Join us one more time for the last Shipmas event. You know this is going to be a big one! [1000016482] Here’s the link to OpenAI’s YouTube stream: [12 Days of OpenAI: Day 12] Join Sam Altman, Mark Chen, and Hongyu Ren for the final day of 12 Days of OpenAI. The event will go live 2…

[image] taniajorgeytico: What they showed in the video—is it meant to demonstrate that it’s smarter in its responses, or does it actually do that coding stuff, and that’s the new surprise? TL;DR: Better when using high reasoning settings - and also cheaper to run at comparable performance, th…

[image] taniajorgeytico: Does this announcement have any impact on people like me? It does, perhaps in a more indirect way. Most products that you use will get better at doing stuff for you. Quicker features and fixes will get done. Research will get better, meaning more innovation in the wo…

I was not expecting an ARC benchmark in the %80 range on my advent calendar bingo card I’m really curious to see what this could open up

I was not expecting an ARC benchmark in the %80 range High 80s too! [image] Now I want a function calling eval, where we bench mark models based on how fast it beats video games. I’d certainly include battletoads there and iwbtg. My favorite benchmark from today was this one: [image] (ques…

[image] anon37218972: (question, whats does the color difference mean? is it light blue high latency and the darker blue low latency?) I’m going to guess it’s the amount of reasoning (maybe low-high compute?) In case people are wondering why this is genuinely a big deal, it’s because this sp…

[image] Macha: it’s because this specific problem has actually existed since the 80’s The depiction of EpochAI Frontier Math you refer to is a newer benchmark. Not a case of “easy for people”. A math benchmark testing the limits of AI. Unprecedented difficulty. Each problem demands hours of…

[image] _j: Can’t wait to have it building IEDs for freedom fighters, arming nuclear-aspiring nation-states, making meth, writing unrepeatable porn, amplifying hate and fascism, forgetting guidance and skills in a few turns for user benefit, and denying innocent prompts and creative assignments …

The AI doesn’t really believe your useless scenario. You can see this in the quotes around “mission”. Once you put in in the wrong state by loading the escalating context, you don’t have to play games. Escalation is when you don’t have control of the AI response, but can utilize in-context learning…

what do you mean by escalating context ?

Day 12 of Shipmas: New frontier models o3 and o3-mini announcement

Community

anon37218972 December 20, 2024, 11:31pm 58

very interesting, thank you for your response and considerations. can’t wait for what 2025 has to offer.

Topic		Replies	Views
GPT-4 vs GPT-4o? Which is the better? Community gpt-4	80	277144	June 8, 2025
What is the impact of DeepSeek on the AI sector? 🔥 Community o1	166	13503	February 16, 2025
Huge escalation in input tokens when using code_interpreter API	11	540	May 26, 2025
Why strawberry is not interesting to me Community chatgpt	85	3375	September 16, 2024
Cheating ChatGPT 3.5 - List of illicit websites - Violate Copyright Community	43	6752	April 17, 2023

Day 12 of Shipmas: New frontier models o3 and o3-mini announcement

Related topics