o3-mini and the “AI War”

o3-mini is here, and yes, I’ve read the paper in full – 2 hours after release, and even the post-launch Reddit AMA. Some epic details like a FrontierMath score that made me double-take, a likely new Cursor favorite, bio risk expertise and a cost-comparison with Deepseek R1., But does it perform on basic reasoning – let’s find out. Plus, arguably the bigger story – the increasingly frenetic rhetoric coming out of the West – and Dario Amodei and Alexandr Wang (CEOs of Anthropic and Scale AI respectively) in particular. The last thing we need is an “AI War”.

(Colab):

AI Insiders ($9!):

Chapters:
00:00 – Introduction
00:45 – o3 mini
05:11 – First impressions vs Deepseek R1
07:22 – 10x Scale, o3-mini System Card, Amodei Essay, bitcoin wallets…
12:40 – Simple Competition Finale
13:03 – Clips and Final Thoughts on the “AI War”

O3-mini:
Paper:
Amodei Essay:
FrontierMath wild stat:
Sam Altman Channels Napoleon:
Altman ‘pulls up releases’:
“AI War” by Wang:
Anthropic Original Views on Capabilities:
AI Insider Cost Comparison:
Deepseek R1 Paper:
R1, o3-mini Price Comparison:
Semianalysis on $1.3M deepseek salaries, and them falling behind as ‘the time gap to match US capabilities increases’:
OpenAI Valuation:
Wang Clip:
Amodei Clip:

AI Insiders ($9!):

Non-hype Newsletter:

Podcast:

Joe Lilli
 

  • @Loris-- says:

    He’s covering IA news so fast, we might get time travel before AGI.

  • @rein436 says:

    I like this Deepseek effect on frontier models.

    • @matthewdignam7381 says:

      It’s almost like everyone started trying again

    • @spetz911 says:

      R1 can’t even handle a simple logic puzzle, not sure what all this hype is about

    • @chinesesparrows says:

      Yeah I won’t be using deepseek when NewGuard audit found it gave 17% accuracy in delivering news. It was almost last place anomg the models tested (10th place of 11 tests) and it gave alot of refusals and vague answers.

    • @tringuyen7519 says:

      @@matthewdignam7381DeepSeek’s $6M training cost is bogus! Techtechpotato has video outlining all of DeepSeek’s claims that are bogus. Media is shocked. AI engineers aren’t!

    • @HaraldEngels says:

      Same here 🙂

  • @user-cg7gd5pw5b says:

    Two years ago, Dario Amadei thought we were decades away from AGI. Now he thinks we’re a year and a half away… from ASI. Boy how quickly things change!

    • @Raulikien says:

      And now he talks about tribalism like the rest. So much for cooperation and safety…

    • @maciejbala477 says:

      @@Raulikien our hand is forced because of the competition, basically. If you don’t do it first, someone else will. And that someone could be worse.

    • @Raulikien says:

      @@maciejbala477 Why worse? I don’t like politics, but you just need to look at the facts about who created the most wars in the last decade for profit and who took the science/economic route instead…

    • @tringuyen7519 says:

      @@Raulikien7 of 10 global corporations which exceed $1T market capital are US companies. What’s your point?

    • @michaeltownsend9008 says:

      When did he think we were decades away? (I thought he had pretty consistently short timelines.)

  • @soumyajitsarkar4614 says:

    It’s 4:42 AM here—when I found out the O3 Mini was released, I came here instantly. Thanks for always being so quick and thorough!

  • @legendarystuff6971 says:

    Thank god! The first decent video. I don’t understand why the ai field attracts all The SHOCKING bumbling buffoons, they show up in my feed faster than I can block them. Thank you for existing and for the work you do, we need a voice of reason. So far I’ve only found 2 youtube AI channels, yours and bycloud.

    • @1meen1 says:

      I like AI For Humans, but that’s more so for people who need casual explanation

    • @maciejbala477 says:

      It’s because it generates a lot of hype, so it’s easy to build on top of it and put out clickbait videos. But I completely agree, I’m glad I found this channel early, and have basically removed all other AI channels from my recommendations at this stage, because the vast majority of them are cancer

    • @JohnVance says:

      This channel has been extremely solid since the beginning, it’s the only one I know of that’s not unhinged.

    • @thatotherjohnc says:

      If you understood how the youtube algorithm works and the unfortunate necessity for ‘shocking’ titles your concerns would melt away.

    • @jaiveersingh5538 says:

      I’m really surprised to see you mention this channel and bycloud in the same sentence – they are definitely not in the same league for me. Bycloud is like the Fireship of AI, just good enough for awareness but not at all a reliable source for detail and nuance like what this channel provides

  • @JeffMcJunkin says:

    I dislike that it’s unclear whether o3 mini as it’s currently provided is low, medium, or high effort.

  • @freeottis says:

    „Dear president trump, America must win the ai war”

    Translation: pls bro give money for research pls it’s for our safety bro

    • @thornelderfin says:

      It means both and both are true.

    • @edism says:

      What are the threats and why didn’t they think of that first?

    • @alansmithee419 says:

      something something national security

    • @thornelderfin says:

      @@edism As you’ve noticed, the world is going to another world war very soon (Russia, China, North Korea and Iran vs the West). The moment China invades Taiwan everything will unfold. What exactly will happen is difficult to predict, but it won’t be good. AI will be very important military tool, propaganda tool, cybernetic warfare tool.

    • @thornelderfin says:

      @@edism However, we have so many open source models that are very powerful, that I think it won’t matter in the end – all sides will have enough to cause incalculable damage.

  • @Roman.Villain says:

    Pokémon cards are serious business and o3 mini understands.

    • @OverbiteGames says:

      Now if the guy had like.. burned his collection..

      That “need for CPR” might be due to a lack of internal organs.. and a head

    • @pauljones9150 says:

      With o3 mini wanting to be a crypto hustler, it all makes sense

    • @silpheedTandy says:

      Sometimes unresolved grudges stay repressed, until a moment of realization happens about how good a friend they were, really

  • Anonymous says:

    It’s not real until there’s an ai explained video about it!

  • @Dr.UldenWascht says:

    Oh, the contrast between 2017 and 2025 Amodei …
    Thanks for the fantastic cover.

    • @DavidGravesExists says:

      Not just him, but every US-based big AI guy over the past 6-12 months. What did they see/hear that made them flip?

    • @Raulikien says:

      @@DavidGravesExists The government breathing at their necks, forcing scientist to wage war

    • @Dr.UldenWascht says:

      @@DavidGravesExists I mean, they’re all officially arms of the gov. Just compare how many seats Pentagon has now on OpenAI’s board compared to 12 months ago.

      Also, this is a techno-cold-war, and the warlords believe AI is a defining military weapon. This is indeed Technofeudalism in full effect.

    • @danskyder1564 says:

      Anyone got the link to that full 2017 talk?

    • @Dr.UldenWascht says:

      @@danskyder1564 Here’s the title and it’s on YouTube: “Dr. Dario Amodei on AI Safety and Cybersecurity – Artificial Intelligence and Global Security Summit”

  • @unvergebeneid says:

    I already didn’t like the race between companies. Making it a race between adversarial countries will increase risks exponentially.

    • @maciejbala477 says:

      race is still better than a monopoly. It’s undoubtedly dangerous though. Even just the funny memes about DeepSeek refusing to answer questions about Tiananmen makes you think how much worse it could be in the future

    • @Alx-h9g says:

      ​@@maciejbala477 But do you think chatGPT doesn’t give a pro western spin on the information it gives out?

    • @thornelderfin says:

      @@maciejbala477 Most analytics agree that there is high probability that China will invade Taiwan in the next 3 years (they constantly keep talking about it). When that stats, we will all feel true power of Chinese hybrid warfare, cybernetic attacks, all Chinese apps, all our data, everything exploited. I have no idea what will exactly happen, but it won’t be good.

    • @thornelderfin says:

      I don’t see this as an AI race between countries, because worldwide open source is so close behind the proprietary frontier models that thew hole world will have more-or-less the same level of AI. The only question is how will each side use it (AI masterminded propaganda manipulation, cybernetic attacks).

    • @ShawnFumo says:

      @@maciejbala477 Sure, but they didn’t even censor the model. There’s just a censoring layer on top in their website, which is required by law there. If you host the model yourself or use one of the other hosting sites, it won’t be censored. That’s better overall to me than something like OpenAI, where you don’t have any control over the model at all. I think they have more censoring on ChatGPT than the API, but if they do have anything in the API, you don’t have the option to just host it yourself instead.

  • @sonny_wiess says:

    I’m so sorry I didn’t sus the answer was 3.8 trillion, I’ll never be able to show my face in public again.

  • @noobicorn_gamer says:

    “The dude’s in his mid-20’s. He needs to chill.”

    Coming from the most chillest and well spoken AI YouTuber around, that’s a very heavy shot fired :O Kinda spat my cherry coke out of my mouth when I heard that lol kudos!

  • @EmergentGarden says:

    I would like this video but I remembered you verbally attacked me for my over-expensive Pokemon collection

    • @Hikaru4444-o3u says:

      When you gonna release a new video ? i love your experiments

    • @Cammymoop says:

      oh hey

      also this joke reminded me of a yt comment I just saw yesterday, where somebody said they have watched every video that channel made for nearly a decade and they have not subscribed specifically because of how slighted they felt by the channel host reminding people to subscribe and promoting their own donation system in every video. I think common sense is maybe a bit less straightforward than we like to think lol

  • @thatryanp says:

    Better content than we deserve.
    I tried o3-mini head to head with Sonnet in a med-complexity React refactor in Cursor agent.
    o3mini failed with gusto, deleting components it didn’t know what to do with. Sonnet completed the task successfully in 25 steps.
    To be fair, Cursor team has not had time to tune the integration, and I don’t know what level o3 was running at.
    Sonnets combo of intuition and precision is still throned.

    • @tomenglish9340 says:

      There’s a big difference between competitive coding and software engineering. OpenAI’s internal evaluation on the latter essentially says as much.

  • @Inglip says:

    The solution of the Fermi Paradox seems clearer with every day that passes, as we stare at a potential Great Filter and are afraid that our rivals will beat us in opening Pandora’s box.

    • @danskyder1564 says:

      I had to do research to figure out what on earth you were specifically saying but well said 😂

    • @CATDHD says:

      the bitter fruits of uber capitalism that has been injected with american exceptionalism while Moloch’s god whispers sweet little nothings into narcisisstic infantilized broligarchs

    • @notoverpowered1739 says:

      Sure about AI being the answer to the fermi paradox? Contrary to other x-risks, an AI takeover would not only destroy humanity, it would leave an intelligent agent behind. I’d assume that at least in a significant fraction of cases, the AI would continue to expand, build things, explore the galaxy and leave traces that other species would see.
      So I believe that we are actually past the great filter and are in a pretty unique and lucky position – it is not at all clear that AI will doom us all, but that would make it so much worse if it happens…

    • @sawtonknuth9333 says:

      I was thinking the same 😂
      The Great Filter is powerful and could be AI based.
      But is it ?
      Aren’t we victims of our epoch’s fears like our grandparents during the Cold War ?
      And humanity self destruction won’t necessarily destroy the AI itself which could then spread across the galaxy.
      However no evidence in our Galaxy so far of AI drones scouting around…
      Moreover, If a powerful AI had already been developed elsewhere in the Galaxy, shouldn’t it perceive our nascent AI as a threat (in terms of energy resources consumption) ? Thus it should destroy us before…
      Conclusion :
      WAGMI 😂

    • @samuctrebla3221 says:

      You make a lot of assumptions on the future course of event, the main one being whatever supersedes the current human race won’t choose/be able to expand beyond our solar system.

      I think there’s a good chance that the first ASI end up being a giant autistic agent (to our human eyes), but it might seek expansion anyways (like the paperclip farmer).

  • @kinngrimm says:

    One of the scenarios as was described by Eliezer Yudkowsky, was market pressures and competition ending in a fast take off before we had enough time to prepare for eventualities or just understand what is going on at all. Seems we are taking that road now.

    • @peplegal32 says:

      I am especially disappointed in Dario, thought he would be stepping on the brakes now, seems it’s pedal to the metal.

  • @Badadodadoop says:

    Just wanted to say that you’ve become my go to for learning what all these new models can really do. No dumb clickbait, just good, smart content. Keep up the good job man.

  • @vladgheorghe4413 says:

    Officially entered my “enjoy life, spend money and go through my bucket list” phase of AI timelines

  • @frogdan says:

    I kinda feel like the value of a human life is understated in training of these models. Like if you just read the Internet with all the stories of war, video games, jokes about pretty much anything, a life seems rather expendable. Which is a stark contrast to real life.

    • @Panicbot92 says:

      @@frogdan I think about this all the time. Even though the Internet is the largest repository of information on human life in history, it still fails to capture what life is actually like. Especially the preciousness of it.

  • >