvement over GPT-4. The next step is AI that can think and act without human hand-holding. What we need is Gemini + AlphaGo. Or GPT-5. Whichever comes first.
Lastly, before we all jump on the Google bandwagon, remember Google is an advertising company. Their specialty is subtly selling us stuff as we look for information. Do you want them to take lead building our AI of the future?
—
—
1
Software Engineer | Writer | AI | Automation | Robotics
Help
Status
About
Careers
Blog
Privacy
Terms
Text to speech
Teams
