Things are getting exciting (again) in the Generative AI space!
ChatGPT is just over a year old (happy birthday), and many of us have gotten used to the magic of large language models. We’ve gotten so used to it that we’re now accusing ChatGPT of laziness.
To appreciate the absurdity of this claim, consider how astounding this “laziness” would have seemed just a year and a half ago! But I get it: we saw what large language models were capable of before they were overly “fine-tuned” by their creators (to ensure no unintended content would be produced by them).
Exploring Alternatives
As a result, many of us, not just researchers but also practitioners, have begun experimenting with other LLMs. I’ve recently started running Llama 2 on my laptop. While it may not always be as robust as GPT-4, the level of control it offers its users makes it a handy tool. Plus, it processes information locally, eliminating concerns about the potential reuse of my prompts.
Easier said than done, you might think: not everyone creates Python environments and runs generative AI models on their machines for the fun of it. There’s some good news coming up!
Google’s New Entrant: Gemini
A few hours ago, Google announced Gemini, a multimodal, multilingual generative AI model that appears highly capable. Check out this video, shared by Google’s CEO Sundar Pichai. It’s well worth a few minutes of your time!
Alongside Gemini, Google is launching AlphaCode 2, a code-generating tool. According to Google, it can comprehend programming challenges and outperforms 85% of developers in such contexts. This is a significant improvement from its predecessor, AlphaCode, which ranked at the median level among software developers (outperforming 50%).
Interestingly, Gemini seems quite good at solving math tasks, where other LLMs typically struggle. Could it assist with your child’s homework or even help prepare your organisation’s budget?
Gemini will come in three flavours. One of them will run on Pixel Phones—we don’t have enough details just yet, but this might suggest local data processing. A crucial feature in some business use cases.
Gemini is said to be included in Bard “on Thursday”, which likely means “on Friday” for those of us here in Australia.
What does it mean for your business?
Google is known for its excellent user experience. If they execute Gemini well, we might see a significant shift from OpenAI’s ChatGPT to Google’s Gemini.
If you already developed governance rules for using generative AI in your organisation, review Google’s new model to ensure your rules consider the capabilities of Gemini.
For instance, the ability to analyse real-time video—just like in the demo above—is of particular interest.
Do your Generative AI Governance Rules cover the case of your employees using a webcam to stream their work environment straight into a generative AI model?
What does it mean for humanity?
I’ll admit it: this header is more of a clickbait, as I won’t pretend to be able to answer this question in a few paragraphs. However, the key takeaway is the intensifying competition in the field. Until now, ChatGPT had no serious rival. Bing’s LLM (Codename Sydney) was a neat experiment but not quite a productivity tool. Google’s Bard almost matched ChatGPT in some areas, but being “almost as good” can make a big difference.
I believe Google’s Gemini, as the first formidable challenger to ChatGPT, will inject some much-needed fresh air into the competition.
Looking Ahead
It’s clear that, just a year after ChatGPT launched, we’re entering a new phase of innovation and competition. The arrival of tools like Gemini and AlphaCode 2 expands our options and pushes existing technologies to evolve faster (let’s see what OpenAI’s response will be!). Staying agile and informed about these developments is more crucial than ever for businesses. It’s an exciting time to be part of this ever-evolving landscape of AI.
Let’s hope all this hype won’t trigger another AI winter.
We have to be careful not to switch places with the rubber duck that AI can be considered these days: https://cdn.gamma.app/ewxm6l34k3c7sds/72be9eaa17004ef5977088a979a25fbd/original/meme-quack-dalle-2.jpeg 😉
Is it likely to be accessible through Bard then?