Welcome to this week’s roundup of AI news written especially for the discerning human reader.
This week AI had us dreaming of reaching new worlds.
Google’s Gemini blew us away.
And machine learning gets a taste for wine.
Let’s dig in.
Is Gemini the new AI star?
The wait is finally over as Google announced the “Era of Gemini” with its new suite of Gemini multi-modal models. The demos look amazing, but is Gemini the GPT-4 killer it’s been hyped to be? Will Gemini Ultra be released just in time to see GPT-5 beat it?
The smallest of the models, Gemini Nano, is really exciting. The functionality this on-device AI makes possible on Google Pixel phones is groundbreaking.
OpenAI’s story is still a closed book
Microsoft’s Brad Smith says Altman’s dramatic firing had nothing to do with safety. Does this mean that the mysterious Q* is perfectly safe or simply doesn’t exist?
The original board’s story is that Altman was less than candid with it. Could it have had something to do with OpenAI’s new multi-million deal to buy chips from a company Altman has a financial interest in?
It’s a good thing OpenAI rehired Sam as CEO or otherwise, this Time Magazine cover would have needed an edit.
— TIME (@TIME) December 6, 2023
That’s not cricket
Media houses apparently haven’t learned from previous failed attempts at using AI to substitute human writers. Sports Illustrated has been caught offside for attributing AI articles to fake AI persona.
AI may not be good at journalism but it’s great at imitating artists. Seven more artists added their names to the resubmitted copyright lawsuit against Stability AI and Midjourney. Stability AI’s CEO regularly shoots his mouth off and there are some awkward Emad Mostaque quotes in the legal filings.
Are you worried that AI could learn to do your job? Meta may just have made that prospect a lot more certain. Meta released Ego-Exo4D, a multimodal perception dataset that is a game changer for AI training data.
You’d think that at least a sommelier’s job should be safe, right? Scientists used machine learning to spot fake wine and identify exactly where a Bordeaux wine comes from. If they need a human wine taster to help in their research I’m happy to fall on that Grenache.
To boldly go
At the risk of appropriating the most famous split infinitive, it’s not just Trekkies that want to boldly go where no man has gone before. Could AI and ML spearhead the journey to colonize new worlds?
Elon Musk said his goal with xAI is to understand the universe. From xAI’s filing submitted to the SEC it looks like Musk is on a funding mission to raise $1 billion for the company.
Or is he? It’s a little confusing and exciting all at once. Is this just Musk being Musk and saying ‘Go fund yourselves’?
Nothing is being sought from anyone
— Elon Musk (@elonmusk) December 6, 2023
Fast forward to AGI
If xAI did get a $1 billion boost it would supercharge the race to AGI. Meta’s AI chief Yann LeCun argues that AGI is far from imminent. LeCun says we’re yet to hit the “cat and dog” level of AI intelligence but there are interesting arguments on the other side of the aisle.
Exciting new research in the use of Fast Feedforward neural networks in language models may have an impact on when AGI is achieved. If you fancy a little dip into some datascience then check out how this new approach could make LLMs 300x faster.
Loose lips sink AI ships
Amazon Web Services announced its B2B generative AI assistant named Q. It’s a corporate AI assistant which is Amazon’s answer to products like Microsoft’s Copilot.
Q is supposed to be better at handling data security and privacy than consumer chatbots. So it was a little embarrassing when Q started leaking sensitive internal Amazon information.
Google DeepMind researchers revealed an extremely easy hack to expose ChatGPT’s training data. They’re not quite sure why this works but it does.
While the world’s governments hold summits to promote AI safety, AI engineers are leaving the keys to the front door exposed in their code. API tokens carelessly exposed on Huggingface and GitHub pose a huge risk to AI companies and users.
AI programmers and their API tokens when they’re done with work:
Across the pond
England once ruled the seas but it now has its eyes on AI ascendency. Microsoft’s planned £2.5 billion investment in the UK AI industry will certainly help them in that cause.
The UK government said it won’t be passing any new AI legislation anytime soon but the EU is on an increasingly desperate mission to drive adoption of the EU AI Act. Will they get it done on time, or at all?
While legislators shuffle papers, businesses are leaning hard into generative AI. AstraZeneca just committed $247m to an AI drug development company to find a cure for cancer.
Stop, collaborate and listen
The growing enthusiasm for AI is reflected in some high-profile conferences and summits held this week.
The GovAI Summit 2023 held in Arlington Virginia focused on AI in the public sector. It’s a timely discussion as this week saw the council of the Brazilian city Porto Alegre passing a ChatGPT-written law.
The Generative AI Europe conference in Amsterdam showcased real-world applications of AI in business.
AI was also featured at the COP28 climate summit in Dubai where the role of AI in combating climate change was discussed.
US-China AI race
The US government is trying to turn off the tech taps that fuel China’s AI aspirations. China has gotten cozy with several tech companies in the Middle East to circumvent this.
This week, the US government ordered a Saudi venture capital firm to divest from an AI chip company called Rain which is backed by Sam Altman. This was in response to potential ties the Saudi firm had to China. Rain is the same company mentioned earlier in which OpenAI is investing millions of dollars.
The inevitability of AI weapons developing faster than AI regulations was reinforced at the Reagan National Defense Forum. There was no subtlety in naming China as America’s main rival spurring on these developments.
The US Commerce Secretary had some choice words for Nvidia at the conference. Nvidia has been developing hardware to bypass the Chinese tech export controls the US recently imposed.
In other news…
Here are some other click-worthy AI stories we enjoyed this week:
- Alibaba Group developed a model that takes an image of a person and creates an animated controllable character.
- OpenAI postpones the launch of its GPT Store to early 2024.
- Creating a single AI-generated image uses as much power as charging your smartphone.
- The Wizard of AI – A documentary about the dangers of AI, made by AI.
- Jimmy Apples giving us a heads up again?
I’m more excited on some open source models coming out soon 😉 but keep an eye out on a potential end of December gpt 4.5 drop and new multimodal from Anthropic. pic.twitter.com/rbMPlo7NbA
— Jimmy Apples 🍎/acc (@apples_jimmy) December 7, 2023
And that’s a wrap.
Have you played around with Bard to see if it’s any better now that it’s powered by Gemini? Once Gemini Ultra drops I think folks might skip OpenAI’s GPT Plus waiting list and head straight to Google.
If we used AI to prepare new worlds for humans do you think the robots would welcome us when we arrive? I think the odds of a welcome mat are 50/50 at best.
What’s your take on the AGI argument? Soon, way off, never, or already achieved?
Let us know, and send us links to any AI stories you enjoyed that we may have missed.