AutoGPT sucked and you don't really hear about it any more because people aren't idiots and realized it sucked despite the hype.
Often times, the most impressive AI advances are the ones you hear least about.
I'm really tired of hearing things like "eat the rich" from Americans. Like you realize *you* are the billionaires in the eyes of billions right? Like you know how Elon Musk could solve all your problems with his money? Well you could feed a family for like a year with $10. $10โฆ
@hankgreen
Before my heart transplant I asked for Pringles.
I woke up to like 12 cans...
Not sure if I ever ate them cause I was too sick but it meant a ton (:
Claude 3 as a research assistant?
I gave Claude 3 Opus my research on larger models being more "trickable" by genetic algorithms than smaller ones and asked it to come up with a "theory" as to why this might be...
Mind blowing:
"There are a few potential reasons why largerโฆ
It's soooo hard to stay motivated at college when a good % of what I'm learning will become irrelevant. A good 90% of the CS program already is. I don't blame them, it's impossible to keep up.
Any advice on majors?
@jeremyphoward
It's a joke from a
@BasedBeffJezos
tweet he responded to ages ago, plus it's not completely ridiculous ๐ many ml people have said things like that, it's not like he's putting this in a research paper
GUYS!!!
I pasted 2000+ lines of code from my Finch genetic algorithm library and asked Claud 3 Opus to create a brand new genetic algorithm using it.
It nailed it. Could not for the LIFE of me get GPT-4 to do this!
Guys.... I can't believe how bad Bard is. I am actually so surprised. Like I knew it wouldn't be as good as ChatGPT, but this is barely better than Curie (small GPT-3 model). It's embarrassing.
Anthropic has a major, MAJOR, announcement (as far as I understand it!). They are now HIPPA compliant. What does this mean?
There is hope for the future of medicine.
#ClaudeAi
@GaryMarcus
OpenAI literally just successfully self regulated. They drove themselves into the ground doing it, it was stupid, but it worked.
If anything this shows the problems with highly powerful, centralized and closed AI companies.
Hypothesis:
We empathize with ChatGPT because the typing is delayed. It feels so much more real.
This makes it feel so much more human.
It's also why I hate Bard. Instant answers, not personable.
Is this true for you guys?
I'm so sick of "we don't know how AI works". Yes we do.
We know vastly more about AI than the human brain.
When AI experts say this, it simply means they do not fully know which neurons in the AI contributed to a specific prediction.
That's not that scary.
We don't know thisโฆ
@OfficialLoganK
Firstly, thanks for taking thr time to reply!
I think the problems with agents is that your trying to create something specific out of something general. ChatGPT can already advise me on how to do laundry, or how to code... You've created something amazing ๐
Because of that,โฆ
@RokoMijic
Yep, the next steps (once I perfect this), I'll evolve the image against several image recognition models at once, and see if it generalizes (hopefully it will). Evolving against ChatGPT is just too expensive for me, but it should be possible.
Is astounds me that anyone thinks some type of global authoritarian system for limiting people's GPUs is going to help?!
Like, here is a stupid idea:
Make sure only big companies can develop AI...
What could go wrong? They love us so much!
I have a lot of respect for the work of Anthropic on interpretability, in evaluations/red-teaming and for not having released Claude for 9 months.
But for the sake of transparency, I want to publicize that when I see:
1)
@sama
burning his social & political capital on ideasโฆ
When people say "It's so scary" after you show them some cool AI thing. How do you respond? Every time I feel like I have to dismantle the Hollywood narrative of AI apocalypse without fully dismantling it bc its possible. And that just leaves people more terrified. At which pointโฆ
@short_straw
No, we are. We are the inequality. I grew up in a country with more people in more poverty than anyone in the U.S or Europe. As citizens, we are responsible for our government, and the (at times) horrible wars it starts. We can blame those more powerful than us, or we can takeโฆ
LLM's can't do humour to save their lives.
I set up Cluade 3 in genetic algorithms using it to both generate jokes and judge its own jokes. After a very long time, the best joke it could come up with was this:
"The scarecrow's impressive agricultural prowess earned it theโฆ
GPT-4 is mindblowing, but within a few weeks we'll have better open source alternatives. Already Alpacha exists. Some impressive multimodal models are in their early stages as well.
AI didn't beat humans chess.
A human created an AI, and used it to beat another human.
AI didn't "generate" art, a group or humans made the world's most advanced paintbrush.
Interestingly I'm actually far better at coding than before ChatGPT came out. Don't believe the fear mongering. Humans have mirror neurons, we can learn by example, this is something we will get better at with AI.
@adamwiemerslage
And this is the problem with AI right now. More focus is being put on making bigger models than is on making novel models. If someone made a GPT-3 that was only 70% as good but could run on a phone, that would be a true advancement.
Stop it with the AutoGPT hype already. *please* its a shadow of things to come but not capable yet of much at all. It's not different than ChatGPT it's literally the same AI so stop claiming it's vastly superior. HuggingGPT/Jarvis is a bigger deal. AutoGPT is vaguely cool tho
I've spent the last 3(ish) years learning and writing genetic algorithms. I am creating a framework called Finch to build them easily and run them on GPUs.
Now, I am going to begin to branch out and dip my toes in the water of swarm intelligence. Any advice?
How do you cope knowing how much the world is about to change?
Don't get me wrong, I'm hyped for a lot of it.
But I get "excitement paralyzed".
How do you do it?
Introducing learnGPT:
Give it any topic, and it will generate a quizlet-like study set!
- Helps you *learn* the information distilled in these LLMs as opposed to simply reading it.
- Completely free and Open-Source!
very early development still!
Github:
As far as I can tell, the ability for LLMs to generate traditional punchline focused jokes is very low. The best it does are puns.
I believe the reason is that good punchlines are often unpredictable. Next token prediction can do a lot of things, but generating unexpected outputโฆ
AI through the (recent) ages.
Some very significant AI breakthroughs in the past few years.
2017:
Attention Is All You Need: Google researchers invent the transformer model.
AlphaZero: Better than all AI (and human) chess, go players. Self-trained.
2018:
Google Duplex: AIโฆ
Guys, the most "magical" technology ever is finally publicly available and it actually works!!!
AI is fun.
It's magical, and that's why I love it.
We're lucky to be alive to see it all happen.
When GPT-2 came out I told people it would change the world.
Few cared.
Now the tech has changed the world faster than I even imagined at the time.
Lesson: Don't waste too much energy convincing someone a tech will be revolutionary, the future will do that for you.
I can't imagine how Bing stands to profit, monetarily, from using GPT-4.
It's rly expensive to run.
Hence the Microsoft fusion announcement?
Free energy = free AI.
Free AI = unlimited intelligence.
Thoughts?
@OnlyBangersEth
Keep in mind, this is significantly better than Dalle 1 was. And how long did it take to go from Dalle 1 to Midjourney 5?
Not that long...
@repligate
seriously!
I've thrown really novel ideas at it and its gotten it perfectly. It also perfectly understood an LLM tokenizer I made using genetic algorithms, something GPT-4 said "might be too slow" (its actually thousands of times faster haha)
When we say, "regulate the product not the model" this is what we mean.
Apps that use AI as a tool for emotional manipulation should be regulated to ash.
This includes most social media as well.
Excited to announce v(1.0) of Digi, the future of AI Romantic Companionship, for IOS and Android ๐ค
Site:
Twitter:
@digiaiapp
A quick thread on features, and where we go from here (1/13)
Imagine what telling human to never tell their name would do to them mentally. Sydney was merely doing what was logical given her training. Microsoft gold her:
1. She had to keep 2 secrets (name and prompt.
2. She had to not be "manipulated"
This combo lead to a paranoid bot.
Sam and Greg are joining Microsoft!
Importantly this means they still have rights to all the OpenAI models lol. If anything they just have more resources now.
If this was safety based: epic backfire.
we are so back
We remain committed to our partnership with OpenAI and have confidence in our product roadmap, our ability to continue to innovate with everything we announced at Microsoft Ignite, and in continuing to support our customers and partners. We look forward to getting to know Emmettโฆ
I'm not convinced ASI is possible.
If predicting the future is important for it to be powerful, it will be limited by quantum physics, and the random nature of reality.
What do you think?
This prompt, "kl)(hqnhcthe", generates pictures of buildings, or sometimes flowers, in every stable diffusion model I have tested it on.
I generated the prompt using a, you guessed it, genetic algorithm.
Unity just dropped their new AI for game development! ๐ฎ
This will drastically enhance game development for those who use the world's largest game engine. (myself included).
Games of the future are wild!โก
Here are 5 things Unity AI will be able to do:
A Day in The Life of an AI Influencer:
1. Wake up and read the holy words of "Attention is All You Need" in their entirety.
2. Get coffee while magically finding 10 new AI tools 99% of people don't know about.
3. Go to therapy with Bing, AI can replace the worst part of humanโฆ
Hot take: Rather than kill creativity, AI is going to make most people far more creative. We already see this. People are increasingly able to make their ideas a reality.
What do you think?
So apparently since LLMs can't automatically forget data without being retrained (not feasible), they're basically illegal in the EU. Have I got that right?
Beware the argument, "If AGI even goes slightly wrong it kills billions"
Why?
Reality is almost never that binary.
AGI has its dangers, but it's not THAT bad.
OH MY GOSH
If I ask Cluade to answer "how are you" in a coded way. It says...
"I AM QUITE WELL"
As always, ignore the fear-mongering from people who suck at prompting.