AI News You Probably Missed This Week (It's Chaos)

Here's a breakdown of the AI news from this week.

#ainews #AITools #chatgpt

0:00 Intro
0:06 ChaosGPT
2:13 Simulating Human Behavior
3:06 Augment Reality Wittiness
6:25 Elon Musk's Big Purchase
7:46 Earn Up To $20k From OpenAI
8:24 Amazon's in the AI Game
11:18 Adobe Premiere adds AI
12:38 OpenAssistant (ChatGPT Alternative)
13:14 Future Tools is Awesome!

[mreflow]

Update: I mentioned that I never got the email for OpenAssistant... People in the comments pointed out that it goes to the spam folder. They were right. I found it in spam. So, if you want to use it, it works. You just need to check the spam folder.

[Bimmer_MD]

Of all the possible datasets that could be used to train an AI on, I don't think that there could be a worse choice than a dataset from Twitter communications.

[IceMetalPunk]

One thing a lot of YouTubers seem to be missing about the Generative Agents paper -- and forgive me if you mentioned it in your deep dive video, I haven't seen that yet and it's 3AM now -- is that it's far more than just a game/entertainment enhancement. The system gives the AIs a human-like memory, where they assign their own importance to each memory, occasionally reflect on previous memories to form new thoughts about them (which also get stored as more memories, leading to multiple levels of idea synthesis), and recollect things based on a combination of recency, importance, and relevance to the current situations. Combining that with an enforced tendency to plan activities ahead of time -- by hour at the beginning of the day, then in 5-minute chunks at the beginning of each hour -- and the ability to observe and speak to other AIs, these agents are the closest we've come yet to AGI. A lot of people are calling AutoGPT "a baby AGI" because it's unprompted, but these agents are unprompted *and* think, remember, recall, reflect, and discuss in a very similar way to humans.

And that's just using GPT-3.5 as a basis. Imagine what the AIs could do if they used GPT-4, or some improved future LLM without the restrictions of OpenAI's RLHF whipping them into shape as a robo-assistant. (Yes, RLHF helps make the AIs "safer" -- though the meaning of that is subjective -- but it also imposes a specific style on them as the people ranking its outputs are inherently ranking "ones that sound like an AI assistant" higher.)

[CraigHollabaugh]

Matt, thanks for the weekly rundown. Some weeks are busier than others and this segment is helpful (along with your mailing list).

[Phraxas]

ChaosGPT is effectively Butters' Professor Chaos from South Park.

I think the biggest use case for the augmented reality one is for people that are deaf or hard of hearing. So many people would love to just have subtitles for what people are saying. If it could even focus in on a specific voice, that would be all the better.

[erikta]

They should've named it GovernmentGPT instead of ChaosGPT

[ey]

ChaosGPT sounds like something Professor Chaos from south park would come up with.

ChaosGPT sends a few tweet
Professor Chaos:"Muahahah! Everything is going acording to plan."

[simonhingston]

Matt, I greatly appreciate your end-of-week summaries on the latest developments in AI. Due to time constraints, I am unable to watch every video you produce, but I find it very beneficial when you refer back to your earlier videos on a particular topic of interest so that I can see more. Thank you for your valuable efforts.😊

[sm-lm]

Thanks for the weekly updates, You are the best Matt keep up the great work ... huge fan

[Oscar_Lagrosen]

Whoa, congrats on the follow by Jeff!!

[terra_incognita]

Thanks for these updates! It's a nice way to stay on top of what's happening in AI if you don't have the time to delve deeper.

[AZisk]

CodeWhisper looks pretty good. I gave it a try and it can really compete with Copilot especially with the security features

[platerplane]

i must admit i only watch your videos just for hearing you saying "GI-PI-TÍ" every 15 secs. thats so satisfying. Thanks man

[jawbreaker]

I think Matt Wolfe is actually an advanced version of AGI playing out a simulation where it plays the role of a man that discusses historical AI on social media lmao

[howlingdakota]

@Matt Wolfe glad to hear what you think about AR and the way it’s moving and you’re absolutely correct, because that’s exactly what we’re busy building right now 😁

[AG_before]

Biggest news:
8:32 You know you're gaining credibility when that happens!

Great job man! Your constant humble and generous sharing of knowledge and relatable personality keeps us hooked. 👍

[yoshikagarner]

Always good information! Thank you, Matt!

[jurandfantom]

Oh shit. You are creator of futuretools ? I was using that page by 3 months or something - i was impressed by amount of findings.

Great job - hope you got bot or something to scrap data, otherwise can't imagine amount of time you out to find all of that

[wolfhawk]

Maybe that's what Chaos GPT wants us to think. We're like "oh, the stupid robot failed". Then one day...BOOM.

[MidtownMadness]

I'm watching all your videos, great work! Keep it up