A week in Generative AI: Gemini, Synthetaic & $7tr
A week in Generative AI: Gemini, Synthetaic & $7tr
News for the week ending 11th February 2024
Last week was a big week for Google, but this week was even bigger - they released their Gemini Ultra model to take on GPT-4 and rebranded all their AI under the Gemini banner. Sam Altman’s also got plans for scaling the hardware required for the future growth in demand for AI.
Bard becomes Gemini and Google launches Gemini Ultra
Alongside rebranding all of their AI features under the Gemini banner, Google also released Gemini Ultra, their competitor to OpenAI’s GPT-4. Overall, it looks so far like Gemini Ultra is broadly on a par with GPT-4 and it will be interesting to see if Gemini Ultra can dethrone GPT-4 in the Chatbot Arena leaderboard over the coming weeks.
In other good news, it seems from work done by Ethan Mollick that any prompts designed for GPT-4 work just as well for Gemini Ultra. Hopefully as GenAI models get more sophisticated, prompt engineering won’t matter as much as this might be the first glimpse of that.
Synthetaic uses AI to find patterns in massive data sets
I’m a big fan of synthetic data and believe there is huge potential in it for solving privacy & ethics issues, legal and copyright concerns and all the way through to challenges on data volumes to train larger and larger models.
It’s good to see this area of GenAI start to get more funding and attention, which will hopefully lead to even more exciting use cases for it!
Hugging Face launches open source assistants to rival OpenAI’s GPTs
It’s great to see the open source community start to push the boundaries of how GenAI models can work as assistants. They’re more of a novelty at the moment, but I expect a huge amount of progress in this space over the next 12 months.
Sam Altman Wants 7 Trillion Dollars
That’s a lot of money, but we are currently in a risky situation where the majority of cutting-edge AI chips are designed (NVIDIA) and built (TSMC) by just two companies. This has to change.
I suspect what is really behind this though is that for robotics to truly take off and hit the mainstream, we will need to be able to build billions of AI chips to power humanoid robots in people’s households. That’s at least another order of magnitude more AI chips that we’re currently capable of producing.
AI Ethics News
Long Reads
Ars Technica - AI can now master your music
Bloomberg - OpenAI’s Secret Weapon Is Sam Altman’s 33-Year-Old Lieutenant
“The future is already here, it’s just not evenly distributed.“
William Gibson