OpenAI’s New ChatGPT: 3 Secrets From The Paper!
❤️ Check out Lambda here and sign up for their GPU Cloud:
📝 The paper is available here:
📝 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:
Alex Balfanz, Alex Haro, B Shang, Benji Rabhan, Gaston Ingaramo, Gordon Child, John Le, Juan Benet, Kyle Davis, Loyal Alchemist, Lukas Biewald, Martin, Michael Albrecht, Michael Tedder, Owen Skarpness, Richard Sundvall, Taras Bobrovytsky,, Thomas Krcmar, Tybie Fitzhugh, Ueli Gallizzi.
If you wish to appear here or pick up other perks, click here:
My research:
X/Twitter:
Thumbnail design: Felícia Zsolnai-Fehér –
Deepseek killing it too
Yes. Running DeepSeek R1 32B model on my mac today in LM Studio and it’s amazing to have access to o1 AI reasoning power locally!
@@JohnSmith762A11B What specific mac do you have which lets you run it? And how many tokens per second do you get?
Deepseek is just a copy of chatgpt. It even says so itself if you ask it. Just another copycat from China. Nothing innovative about it. They just want to grab money from investors or gullible customers.
@@GateOfSteins wondering the same thing
@@JohnSmith762A11Bhardware specs? Thats crazy
Im Excited to possibly try this in a mobile port sometime?
What a time to be alive!
I can’t wait for models of this quality to be open source!
It’s just a system prompt, nothing more. I can get me llama to do the same on ollama. Software doesn’t “think” 😂
6 minute papers
Watch it in 2x, it’s still 3 minutes!
He speaks as fast as you hear in 2x speed. Has to slow down so we can understand him
How can I get Grok without being on X?
Yeah O-3 is going to dominate as it makes o-4, 0o-5, and then o=7 will just replace everything. No more anything for humans to do.
That’s not how it works bro but ok
Bold of you to assume sequential naming conventions.
@@Paull2here for this comment lol – hardly any tech companies have consistent naming
babe wake up new paper just dropped
Hey I have been a fan since your oldest videos. So glad to see the channel doing well. It’s amazing also how our interests converged – I was doing my PhD while you were working on your own research.
You are too kind, thank you so much!
Honestly i believe uncensored open source models will be the future, all this censoring is kinda worrying and im so tired of it 😫
Yes. But remember that if you use the frontier models through the API and give them a good and allowing system prompt.
They will censor WAY less.
The app and web versions are much much harder censored.
Kinda sad that it gets more resistent against jailbreaking.
I guess… this is now a race and Deepseek R1 will beat o3 two versions down the line…
The improvements are impressive, but at only 46%, it still isn’t “useful” yet. Not that that has stopped companies like Apple or Microsoft before, rushing out “new Features” that barely work
I once discussed inventing a fictional government office with ChatGPT. It started with something like “Federal Unilateral Multi-Kiosks for Young and Overdeveloped Unemployed”. Then I changed Multi-Kiosks to Central-Kiosks, and started using the abbreviation. I added more departments in the same way, and before long we were both swearing at each other.
is this voice AI?
I’d love some more light transport research coverage. 🙂 I recently tried to get o1 to help me write a procedural ocean shader… using HLSL using a compute and geometry shader and it failed terribly. lol.
Deepseek is the real openAI
wow 13%- so you have about a 1 in 8 chance of getting a correct answer if you ask it a halfway serious question.
we should have never openly allowed them to know about jailbreaking
Step on the gas!
Go fast!
XLR8!