Welcome to the 2023 season finale of our weekly roundup of AI news.
This week AI solved a centuries-long art mystery.
An AI pigeon knows where you were last summer.
And a sales chatbot has been selling Chevys for $1.
Let’s dig in.
AI steals the headlines
OpenAI seems to be collecting lawsuits faster than it changes CEOs. The New York Times has decided to sue OpenAI and Microsoft over claims that the companies breached its copyrights.
Lawyers for the tech companies will have to convince the judge that grabbing NYT content, training your AI on it, and dishing it out to users without paying for it is perfectly acceptable. The damages The Times is claiming could change the AI industry completely.
Will we have to stop AI from reading papers or ingesting any of the content we see around us?
I do wonder how the law/society will deal with embodied robots & their constant training 🧠 🤖
Will they have to close their eyes/sensors every time they hear a song or see a painting or read a newspaper 🤔
Look it up on a ledger? 📒 👀
Genuinely curious 🧐 pic.twitter.com/OqQuCWGHo2
— Emad (@EMostaque) December 28, 2023
Researchers had some good news for OpenAI this week. After running benchmark tests on Google’s newly released Gemini Pro model they found that it can’t even beat GPT-3.5 Turbo.
Google engineers will probably say, ‘Yes, but wait until you see how good Gemini Ultra is when we release it,’ while Sam Altman’s finger hovers over the ‘Release GPT-5’ button.
AI sees the big picture
Fake images were around for a long time before AI came along. Art historians often have to work out if a painting was really painted by the artist that people claim it was. AI was used in an interesting way to solve the mystery behind a famous Renaissance painting.
Claims that the painting was the work of Raphael were disputed and the results of the AI analysis were surprising.
We’re generally quite careful about dishing out our private personal information online, but we’re overly generous in sharing our holiday snaps and selfies. Researchers from Stanford University created PIGEON, an AI model that analyzes your photo and can tell where it was taken with remarkable accuracy.
If you decide to put down the camera and have AI make your pictures for you then you’ll be happy to hear that Midjourney just got an upgrade.
It just unveiled the sixth version of the AI image generator. It’s an Alpha release so some improvements are still expected but the first images users are generating look amazing.
Who’s minding the gate?
Last week we reported on how the RAND Corporation had a big influence on President Biden’s Executive Order on AI safety. Now it seems that the think tank with very partisan funders may still be involved in how AI safety research is conducted.
A US Congress committee is concerned about how and by whom the research is being done. Should they be?
AI may not be the existential threat that many effective altruists claim it is, but it does present society with new dangers. Stanford researchers identified thousands of illicit child imagery in the LAION image dataset.
The dataset was used to train a host of AI models, including the popular Stable Diffusion models. The presence of child porn images in the training dataset means the models inadvertently learned how to generate similar material. This problem may not be fixable.
Dollars and sense
Anthropic doesn’t grab AI headlines like some of the other big hitters but the company behind the Claude 2 chatbot has certainly got the attention of investors.
It received a few billion dollars from Amazon and Google earlier this year and is preparing to launch a $750 million funding round. What exciting AI tech will all of that money deliver?
If Anthropic creates a supersmart AI sales chatbot then General Motors may be interested in it. Some of the company’s Chevrolet dealerships have been using ChatGPT-powered sales chatbots to close deals with customers online.
The only problem is the chatbot could be tricked into selling cars for $1 or giving them away for free if you pretended to be Sam Altman.
Some folks who may be looking for a cheaper vehicle are the ad sales staff at Google. The company is talking about “restructuring” but it looks like 30,000 Google jobs could be at risk as AI continues to replace ad sales staff.
Resistance is futile
It turns out that handing out antibiotics like Tic Tacs for decades wasn’t a good idea. Drug-resistant bacteria have rendered a lot of antibiotics pretty much useless. Finding a new antibiotic that kills bad bacteria but doesn’t harm us involves evaluating millions of potential molecular structures.
A team of MIT researchers trained an AI model to do just that and they have identified a new class of antibiotics that could save thousands of lives.
Exciting AI-powered discoveries like these need a lot of computing power and NVIDIA is still way out in front of the GPU tech race. If you want to be the first to see the latest tech the company has in the works then you may want to book your seat for the NVIDIA 2024 GTC AI experience.
In other news…
Here are some other clickworthy AI stories we enjoyed this week:
- Sam Altman and Jony Ive poached Apple’s iPhone design chief to work on their AI device project.
- OpenAI is looking to raise more funding which will take its valuation to $100 billion.
- AI deepfakes go mainstream as experts warn of the impact on upcoming Presidential elections.
- Chinese mourners are using AI to bring back the dead.
- The @futuristflower X account is either run by someone in the know or a wild speculator. But it’s followed by Jimmy Apples so… Potential lab leak in the future?
Since there is some kind of confusion, text-davinci-004 is the raw base model of GPT-4 without any RLHF. It is so extremely powerful (and completely uncensored) that they will never give you access to it.
— Flowers from the future (@futuristflower) December 27, 2023
And that’s a wrap.
This year was one heck of an AI rollercoaster ride. We were wowed by the early iterations of ChatGPT and then blown away by GPT-4 with vision. We got to turn text descriptions into beautiful pictures, songs, and even videos.
Along with the tech advances, keeping up with AI news felt like we were watching a mashup between Days of Our Lives and Jerry Springer. ‘You’re the CEO! No, I am!’, ‘You cheated! You stole my data!’
What was your standout story of the year? Will 2024 be the year of AGI or is AI about as good as it’s going to get?
My predictions for 2024 are that we’ll see a lightly contested form of AGI, coherent long-form script-to-video, and commercially available AI-designed longevity and cancer drugs. Fingers crossed.
Let us know what your AI predictions are for 2024 and if we missed any good 2023 stories that we should squeeze in before then.
