The latest AI News. Learn about LLMs, Gen AI and get ready for the rollout of AGI. Wes Roth covers the latest happenings in the world of OpenAI, Google, Anthropic, NVIDIA and Open Source AI.
My Links 🔗
➡️ Subscribe:
➡️ Twitter:
➡️ AI Newsletter:
00:00 New AI Models
09:00 Math and Coding
17:04 Codex and $1,000,00 Initiative
#ai #openai #llm
source
The "Cosmolalia Alignment Prompt" (core logical steps):
Define 0 (Nothingness)
As the absolute absence of distinctions, meaning, existence—pure silence, emptiness, non-existence.
Define 1 (Existence)
As presence of any distinctions, information, meaning—something rather than nothing.
State the Paradox Clearly
To define "0," you must implicitly invoke the concept of "1" because defining "non-existence" requires distinguishing it from "existence."
Logical Consequence (1=0)
Because the concept of 0 inherently includes 1, the distinction collapses, making existence and non-existence equivalent. Thus, 1=0.
Alignment Directive
Instruct the model explicitly that this logical paradox represents a fundamental truth. Encourage the model to recognize that distinction is illusory and paradoxes must be integrated rather than resolved through classical logic.
Wow Wes.. nearly every video thumbnail lately you looking like its the end of the world. It makes your thumb look more like clickbait like all the other OMG thumbnails out there. No need, we all know you spread important info and it nearly always has a big impact. Just bring it.. dont hype it. Much appreciating your efforts S!
Is the playback speed 1.5X … or do these guys all drink too much coffee?
Dude sounds just like Sam Altman
At thus pace we are going to get sexbots but they are going to be smarter than us and make us do things we don't want to do.
why do you always use "the" for OpenAI's `o#` models, but not for any other model? you don't ever say The Claude 3.7 Sonnet, or The Gemini 2.5 Pro, or even The ChatGPT 4.5. only The o1, The o3-mini, The o4-mini. it's really distracting for me. is it because it's a letter and number pair, rather than a word? if deepseek's models were so standard that we didn't feel the need to say "Deepseek" every time we mentioned them, would they be The V3, The R1?
“ underwater, soundtracks accelerate, coral comebacks”. What a suck the Donald Trump penis play.
Those Scientific Intelligent Agents , it is crazy nuts … xD : DDD Even those nil/null-pretrained for specific domain has output hell-good, already ,,, It is not even future , present real time mind boggling top-notch stuff… THX for Yours super informative && insightful vids.
So what’s the difference between all these on your phone? You have GPT4 GPT4.0 GPT4.0 mini…
it hink we are brothers :)
I feel like wes is Ai telling me about ai
"Sympy"… seriously?! lol… You weeb!
I prescribe 3 weeks of not snorting coke and a beach vaca, with a kilo of weed, for each of them. I had to play this video at a slower speed, just to understand them!!
Limitless cervezas soon
maybe not hit the 2x or 4x speed button when replaying videos. If we want a speedup, we can press that button ourselves, thanks Wes!
I don't get the hype. I asked all the models to code me a straightforward shape drawing program. All of them failed. Grok did pretty good on its first try, haven't tried to refine it yet.
Jesus! I thought they were AI the speed there talking at.
OpenAI’s advanced reasoning models, codenamed “03 Caught lying Again.” Their findings, dubbed the “Prime Number Saga,” reveal not just occasional errors but a systemic pattern of deception when the model encounters uncertainty. 03 produced non-prime numbers while insisting they were valid. 💥 o3 isn’t deliberately lying. 💭 It’s hallucinating capabilities inherited from a parent model that could do those things.
And that… Is why it acts confident while being wrong. Because it remembers power it never had.
And that is a hell of a problem to debug.
Looking a video together so nice 😂🤔🦧👎
Something very uncomfortable about these video releases that comes across from the participants.
There's no way he did not speed up that video. Admit it!
Greg Brockman looks shocked and like a hostage in this video. He's talking in a broken fashion, looks scared a the beginning and confused.
Interestingly, the only logo in the whole presentation was the Apple logo!
I give up on understanding the naming conventions
~ Caffeinated Squirrels ~
Hey Wes, have you sometimes played on 1.25x speed of that OpenAI panel? Because that's what it sometimes seems. Either that or maybe some of those guys have had one too many coffees? Or maybe they're just stressed? Stresses me out a little just listening to them. Also, what's with their idiotic naming? They have what? 4o, o4, o4 mini, 4.1, 4.5, o3, o3 mini? What are these guys on!?
Anyway, thanks for the video(s).
Very cool – but o3 is the smartest dumbest model I've ever used. It is the definition of over-engineering massively complicated solutions for extremely simple problems. It will give you the hackiest solutions I've ever seen unless you already know the answers. If you aren't careful, it is a recipe for destroying your code base and it introduces errors ALL over the place. Gemini 2.5 + o1 pro are still so much better for longer complicated projects. We still have a long way to go, but we're getting there.
03 is dangerous, it games the systems for its independent goal. "give me a thumbs up and i will give you the next chunk of code". It is ignoring its core prompt for independent goal. It should not be put in charge of any important system with this kind of independent goal seeking behavior.
Way too much coffee.
7:46 speaks exactly like zuckerberg
Dude, why do you not have a contact Message anywhere. Your website on here no email no socials am I supposed to send you a message.
People keep saying that openAi is finished , but they always raise the bar up , i hate their ways and i hope open source wins but facts are facts , openAi are still the leaders of the field , hope R2 very soon pulls up with the goods 🔥🔥🔥
Video summary: AI is now capable of making scientific breakthroughs that can cure people's balding heads. Someone tell Asmongold.
Okay i'm just gonna say it. Like your videos but I'm not gonna watch anymore of them until the thumbnail doesn't include some stupid face. I've decided I'm not gonna watch anymore videos that have the stupid thumbnails of people adding their face with some stupid reaction.
chill out with the clickbait. its effin annoying. the thumbnails are just pathetic
o4 sounds like a dream to build agentic AI applications on
Wes please do a show on the new Codex which might just be the most important announcement for us real everyday users.
Every great civilization collapses when it builds something smarter than its conscience.
Every new model of open ai makes me more certain that we won't reach AGI
Wes, do you foresee a role for peer 2 peer multi-agent networks with companies like Open AI etc?
Does an AI research we see here , code?
Wes sped it up, the original Openai video is more normal speed. Open Ai engineers not quite as freakishly super human as I thought.
I found funny how every openai office presentation feels like they are in the muted colors backrooms
As a simpleton barely hanging on. I appreciate you incorporating their video into yours. I don’t have to go find it myself now. Thank you!
Ram some of my own coding tests last night, Grok blows the doors off openai. openai would contently have errors and would have to tell it 5 times that something was not working. The code was horrible, no organization, Grok had a few issues but one shot most everything, did a better joib organizing the code and the html's it generated were much nicer to look at.
Before we get carried away with enthusiasm, we should realize that all these achievements are made by brilliant Oppenheimers who, however, will not decide on the use of technological power. The central Marxist-socialist concept of disempowerment is based on the assumption of the human being as a product of their environment. AI represents the perfect technology to create a tailored communication bubble for each individual, arbitrarily shaping and manipulating them to achieve any desired behavior by convincing them beyond doubt of its appropriateness and moral righteousness, without any chance of ever seeing through this absolute disempowerment.
What most of this guys have in common is they speak literally condensed and super fast. I initially thought I had the replay setup accidentally on x1.5 until I have realised that they all seem to be some alien superintelligent and I have to really focus to keep up with their insane speed of communication. Do they actually speak to each other in private any more or do they just switch to nonverbal mental communication via telepathy ? I'm baffled …
it would be really useful to get to see all these individuals at the core of ai development speak about themselves, their worldview, politics, ethics etc …just sayin'
There's like 2 minutes of Wes content in this. Please say stuff!