Sam Altman goes NUCLEAR (CODE RED)

In response to Google getting all the love for Gemini 3 and their TPU architecture, Sam Alman has declared code red. And according to the

code red. And according to the information, they have a new secret model in the works called Garlic. All

right, let me take a step back. I just

made a video about how Google is probably the best positioned company to win artificial intelligence. They have

everything. They have a frontier model.

They have AI infrastructure. They have

custom silicon. They have a ton of revenue. They have a ton of great

revenue. They have a ton of great researchers and of course a massive amount of data. And so Google after last week has really planted their flag on

the AI hill. And it seems everyone was mogging Open AI. According to semi analysis, OpenAI's leading researchers have not completed a successful fullscale pre-training run that was

broadly deployed for a new frontier model since GPT40 in May 2024. That is over a year and a half ago, highlighting the significant

technical hurdle that Google's TPU fleet has managed to overcome. So apparently

we have hit a wall in pre-training scaling. Ilia Sudskver, OpenAI

scaling. Ilia Sudskver, OpenAI co-founder, current founder of safe super intelligence, one of the leading minds in AI, just did a podcast with Darkesh Patel in which he basically

said, "The age of scaling is over. We

are transitioning back to the age of research." Meaning, we need new ideas.

research." Meaning, we need new ideas.

We need new algorithms. Just scaling up what we're doing currently is not working anymore. Watch this.

working anymore. Watch this.

>> From 2012 to 2020, it was the age of research.

Now from 2020 to 2025 it was the age of scaling but now the scale is so big like is is it is the belief really that oh it's so big but if you had 100x more

everything would be so different. I

don't think that's true. Andre Karpathy

another famed AI researcher AI thought leader also said something similar on the Darkh Patel podcast and of course Yan Lun has been saying this for a

while. Large language models in their

while. Large language models in their current form are running out of space to grow. We need new ideas. This is

grow. We need new ideas. This is

becoming so prominent it's even becoming memes. Look at this. So here's Ilia.

memes. Look at this. So here's Ilia.

Scaling is over and LLMs are a dead end.

Ah, you're so sweet. And then of course Lun comes in. Scaling is over and LLMs are a dead end. Hello human resources.

And Lacun loved it. He thought it was hilarious. And you know what it really

hilarious. And you know what it really was cuz Lacun has been saying this for years, years and years. Then Google came out with Gemini 3. And Gemini 3 showed

scaling is far from dead. But maybe

there's something unique about Google, something unique about the way they're doing things or their architecture that allowed them to scale past what everybody else said was the wall. And a

lot of people are pointing to their TPUs. The TPU is Google's own custom

TPUs. The TPU is Google's own custom silicon to run artificial intelligence.

They've been building TPUs for years and years. And now the benefit of all that

years. And now the benefit of all that work is finally coming to fruition. And

they put out an absolutely massive model that performs incredibly well. And

people started counting out OpenAI. But

don't count them out quite yet. Sam

Alman just this week told OpenAI, "We are going code red." This means working to improve the quality of Chad GPT and

the grand model race at the sacrifice of progressing other products. According to

this Wall Street Journal article, Altman said OpenAI had more work to do on the day-to-day experience of its chatbot, including improving personalization features for users, increasing its speed and reliability, and allowing it to

answer a wider range of questions. Now,

what is interesting about this quote is they don't really talk about the quality. And to be honest, and I've been

quality. And to be honest, and I've been saying this for a while, the quality, the core intelligence of these models, they're there. we really for 99% of use

they're there. we really for 99% of use cases don't really need higher intelligence models. Now, of course, I

intelligence models. Now, of course, I will be happy if these labs continue to improve the quality of the models and continue to improve the intelligence of the models. Yes, of course. But what we

the models. Yes, of course. But what we need is all the things around it, the scaffolding, and that's what he's talking about here, the day-to-day

experience. And for 99% of use cases and

experience. And for 99% of use cases and 99% of users, the day-to-day experience is everything. the models are good

is everything. the models are good enough to serve the vast majority of most people's use cases. And so we're at this interesting time. Google finally

caught up. They released an incredibly good frontier model, beating OpenAI's models on many benchmarks. Their usage

is growing. They went from 450 million active users to just a few months later 650 million active users. We have OpenAI whose growth seems to be slowing down.

They are nearing or almost at a billion active users. But as I said in a

active users. But as I said in a previous video, Google is incredibly well positioned. They have diversity of

well positioned. They have diversity of revenue. They have a ton of cash on

revenue. They have a ton of cash on hand. They have multiple business lines

hand. They have multiple business lines that they can tap into to fund and lose money on the AI side of their business while they compete with OpenAI and

Anthropic. But again, don't count OpenAI

Anthropic. But again, don't count OpenAI out. And by the way, if you want the

out. And by the way, if you want the easiest way to test all of these models without just having to pick one subscription to go with, try out the sponsor of today's video, Chat LLM by

Abacus. There are so many AI tools out

Abacus. There are so many AI tools out there, it becomes incredibly expensive to get access to them all. Trust me, I know. And there's a better way. With

know. And there's a better way. With

Chat LLM by Abacus AI, you get access to Chat GPT5, 03 Pro, Gemini 3 Pro, Claude 4.5, Grock 4, Deepseek, and many others.

And with Route LLM, it automatically picks the best model for whatever prompt you give it. And you get the latest image generation models like Seedream, Nano Banana, Aiogram, Flux, and more.

And you can even generate video, cling, runway, Helio with just a simple prompt.

And now you can access the latest Opus 4.5 and Nano Banana Pro easily in one place with Chat LLM. And their humanized

feature turns obviously AI written text into something that looks more human. So

it goes from boring AI generated to something unique and compelling. And

last they also have deep agent which is a powerful general agent that can build websites, apps, AI agents, chat bots, presentations and complete research

reports. And all of this is just $10 per

reports. And all of this is just $10 per month. Visit their website by clicking

month. Visit their website by clicking the link down below that lets them know I sent you. And thanks again to Abacus.

Now back to the video. Mark Chen, chief research officer at OpenAI, just did a podcast in which he said pre-training specifically supercharging our pre-training efforts for the last half

year. Um, and I think it's a result of

year. Um, and I think it's a result of some of those efforts uh together with Jakob focusing and building that muscle of pre-training at at OpenAI. uh you

know crafting a really superstar team around it, making sure that all of the important areas and aspects of pre-training are emphasized. Um that's

what creates the artifacts today that feels like we can go headto-head with Gemini 3 easily on on pre-training.

>> And so there it is. He explicitly said it. Yes, we have some tricks left. And

it. Yes, we have some tricks left. And

yes, we're forming and continuing to build out our superstar team to focus on pre-training to specifically compete with Google. He didn't mention

with Google. He didn't mention Anthropic. He didn't mention XAI. He

Anthropic. He didn't mention XAI. He

mentioned Google by name. Um, we think there's a lot of room in pre-training.

You know, a lot of people say scaling is dead. Um, we don't think so at all. I

dead. Um, we don't think so at all. I

think um it's a little bit of an alpha for us because we think there's so much room left in in in pre-training and and I think as a result of these efforts you know we've been training much stronger models and that also gives us a lot of

confidence carrying into you know Gemini 3 and other releases coming this end of the year >> and so there it is chief research officer at OpenAI saying pre-training is not dead RL was incredibly important we

spent time on it we maybe had a little bit of muscle atrophy on the pre-training side but now we are back we are back to pre-training and we are going hard and they have a model

cooking. Now that might be what the

cooking. Now that might be what the information just reported on. Let's take

a look. So according to the information this article just came out, OpenAI is developing garlic, a model to counter Google's recent gains again on the

pre-training front. Google really just

pre-training front. Google really just set the industry ablaze on how much they were able to scale up pre-training and have this incredible model. Now, garlic

is the counterargument to Google's Gemini. Last week, OpenAI's chief

Gemini. Last week, OpenAI's chief research officer, Mark Chen, told some colleagues about the new model, which was performing well on the company's evaluations, at least when compared to Gemini 3 and Anthropics Opus 4.5. Now,

remember, it might be performing well on the benchmarks, and they might not even be done cooking it. It just might be a checkpoint that they're testing against, and that just makes it super exciting.

Of course, the pre-training might plateau, but it is exciting nonetheless.

And as part of Sam Alman's code red inside OpenAI, OpenAI has a new reasoning model it is preparing to launch that is ahead, quote, ahead of Gemini 3 and OpenAI's internal

evaluations, Altman told colleagues Monday. And we might get a December

Monday. And we might get a December surprise that might mean GPT 5.2 or even 5.5 by the end of this year. But Garlic

is different from this other model codenamed Charlotte Pete which has had a lot of rumors circling around it lately.

But Garlic incorporates bug fixes that the company used in developing Charlotte Pete during the pre-training process.

And according to the Kobaysi letter, internal memos show that OpenAI is pausing projects like ads and shopping to focus on improving Chat GPT's performance. So kind of glad to hear

performance. So kind of glad to hear that. I don't need ads. I want them to

that. I don't need ads. I want them to focus on everything else that makes the experience of chat GPT better. And let

me talk about that for a moment. Now, a

handful of people that I've spoken to recently, including Jonah from my own team, has said that they have almost entirely switched to Google Gemini, but they're not the normal user, and I'm not

the normal user. When we're talking about the average general AI user, they don't care about these incremental improvements in the core intelligence of the model. What they want, as I said

the model. What they want, as I said earlier in this video, is the best overall experience. And right now, Chad

overall experience. And right now, Chad GPT is still at the frontier. Google's

Gemini is incredible. And of course, because they have distribution with Google search, they are going to continue to grow their user base and share of the market. But for most

people, Chad GPT was their first exposure to AI and likely will just continue to be. There would have to be major issues with chatbt for the

majority of people to want to switch off of it. Remember chatbt is the verb

of it. Remember chatbt is the verb similar to how Google was with search.

Go Google it. Well, with AI, the verb is still chatbt and the chatbt experience is still killer. And so even though we

have some people at the frontier of usage of these models of these products switching to whichever one is best at the moment the vast majority of the

population who uses AI won't be doing that. Don't count OpenAI out. They

that. Don't count OpenAI out. They

definitely have increased competition in this moment. But you know what? You know

this moment. But you know what? You know

who benefits from all of this cutthroat competition is us, the consumers of these products and services. If you

enjoyed this video, please consider giving a like and subscribe.