Welcome to our weekly roundup of premium-grade AI news.
This week OpenAI magically turned words into video but lost the plot.
Google went large with yet another Gemini release.
And giant AI-generated rat testicles made it into a research paper.
Let’s dig in.
Sora for your loss Google
Our 2024 AI wishlist was satisfied in dramatic style when OpenAI introduced us to Sora, its advanced text-to-video model. OpenAI’s developers must have had a quiet giggle when Google announced Lumiere a few weeks ago, knowing what they were about to unleash.
‘Oh, your model makes 5-second clips that look pretty realistic? Cute.’
OpenAI staff with equity in the company are in for a big payday. OpenAI sealed a tender offer that sees the company’s valuation hit $80 billion, triple what it was 10 months ago.
It wasn’t all good news for OpenAI this week. ChatGPT went from lazy to crazy as the chatbot spewed out a bunch of weird outputs in a strange temporary meltdown. It would be nice to know why this happened, but OpenAI hasn’t commented.
Gary Marcus is a bit of a crybaby when it comes to AI dangers, but he asks some valid questions in light of ChatGPT’s momentary madness.
The Great ChatGPT Meltdown has been fixed. Has OpenAI said anything about what caused it?
With society’s increasing dependence on these tools, we should insist on transparency here, esp. if these tools wind up being used in defense, medicine, education, infrastructure, etc.
— Gary Marcus (@GaryMarcus) February 21, 2024
Google in context
Google’s pursuit of AI relevance continued as it announced the release of Gemini 1.5 Pro. Are you struggling to keep up with the different names of Google’s AI products? Yeah, us too.
Gemini 1.5 Pro beats Gemini Ultra in most benchmarks but the real news is the model’s 1 million token context window. This huge context window combined with impressive recall is a game changer, especially for programmers working on big coding projects.
A lot of Google’s income comes from ad revenue which is heavily dependent on its search tool and display ads. Both of those features are under threat from the Arc AI browser. This could be the new way we search the internet, but there are some risks that need to be addressed.
Could Google be the mystery company engaging with Reddit? Reddit has signed a $60 million per annum deal with an unnamed company to license access to its user-generated content.
Guys vs Girls
AI has confirmed what we knew all along; men are from Mars and women are from Venus. A Stanford AI model accurately differentiates male and female brains by analyzing brain scans. It has interesting implications for neuroscience, psychiatry, and arguments with your partner.
‘I just don’t understand you!’
‘Here Babe, read this research paper.’
Maybe those researchers can explain why AI models on OnlyFans are earning 10 times more than the average American.
Biology researchers made an AI balls-up when they included bizarre AI-generated images in the paper they authored.
The paper was apparently peer-reviewed before being published in Frontiers Journal, but we’re not sure how the giant AI-generated rat testicles made it past their scrutiny.
A new AI tool will make reading those voluminous research papers much easier. Adobe is rolling out a new PDF AI chatbot. AI Assistant will be integrated into its Reader and Acrobat products and will have a huge productivity impact for anyone who needs to make sense of long documents.
Fighting fakes
There’s been a lot of talk about fighting AI fakes but the problem is only getting worse. This week a mass deep fake campaign targeting the US election came to light. It looks like China may be behind it, but the images are clunky and people may not even be paying them much attention.
Does the reported low level of engagement mean that deep fakes aren’t as dangerous as we think? Sam explores the surprising effects deep fakes can have on our memories and capacity for critical thinking.
Big tech companies are banding together to tackle deep fake electioneering, but will adding safety measures to their tools be enough?
Some experts think the solution to AI safety is to add hardware regulation into the mix. SoftBank’s CEO just announced that he’s planning a $100b AI chip project. It’s going to be tough to regulate who ends up with access to these chips as more manufacturers enter the fray.
It makes me imagine a future where some dodgy guy in a trenchcoat says, ‘Psst. Hey buddy. Wanna buy some NVIDIA GPUs?’
But I thought $7 Trillion is all you need pic.twitter.com/6pcQXhj9jQ
— Min Choi (@minchoi) February 19, 2024
In other news…
Here are some other clickworthy AI stories we enjoyed this week:
- Air Canada has been ordered to honor a chatbot’s advice to one of its passengers.
- Meta released V-JEPA, a predictive vision model that learns much like we do.
- The Pentagon engages Scale AI to produce a way to evaluate LLMs for military planning and decision-making.
- The Lincoln Project used AI to make a video of Donald Trump’s late father calling him a “disgrace”.
- Elon Musk hints that MidJourney may bring image generation to X.
- ‘Slow Horses’ & ‘One Life’ director predicts a show made entirely by generative AI is only three to five years away.
And that’s a wrap.
Has Google’s Gemini Pro tempted you to reconsider your ChatGPT Plus subscription or are you holding out for GPT-5 with Sora built in? That would be awesome!
Now we’re not saying you do hang out on OnlyFans, but hypothetically, if you did, would you care if the person was AI-generated? Maybe we’ll see human OF models on the picket line soon.
2024 has barely started and we already have Sora. Can you imagine what AI tech we’ll have by the end of the year? Mind. Blown.
If you find a cool Sora video please send it to us. And let us know if we missed any exciting AI news we should have covered.
