My week in LLMs #3
You can actually find weeks #1 & #2 as a Twitter thread.
So everything is still going as fast as the last 2 weeks, and it’s hard to monitor everything. So I’ll try to sum up everything that’s caught up my attention
Disclaimer, AI helped with the header image and a couple of phrasings
OpenAI wants to become Apple
OpenAI released a plugin system for ChatGPT; and of course, it’s on a waitlist. Both for users and developers. I think that it’s one of the smartest moves OpenAI could have made to position itself as a key player. We’ll still have to see if they’ve got enough of a headstart for it to pay out.
Because while we’re talking about Apple, they’re still completely silent. But remember that their entire customer base is already equipped with chips tailored for AI usage. So I’m not the only one to expect them to do the usual: make it hard for others to get onto their devices under the guise of security/privacy and release a Siri 2 tightly integrated with their ecosystem to swipe up the user base.
While Google is fumbling far behind
Every preview I see of Google’s Bard (their Bing Chat competitor) paints it as lackluster. Alphabet actually has 2 AI teams, Google and Deepmind, but of course, they were at each other’s throat 🤦
It also seems that Bard was partially trained on ShareGPT, a public page listing conversations people had with ChatGPT… Which could explain it being a subpar ChatGPT3.5.
But let’s remember that it’s Google, so we shouldn’t count them out of the race. Remember that they’ve got ~75% of the mobile market with Android, even with Apple getting a huge head start.
Open source is hitting back
Do you remember that Meta’s Llama model leaked, and they promptly handed out takedown notices to everyone sharing it? Well, one of those takedowns is being fought back against. If successful, it could be a game-changer for the openness of LLMs.
There is also a great thread about why Meta could be botching its “Android” moment by enforcing takedowns instead of embracing open source.
GPT experiments are getting really crazy
Because it’s been almost a month, a lot of cool projects are getting released on a daily basis. But I want to focus on a couple of free/open-source ones you can toy with:
“Literally anything” can produce a functional website with just a few lines of text. You can try it at literallyanything.io
BabyAGI: a self-directed integration of ChatGPT that assigns itself fresh tasks based on a goal. It also blew my mind to think that its creator is actually a venture capitalist by day and didn’t know how to code, illustrating how the release of ChatGPT has democratized AI integration and experimentation.
AutoGPT: another autonomous integration worth checking as it can access commands
And of course, Microsoft had to release JARVIS right before I sent this newsletter. Yet another project aiming for autonomous AIs.
Open letter to stop LLMs development for 6 months
An open letter urging a 6-month moratorium on AI development has recently been circulated, garnering over 1,100 signatories, among them several prominent figures in the field.
While there have been instances of falsified endorsements, the genuine support for the proposal is noteworthy. But some signatories may only be motivated by the desire to slow the race in order to catch up and seize a share of a market projected to be worth $1.5 trillion by 2030.
While I agree with some of the points of the letter, it’s absurd to think it’ll have any effect apart from possibly raising public awareness.
And a tad of gloom
Remember the paper about GPT4’s impact on jobs? Well, Goldman Sachs predicts 300 million jobs to be lost to automation. Not much to add to this one, unfortunately…
Before a pinch of hope
There was an “AI for Good” hackathon, and a group released a tool to detect dangerous moles for skin cancer prevention. So there’s still hope in the world 🤞.
The cat’s out of the bag anyway, so we might as well get ahead and learn to use these new tools as best as possible. Feel free to tell me in the comments if I've missed anything.
See you later for week #4 😉.