Thank you everyone for watching and thanks to Airfoil for our new look! Find their link in the show notes. Also, next Monday we've got a video walking everyone though how we make our AI co-hosts! You asked, we deliver (very late). xoxoxox Kevin & Gavin ps tell someone about AI For Humans today!
At the end of this show, you promised to finally create a video showing step by step how you create a virtual co-host and that you would release aid video on Monday. I have been waiting for this information for months...alas, Monday has come and gone and still no information on how you do this. When can we expect such a video?
@@WebStoresLtd lol we are SORRY we expect it is coming THIS coming monday now (a week from today) because this Runway video went up.... also Gavin is going to the UK for a week so we've now got a video ready for that (we'll still be doing the weekly pod / vid)
We're so thrilled with our redesign and just absolutely love that Phil helped us out. PEOPLE ARE GOOD, THE WORLD IS OK. Also, apologies for bringing Safety Steve into the world.
New subscriber…idk how you are just popping up on my algo, or how I’m just realizing it. I drink up AI You Tube like an energy drink. Good job guys, I’ll be returning.
haha we've been here for over a year but kind of just now finally figuring out the YT algo so that makes us happy it's helping us find the right people we too drink AI YT so thanks for getting drunk on LLMs with us (ps, in case you hadn't guessed, we're a lil weirder than your average AI YT channel)
hey thank you so much for both a) the sub and b) telling us this. Algos are so freaking weird. We made a few tweaks (different thumbnails, better titles & shorter episodes) and that prob had a lot to do with finally getting better algo love
Based on their pattern of training and release, the training they just announced is most likely about GPT-6. They already finished training GPT-5. So far they seem ship a model to the public about 9 months to a year after completing the training (unsupervised) and 6 months after the red teaming testing.
yep this makes sense. what's crazy is that the 'blue whale' model that the Microsoft CTO was talking about is prob 6 and I guess we'll see exactly what insane surprises that has in store for us
@@AIForHumansShow Another thing I found interesting is this statement from that announcement: "OpenAI has recently begun training its next frontier model and we anticipate the resulting systems to bring us to the *next level of capabilities on our path to AGI* ." I've watched multiple interviews from Sam regarding GPT-5, and all he's been saying is the major change about that model is how "much smarter". So its reasoning capabilities are going to be greatly improved, but I don't expect much more from that update. Of course, reasoning is a big deal, but the model coming in 2025 will be much more shocking, in my opinion. So combining that improved reasoning with the omni-model interface as well as opening Sora to the public will give them the edge again.
@@1sava yeah pretty curious as to how much this whole paradigm can actually scale. will be fascinating if we get at least two more major jumps in scaling up.
Great New Rebrand and congratulations on the growing success. The way I see it ….. I look at your subscribers amount and transfer it to a stadium seat chart ….. always good for perspective.
OK guys, there's only so much fun with Steve that can be legal. So my question is: Who do I sue for getting that fucking song stuck in my head now? Suno? Or Gavin?
I've played with the music-generation AIs as well and I agree about Suno being better at traditional mainstream genres whereas Udio does pretty well with more "out there" options. Udio can do some very decent "prog rock" and I even managed to get a very nice Balkan folk song out of it, which is pretty wild. Suno also seems a bit looser with following the prompted genres, in my experience. I find that Suno is better with a more generic prompt where you use plain language to describe a song whereas Udio isn't as good at that, but is much better with a detailed-oriented approach where you specify a series of features and custom lyrics that feed in to the AI.
Totally agree. I think it’s because I (this is Gavin) tend to move towards weirder music that I write lyrics to myself I lean towards Udio but I’m excited for Suno 4.0.
I will try to provide you some better timelines. When GPT-4 was released around April 2023, the training was already completed at least 1 year prior (2022). Sam mentioned GPT-4 was in red-teaming for at least 10 months. GPT-5 started was in training from around 2023 Nov and completed in April 2024. GPT-5 is now in red-teaming phase will release little by little with different names I assume for this year. They cannot keep tons of H200 GPU racks in idle mode (too expensive). The so-called "NEXT" model is GPT-6 and training has started this month. GPT-6 training could be done by the end of this year but will need another 6 /10 months for red-teaming so the release will be around late-2025 for the "NEXT" model which OpenAI is referring to. Just need to understand the life-cycle of high-end LLMs from training to Red-teaming to release can span out to 2 or maybe 3 years.
Yeah this makes a ton of sense. It's gonna be interesting to see how much a leap GPT-5 is over 4o and then project out to whatever NEXT could actually be.
oh man i'm really excited for that next episode where you guys are gonna go into detail how you do the co-host thing i was really hoping you guys would do that.
As someone very familiar with the DevOps process…..they are probably on 5.5 or 6. The QA process has to be vigorous due to all the controversy. AGI is still years away, but if they can solve the compute issue, then it could grow exponentially.
Keep it in mind that ai has no motivation it has no feelings and it is just a very advanced search engine. People think it is alive because of movies, i don't think agi/si is even possible.
Hey Kevin, can I ask where you got that sweater? I think it's the same one you wore on the Kevin Rose interview, but it was red. I'd like to get one myself.
So, I’ve been saying lately Sora + Apple Vision Pro = text to simulation I’m already building the basics for my simulation, a 6 burrough Filipino Steampunk Metropolis. So? I agree, that is coming for sure.
we're working on it -- spread the word, finally getting the algo to take a lil notice everyone says it takes time to learn YT and good lord, that was very true for us
well, we do make a podcast and it's ok if you don't love it. there are literally millions of other ones and other TH-cam channels to watch. but we appreciate you giving our lil channel a try.
Ya'll are some Sam Altman groupies🤣 I SWEAR every AI channel repeats what the last AI channel said. Yeah we know Sam did some stuff...can we move on? Like who else is innovating and building stuff? Tired of looking at Sams face everyday🙄
Thank you everyone for watching and thanks to Airfoil for our new look! Find their link in the show notes.
Also, next Monday we've got a video walking everyone though how we make our AI co-hosts! You asked, we deliver (very late).
xoxoxox Kevin & Gavin
ps tell someone about AI For Humans today!
can't wait!
At the end of this show, you promised to finally create a video showing step by step how you create a virtual co-host and that you would release aid video on Monday. I have been waiting for this information for months...alas, Monday has come and gone and still no information on how you do this. When can we expect such a video?
Still waiting... which Monday were you talking about?
@@WebStoresLtd lol we are SORRY
we expect it is coming THIS coming monday now (a week from today) because this Runway video went up....
also Gavin is going to the UK for a week so we've now got a video ready for that (we'll still be doing the weekly pod / vid)
We're so thrilled with our redesign and just absolutely love that Phil helped us out. PEOPLE ARE GOOD, THE WORLD IS OK.
Also, apologies for bringing Safety Steve into the world.
New subscriber…idk how you are just popping up on my algo, or how I’m just realizing it. I drink up AI You Tube like an energy drink. Good job guys, I’ll be returning.
haha we've been here for over a year but kind of just now finally figuring out the YT algo so that makes us happy it's helping us find the right people
we too drink AI YT so thanks for getting drunk on LLMs with us
(ps, in case you hadn't guessed, we're a lil weirder than your average AI YT channel)
@@AIForHumansShow your humor is more out there and entertaining, where a lot of the AI YT guys are cleverly dry. I enjoy both
tysm we appreciate it!
hey thank you so much for both a) the sub and b) telling us this. Algos are so freaking weird. We made a few tweaks (different thumbnails, better titles & shorter episodes) and that prob had a lot to do with finally getting better algo love
aaaye! welcome, and thanks.
It’s not Thursday without this podcast 🎉
it's the only day i now recognize.
ayeeeee thank you for this message, means a lot
Why is this so cool and not viral? You deserve at least 70k views
YOU CAN START THAT TRAIN
lol we've been here for a while, just grinding away, happy people are starting to see what we do
slow, but surely. with your assist. :)
it's a slow build but we're out here grinding the best we can
“AI Safety with Steve” song has some strong “Still Alive” energy from the Portal game end-credits
hahaha hadn’t thought about that but you’re absolutely right
well if you mean it's been stuck in my head since I brought the cursed thing into being then yes
Humans for AI podcast when?
uhhhhhhhhhhhhhhhhh soon?
My brain is hurt.
Based on their pattern of training and release, the training they just announced is most likely about GPT-6. They already finished training GPT-5. So far they seem ship a model to the public about 9 months to a year after completing the training (unsupervised) and 6 months after the red teaming testing.
yep this makes sense. what's crazy is that the 'blue whale' model that the Microsoft CTO was talking about is prob 6 and I guess we'll see exactly what insane surprises that has in store for us
@@AIForHumansShow Another thing I found interesting is this statement from that announcement:
"OpenAI has recently begun training its next frontier model and we anticipate the resulting systems to bring us to the *next level of capabilities on our path to AGI* ."
I've watched multiple interviews from Sam regarding GPT-5, and all he's been saying is the major change about that model is how "much smarter". So its reasoning capabilities are going to be greatly improved, but I don't expect much more from that update.
Of course, reasoning is a big deal, but the model coming in 2025 will be much more shocking, in my opinion.
So combining that improved reasoning with the omni-model interface as well as opening Sora to the public will give them the edge again.
@@1sava yeah pretty curious as to how much this whole paradigm can actually scale. will be fascinating if we get at least two more major jumps in scaling up.
Steve definitely knows the location to some buried bodies in the woods
So. Many. Bodies…
Another great one guys!
But maybe you should just stop questioning Steve's credentials, for your own Safety
we haven't seen or heard from Steve since he walked off, so yes you're prob right
8:00 I really liked the analysis by Wes Roth on Helen's motivation.
we love Wes' channel but haven't seen this yet, we'll take a look
love the new logo! this week's AI (Steve-o-rino) makes me happy about the future of AI security (oh, and I did take something before writing that )!
thanks so much for this! we're thrilled about it too.
Hahaha what…did…you…take???
@@AIForHumansShow just a little something I got at the ai made easier store
I love Steve, we need a AI battle between Gash and Steve. Awesome show guys.
thank you so much and thanks for always coming back for more of us two dummies!
Great New Rebrand and congratulations on the growing success. The way I see it ….. I look at your subscribers amount and transfer it to a stadium seat chart ….. always good for perspective.
we appreciate that so much - and thank you for sticking with us!
shoutout to AirFoil for doing something just awesome for us!
Helen spoke out right after her benefits were legally safe allowing her to chat about OpenAI.
🤣
OK guys, there's only so much fun with Steve that can be legal. So my question is: Who do I sue for getting that fucking song stuck in my head now? Suno? Or Gavin?
WE TAKE NO LEGAL RESPONSIBILITY FOR ANYTHING ON THIS SHOW
that said, it's prob Gavin's fault
I'm so very sorry
I've played with the music-generation AIs as well and I agree about Suno being better at traditional mainstream genres whereas Udio does pretty well with more "out there" options. Udio can do some very decent "prog rock" and I even managed to get a very nice Balkan folk song out of it, which is pretty wild. Suno also seems a bit looser with following the prompted genres, in my experience. I find that Suno is better with a more generic prompt where you use plain language to describe a song whereas Udio isn't as good at that, but is much better with a detailed-oriented approach where you specify a series of features and custom lyrics that feed in to the AI.
Totally agree. I think it’s because I (this is Gavin) tend to move towards weirder music that I write lyrics to myself I lean towards Udio but I’m excited for Suno 4.0.
Big fan, first time caller. You guys have THE best and most balanced takes on AI I've seen/heard to date. Thank you and keep up the great work =)
Hey thank you so much! We really appreciate this and work hard to make it that way.
I will try to provide you some better timelines. When GPT-4 was released around April 2023, the training was already completed at least 1 year prior (2022). Sam mentioned GPT-4 was in red-teaming for at least 10 months. GPT-5 started was in training from around 2023 Nov and completed in April 2024. GPT-5 is now in red-teaming phase will release little by little with different names I assume for this year. They cannot keep tons of H200 GPU racks in idle mode (too expensive). The so-called "NEXT" model is GPT-6 and training has started this month. GPT-6 training could be done by the end of this year but will need another 6 /10 months for red-teaming so the release will be around late-2025 for the "NEXT" model which OpenAI is referring to. Just need to understand the life-cycle of high-end LLMs from training to Red-teaming to release can span out to 2 or maybe 3 years.
Yeah this makes a ton of sense. It's gonna be interesting to see how much a leap GPT-5 is over 4o and then project out to whatever NEXT could actually be.
thanks so much for this -- agree with this pretty much
I really liked Phil’s take on the role of designers and interfaces!
yeah it's so nice having people with different creative jobs on the show, lotta different angles to approach the AI conversation from as well
I remember actually wanting an encyclopedia set when I was young, and then using Encarta for the first time and how cool it was
Encarta was absolutely the coolest. Like a real 'omg' moment of early computer stuff.
oh man i'm really excited for that next episode where you guys are gonna go into detail how you do the co-host thing i was really hoping you guys would do that.
WE ARE. JUST SLIGHTLY DELAYED. It will happen, we've got it ready to go need to do a few more specific things on it.
Came in sad and anxious about AI and left with a smile on my face...Thanks Gents!!! don't know how do you do it
well, this is just the best message and exactly what we hope to do
Where is the link to the Canva show, yo? I wanna start my weekend right!!!
hahaha oh i forgot will drop into the show notes now but here you go...
x.com/asallen/status/1794467197039079654
just happy this is the first thing you thought to ask, our fans understand us
As someone very familiar with the DevOps process…..they are probably on 5.5 or 6. The QA process has to be vigorous due to all the controversy. AGI is still years away, but if they can solve the compute issue, then it could grow exponentially.
So if you had to guess, how big is the leap gonna be that we get with 5? For us plebs.
Wicked! Great job to everyone out there! Have a great weekend!
same to you!! and tysm for watching
I’m having nightmares about safety Steve tonight
you and me both brother
that's precisely what he wants.
That train car video is interesting, AND TERRIFYING. The AI can single out the humans? Next step is the T2's...😂
yeah we are maybe three lil metal steps away from saying huh, maybe we SHOULD slow down
Keep it in mind that ai has no motivation it has no feelings and it is just a very advanced search engine. People think it is alive because of movies, i don't think agi/si is even possible.
I actually hit 'like' when you said it...imagine that. :) Love you guys.
This is your lucky day now. Buy a lottery ticket and share 50% with us for the blessings. JKJKJK or are we
next week "like, subscribe, donate a billion dollars..."
Safety Steve looks like a failed Fred Armison SNL skit :)
we know we've leveled up when we're getting compared to failed SNL sketches haha
tysm for watching as always
11:00 is that Chris Pirillo?
hahaha we love CP but no this was not inspired by him
Hey Kevin, can I ask where you got that sweater? I think it's the same one you wore on the Kevin Rose interview, but it was red. I'd like to get one myself.
Love the content. In the name of nitpicking, the colored contact thing is kinda distracting. Cheers!
tysm!! oh which specific thing are you talking about??
@@AIForHumansShow Am I crazy or does one of the hosts not wear colored contact lenses? It's 100% irrelevant but it's spooky and stuff
I'm here for the text-to-AAA-video-game AI.
looking forward to 'vaccum man' which is just me vaccuming up corners of my virtual home ala power washing, but yes cannot wait
@@AIForHumansShow 8-bit style? 😁
would play
Safety Steve reminds me of CodeMiko. I wonder if it is her...🤔
it's not (it's really and truly only AI) but we do love her
Chat gpt 4o is freaking awesome
cannot wait for the full unlock on the rest of the toolset within it
So, I’ve been saying lately
Sora + Apple Vision Pro = text to simulation
I’m already building the basics for my simulation, a 6 burrough Filipino Steampunk Metropolis.
So? I agree, that is coming for sure.
hahaha yes, this is def the future, especially in a few years when Sora generations could pop up as fast as say Suno songs now...
It sounds like it is GPT 6 they are working on to be honest.
def feels that way -- we assume GPT-5 is coming later this year
Oh one thing chat gpt needs to do, get one more voice to replace sky. Devastating loss with no sky voice
honestly they should do this asap -- they need more voices to choose from
Open ai will be just fine.
prob more than fine
Yes...YYEESSSSSS
YE-YE-YESSSSSSSSSSSSSSSS
Kevin Pereira???
YES THIS IS RIGHT
I can't wait to see ads for random shit when I use chat GPT for anything in the future.
oh they are coming faster than you prob know
What Roblox stranger things add on? So confused about that lol
it was a 'live event' they did with stranger things
Hahahaha. Sasasaaa Safety ssssSteve
There's only so many times on AI can say SA-SA-Safety
Ooops! Missed bleeping an f-bomb. Maybe you should use AI to help censor your language.
er, yeah we saw that. mistake on our part.
we have submitted a lil trim -- thank you for mentioning, might take a lil while to show up in the feed
🎉
🙏
There should be a few zeros on your subscriber count
we're working on it -- spread the word, finally getting the algo to take a lil notice
everyone says it takes time to learn YT and good lord, that was very true for us
First?
YOU DID IT huzzah!
Commit to the bit, yo!
🎉🎉🎉🎉❤❤❤👏👏👏
🙏
This show is a BIG NOTHING BURGER.
That sounds delicious!!
50 minutes of conjecture.
well, we do make a podcast and it's ok if you don't love it. there are literally millions of other ones and other TH-cam channels to watch. but we appreciate you giving our lil channel a try.
Ya'll are some Sam Altman groupies🤣
I SWEAR every AI channel repeats what the last AI channel said.
Yeah we know Sam did some stuff...can we move on?
Like who else is innovating and building stuff?
Tired of looking at Sams face everyday🙄
hahaha fair enough, but we do talk about a lot of other things both on this channel and deeper in this video etc
Another great episode fellas. Will never get tired of your unhinged ai guests 😂
hahaha thankfully neither do we!
chanell made by people with real people? Its so oldschool. 2 hours of watching you faces. WHY?
Please. Automate our pain…
I don't comment usually but I like the show and want to boost that algorithm.
you know what, this here? THIS IS AMAZING
thank you so much and we really appreciate it
You went Dos Equis on Ai *XX*