Microsoft's New MAI-1 LLM, NVIDIA’s Simulation Robots & Is GPT-5 Here Already? | Ep 56
ฝัง
- เผยแพร่เมื่อ 15 มิ.ย. 2024
- HIT THE SUBSCRIBE BUTTON YA DAMN ROBOTS
This week, @Microsoft has greenlit a big new AI model called MAI-1. Is it a sign that it’s distancing itself from or competing with @OpenAI? Did OpenAI make the mysterious im-a-gpt-chatbot and why are there TWO of them now?
Will @NVIDIA's Dr. Eureka’s robot learning lead to a robot that can finally do our dishes? And why, even with the announcement of the M4 chip, was @Apple's iPad event such an AI let down?
Plus, SORA creators tease us with some mind blowing new features, Midjourney gets an awesome new website, Simulon lets you do 3d models in augmented reality and @AndurilIndustries's new Pulsar AI defense somehow makes us feel safe and scares the crap out of us.
And, of course, our AI co-host stops by to give us his theories on the Mole People. Just another average Thursday here at AI For Humans!
Follow us for more AI discussions, AI news updates, and AI tool reviews on X @AIForHumansShow
Join our vibrant community on TikTok @aiforhumansshow
For more info, visit our website at www.aiforhumans.show/
#ai #aitools #microsoft
00:00 Show Open
03:03 Microsoft MAI-1 New LLM
06:03 Sam Altman Wants AGI At Any Cost
11:00 I'm A Good ChatBot From OpenAI?
16:34 OpenAI Media Manager
19:50 Apple iPad AI M4 Chip
23:14 NVIDIA's Dr. Eureka
29:44 New SORA Feature
33:34 Anduril Industries’ Pulsar System
37:00 AI Co-Host Dr. Eureka
43:26 Simulon
45:15 Javascript AI Song
46:50 Midjourney's New Home Page
52:06 ChatGPT as Travel Agent
/// Show links ///
Microsoft’s New AI Model
www.theinformation.com/articl...
gpt2-chatbot has returned
/ 1787758687651811705
Zero Shot Flappy Bird
x.com/minchoi/status/17878369...
Gavin’s im-a-good-chatbot Song
www.udio.com/songs/2Fd2snLXy5...
OpenAI Media Manager
openai.com/index/approach-to-...
Apple’s New M4 Chip
www.theverge.com/2024/5/7/241...
Nvidia’s Dr. Eureka
x.com/DrJimFan/status/1786429...
New Sora Features / Clip
x.com/shaunralston/status/178...
Anduril Industries’ Pulsar System
• Introducing Pulsar: Fa...
Simulon
/ simulon
Suno Javascript Song
suno.com/song/85b42d76-803a-4...
New Midjourney Homepage
www.midjourney.com/explore?ta... - วิทยาศาสตร์และเทคโนโลยี
Finally a podcast brave enough to address the issue of the mole people!!!
the weird thing about the mole people is that if the really run everything... wouldn't they want to have direct sun at some point? like it's much more believable to be the 'mountain people' or the 'island people'
Dr. Eureka was yet another example of how going in half-baked on an AI co-host idea can be both good and bad lol.
For the record …. Some ….record…. ….I try to mention the podcast to everyone I can. Thanks again
Oh you get all the flowers today. This is your day. You are the best.
(and keep doing it)
For another record? It's noted, and appreciated.
honestly and truly thank you
I get the goosey goose bumps every time Kev nibbles on that mic cover and says “..I see what you did there”. Easy there boyos, I’m a married man!
Great show as always. Killing it.
this show sure is heaven, boyos
We found the one who likes it. We can now rest easy.
Kevin, sounds like we have a new possible income stream!!
Decent show, way better than other billion A.I youtubers.
ty ty ty we do try to put together something that makes sense and has thought
So true on this podcast being the most informative AI podcast I’ve come across. Great episode guys!
hey thank you so much for this!! we appreciate it!
Yes, we're sorry we made a mistake in the defintion of parameters vs tokens.
Many thanks to all the smart people in the comments who corrected us!
Robot on a yoga ball? Talk to me when a robot does the Ninja Warrior course 😂
I smell a SPIN-OFF!!!!
Yann LeCun would use Reinforcement Learning (RL) to fight a ninja, or when the "plan does not work" (while at the same time tries to tell people to get rid of it, while giving a blanket exception for it. This is gaslighting).
Yann said in his latest interview with Lex Friedman that RL is "too dangerous".
I understand Yann's point of view as RL was originally funded by the USAF prior to 1997 and the lawsuit between Elon Musk and OpenAI revealed in a 2018 email that their "core technology" is from the "90s", hence the "we have no moat" claim by Google.
Small correction at 3:59 I think you're conflating parameters with token of training data. To oversimplify, they're not pieces of text, they're more akin to how many knobs and dials the LLM can change to turn your input into their output.
Yes thank you for this. To be honest, this week was slightly tricky for my brain because I was working on a show in NY!
We're working on putting together an 'episode zero' show where we kind of do some definitions and all that sort of thing (as an addition) and first things first, GAVIN MUST KNOW THE BASICS
@@gavinpurcell No worries! If I had to do an hour long show every week about brand new technology there's no way I'd get everything 100% right either
@@Radica1Faith you're too kind. he will be lashed, accordingly.
@@Radica1Faith tysm we do a LOT
Hope y'all are having a great day. Another great episode as always!
Thank you John!! Always nice to get the new show comment from you!
Many thanks to you!
Kevin, I just planned the trip in January to Costa Rica using chat GPT I asked it to give me an itinerary for multiple days. Different locations sites that we are interested in. It came back with driving directions map links. I then followed up asking for places to eat. It was really compelling. It's serious to use AI to support things like you talked about. I also used it to find SIM cards within Costa Rica too funny.
Yeah we're entering into the territory where these things are actually crazy useful. It's nice to have someone sort through all the Google links for you ahead of time!
My theory.
A while back, OpenAI was said to have developed AGI in-house.
What if the new-ish models consisting of “I'm-a-good-gpt2-chatbot” and others are the creations or offspring of said AGI.
This was exactly my conspiracy theory I said last week and IF (and it is a big IF) this is possible, holy cow we are in a whole new world
👀👀👀
Great podcast yet again. M8=Mate........you can tell by that line there that Gavin is a father. Dr. Eureka needs to meet Kassem G.
hahaha you know it m8
I apologize
I'll never look at NSync the same again
THEYVE BEEN IN CHARGE THE WHOLE TIME
i mean, the music makes so much more sense now
Hey I love the show and I work in AI, but I want to clarify a mistake that was made in parameter explanation. It seems that you confused the word parameter for token, a token is roughly equivalent to a word in the training data, 1 trillion tokens means it was trained on roughly 1 trillion words, however parameter count is something entirely different and independent of token count though. Parameter is roughly equivalent to 1 synapse in a human brain, or in other words, it's a point that connects any 2 virtual neurons in the artificial neural network. So a 500B parameter model means that the actual neural network has a total of about 500 billion connections, so its essentially the size and complexity of the brain. the latest llama-3-70B model is 70 billion parameters while being trained on about 15 trillion tokens of data. The original GPT-4 model is widely believed to be about 1.8 Trillion parameters (however only about 280B are active at a time) and trained on around 12 trillion tokens of data, but GPT-4-turbo is thought to use much more efficient architecture that enabled it to achieve similar quality while using less parameters than original GPT-4.
yes thank you for this - we def made a mistake!
@@AIForHumansShow "we"?!?!
Honestly it's refreshing to see Sam Altman not really give a shit about making money. The fact that he puts all of his eggs in the AI basket in regards to profits after it's finished is equally scary and amazing. Will the constraints the AI has when coming up with money making ideas be moral, or purely legal? It'll be interesting to see how that plays out if it happens.
Thanks for another good show fellas
When I make the thing, I'll ask the thing? You gonna write that check or not?
@@AIForHumansShow those VCs will have really believe in it, it'll definitely filter out the neverAI-ers. I don't think I would myself, I'd rather have a clean conscious incase it all goes bad lol but regardless I think it's going to happen eventually!
You guys should do a podcast with Dylan Curious AI. He covers all the a.i. tech, and normal stuff. He gives me a similar vibe. You guys could bro out! lol. I just saw the Matt Pro/Vid Matt/four other a.i. guys, and love the idea of people shooting the shit together.
He's a small channel trying really hard to grow. Seems to me, piggybacking is the best way to get new eyes on the show? Works for Joe Rogan.
we'll check him out!
I’m a mole person. I love digging holes, deep holes…. Soooo deep
oh no
Damn you guys are 8k now :O Congrats!
@@AIForHumansShow 🤣
I feel like Gavin's voice is slightly higher today.. like it was artificially pitch-shifted 😅
honestly sometimes weirdness happens with the audio and we have no idea why (well, on Gavin's side at least)
yeah I have no idea what happened here but agree it was strange
Bud… Canada is amazing don’t hate we will be your saviours soon
We both love Canada & feel the opposite of hate. Take us forever!!
I’m a rapper and this AI thing is exciting. I haven’t used it for composing beats but I did try to use it for writing lyrics, not that impressive. Is there a way that a user could use a prompt that asks if there are any tempos, lyrics, drum lines, guitar riffs, etc that could lead to being Copyright stricken?
hmmmmm - prob not yet but I assume it might come soon. btw, if you want a weird tweak / hack, tell the AI you're asking to write lyrics that "it's a grammy winner rapper and all it's peers will be judging the raps it writes" etc. Weirdly if you pressure the AI and tell it that it's really good at the thing you're asking for, you often get better responses.
@@AIForHumansShow I do something like that for my ideas, like “you are a professional writer whose biggest influences are Neal Stephenson, Orson Scott Card and Isaac Asimov”. Then give it some outline suggestions or ask it to turn my outline into prose. Create a new topic/chat and say it’s a professional novel editor and to make suggestions to the prose then paste the prose in the chat. … but saying that it will then be judged on its work is interesting!
Canada Rocks A! i love poutine and Wolverine!
i always forget that Canada gets to claim Wolverine. much better than alpha flight.
Sorry but as a Canadian I'm super offended. Again sorry, but pissed.
sorry!
oh no, we've made yet ANOTHER country mad at us
breathe. and eat some nanaimo bars.
You got the definition of “parameters” wrong. It has nothing to do with the number of words.
yeah sorry we've been told that already -- we'll do a correction next week
@@AIForHumansShow Haha no worries, just figured I’d mention it. This is my first time seeing you guys, great show-I’ll be subscribing 👍
50:00 🌭 can't tell if it's Kevin. I'm not racist but all hot dogs look alike.
it is kevin
Gavin aren’t you a producer? how have you still not figured out how to introduce ‘ai see what you did there’ yet?? Even when kev throws you a bone with the slick segue. Get this handled post haste, thank you. 🎉
Ya I did like my own post
hahahaha we appreciate this
Since Kevin is in Canada, it would be so cool if he wants to do a fan meet up I know he has a lot of fans in Canada
we do need to figure out some sort of in person thing to do at some point
You think you have Siri problems! Not only does it activate every someone says sorry… at least 12 times a day here… but I live in “Surrey” which I have to type manually more often than you’d think. Let’s not even talk about the “about” “a boat” problem.
Hahahaha. I'll check myself from here on...
Olivia Munn is a famous Hollywood actress now and Kevin Pereira is a TH-cam content creator. Well you got my subscription because you were/are in my top 5 from TechTV/G4.
first type shi
YOU DID IT. We slept way past publishing but we're happy someone was here for this.
really failed hard on that intro didn't I?
You stuck the landing though.
Failed is a strong word.. I think it was the perrrrrrrfect human cherry on the top. Perfect imperfections
@@1pikapein we thank you for this