OpenAI’s GPT 4.1 – Absolutely Amazing!

❤️ Check out DeepInfra and run DeepSeek or many other AI projects:

GPT 4.1 (once again, likely API only, not in the ChatGPT app):

📝 The paper "Humanity's Last Exam" is available here:

Sources:

📝 My paper on simulations that look almost like reality is available for free here:

Or this is the orig. Nature Physics link with clickable citations:

🙏 We would like to thank our generous Patreon supporters who make Two Minute Papers possible:
Benji Rabhan, B Shang, Christian Ahlin, Gordon Child, John Le, Juan Benet, Kyle Davis, Loyal Alchemist, Lukas Biewald, Michael Tedder, Owen Skarpness, Richard Sundvall, Steef, Taras Bobrovytsky, Thomas Krcmar, Tybie Fitzhugh, Ueli GallizziIf you wish to appear here or pick up other perks, click here:

My research:
X/Twitter:
Thumbnail design: Felícia Zsolnai-Fehér –

Joe Lilli
 

  • @WompWomp656 says:

    What a time to be alive!

  • @wisdomking8305 says:

    Is OpenAI intentionally horrible at naming their product.

    There is no way they are this bad, this has to be a joke or marketing scheme

    • @tnix80 says:

      @@wisdomking8305 I thought it was like a nod to how over hyped they got with 4.5. they seem hesitant to call anything 5 unless it’s legit now. Hopefully they learned something, we’ll see.

    • @kittywampusdrums says:

      We are heading towards the Minus World

    • @EGarrett01 says:

      They’re not a polished company like Apple. They’re a research organization that happened to make a world-changing technological breakthrough. The branding, product names and all that are secondary.

    • @takkik282 says:

      Got the feeling they are releasing older models they didn’t mean to release to public, but because they are low on GPU they can’t push their master models.

    • @slacker2016 says:

      @@EGarrett01 debatable

  • @kipchickensout says:

    they always say the context menu gets larger, but it forgets more and more the more you iterate over some code, it’s really obvious, they still remember if you tell it 2 or 3 times to correct the code but at some point it just makes other mistakes instead or just runs way too slow because the conversation is already 10-15 minutes long

    i was using mostly 4 and then 4o though, so I’m not sure if that got improved too

    • @BRUXXUS says:

      Yeah. The chat and responses really start to degrade and fall apart over longer projects and chats. Starts being anti-productive at a point

  • @disconnect8873 says:

    They should ask GPT 4.1 to solve the naming problem.

  • @itsthelittlethings100 says:

    I also feel grateful to have these tools at this time. Thank you for this catch-up. =)

  • @NeoFighterX says:

    I remember breakthrough uploads on this channel being like once every few months…. what a time to be alive

  • @spartan117ak says:

    I miss when this was mostly videos about computer graphics innovations, now its all impressive footage of inevitable disappointments that will ruin everything they’re forced into by sweaty investors

  • @midnightmoves7976 says:

    why do you not cover open source?

  • @WinonaNagy says:

    DeepInfra’s AI projects are looking solid. GPT 4.1’s context window of up to 1 million tokens and the new mini and nano models sound like a perfect mix of speed and intelligence. Can’t wait to see them in action!

  • @coldlyanalytical1351 says:

    Very true.
    A few days ago I realised thta I no longer have to religiously monitor every AI development.
    I now use several very powerful AI tools for my research work – they are almost a commodity.
    I can now switch to focusing on my NON AI goals … with my AI buddies at my elbow, running on a dedicated monitor.
    Sure, the AIs will improve .. but will I notice?
    They are already cleverer, quicker, more knowledgeable than I am!

  • @cleroth says:

    What an alive to be timed!

  • @joel16961 says:

    Love your channel – but got to say, a 12 minute part one of 2 video is really stretching the “2 Minute Papers” moniker 🤣🤣

  • @twokidsmovies says:

    Im very confused, is GPT 4.1 better than 4.5 and o3? The naming is throwing me off

    • @korneldekany6689 says:

      In short: o3 > 4.1 > 4.5
      Explained:
      GPT 4.1 is much faster (and cheaper) than 4.5, and also does similarly. o3 on the other hand is a reasoning model, wich means that it ‘thinks’ before it answers a question. That makes it a LOT more time consuming to run (not necessarily more expensive), but also much more powerful. The reason for non-reasoning model (4.1, 4o, 4.5) is that the resoning models (o3, o1) are made using these non-reasoning models, and that their speed makes them a lot more usable for most cases.

    • @julien5053 says:

      4.5 is for writting, exploring new ideas, finding a path to the solution to a problem that other models can’t solve on their own.
      o3 & o4 will be particularly good for complex math, complex logic and complex programming.
      4.1 will be good for your regular math & programming.

    • @__________Troll__________ says:

      @korneldekany6689  *Hopefully, we’ll see o3-mini with access to the up-to-date information. For example, it believes Luka Doncic still plays for Dallas. This limits its scope greatly. It was fun trying to convince it that the sources it was pulling from were much older than today’s date, but it didn’t see reasoning. It’s really good with puzzles though*

    • @photelegy says:

      1:15
      The solution from the AI-image is to ad a thero the make 4.5 > 4.10 😂👍🏻

    • @fergalhennessy775 says:

      o3 doesn’t seem to be good at much other than codeforces problems to me.

  • @freeottis says:

    12 minute papers

  • @HelamanGile says:

    Can’t wait for Chat GPT 3.1

  • @TriSamples says:

    I’m so over AI. Can we have some non AI papers please please please?

    • @Blackhearts60 says:

      @@TriSamples My guess is because its a science channel and generally that means pushing the bounds of what is possible. Open source, while good, is usually a bit behind commercial AI. So its not as notable for a video.

  • @madcatlady says:

    I am still waiting for the translation of the Voynick Manuscript

  • @dirtydevotee says:

    Please stop supporting OpenAI. These are not innocuous computer people. They are actively working against you for their own profit.

  • @EVILBUNNY28 says:

    I was seriously frustrated with 4.5. It felt like the dumbest model since 3. Honestly didn’t listen to instructions and just spat out what it wanted. To top it all off you had limited responses (despite paying $20 a month) that wouldn’t reset for days after you used your limit.

    I seriously hope 4.1 is better

  • @pandoraeeris7860 says:

    And we’re getting two more today!

    What a time to be AI!

  • >