SONY NIGSTATION 9
#1
★★★★★
- Joined
- Oct 14, 2020
- Posts
- 23,488
We're ahead of schedule :clown:
Over the past few weeks in the AI space there's been lots of interesting news, from the scam of Reflection 70b to rumors of OpenAI oiling the gears for two brand-spanking new, sophisticated models to drop in the next few weeks. We shall wait no more as the first model (codenamed Strawberry) was just dropped, dubbed 'OpenAI o1'. This model is specifically designed for reasoning-based tasks and insider reports already suggest it may be used to generate synthetic datasets to train their second upcoming model (GPT-5 speculated) - codenamed Orion.
OpenAI o1
https://openai.com/index/learning-to-reason-with-llms/
As we can see OpenAI-o1 specifically VASTLY outperforms current models in fields of coding, mathematics and reasoning, as they've baked in CoT (chain of thought) technology to allow it to emulate higher level thinking. For this new OpenAI-o1 model, the main feature is increased reasoning and videos on the official OpenAI YT channel hint at proficiency in many STEM fields. Obviously benchmarks have their issues when evaluating AI sometimes but demos of the new model show it to be highly competent.
So already we are ahead of schedule as Aschenbrenner claimed PhD level AI by 2027 :feelsLUL: ; keep in mind we are still in September of 2024. Absolutely over for the wagies, and I can see why Leopold was so fervent in his claims. If he was aware of the progress of 'Strawberry' and 'Orion' then this rate of development must have spooked him; perhaps the real beast is yet to come.
Last edited: Today at 7:45 PM
Over the past few weeks in the AI space there's been lots of interesting news, from the scam of Reflection 70b to rumors of OpenAI oiling the gears for two brand-spanking new, sophisticated models to drop in the next few weeks. We shall wait no more as the first model (codenamed Strawberry) was just dropped, dubbed 'OpenAI o1'. This model is specifically designed for reasoning-based tasks and insider reports already suggest it may be used to generate synthetic datasets to train their second upcoming model (GPT-5 speculated) - codenamed Orion.
OpenAI o1
https://openai.com/index/learning-to-reason-with-llms/
As we can see OpenAI-o1 specifically VASTLY outperforms current models in fields of coding, mathematics and reasoning, as they've baked in CoT (chain of thought) technology to allow it to emulate higher level thinking. For this new OpenAI-o1 model, the main feature is increased reasoning and videos on the official OpenAI YT channel hint at proficiency in many STEM fields. Obviously benchmarks have their issues when evaluating AI sometimes but demos of the new model show it to be highly competent.
- coders, mathematicians and researchers 2 weeks ago when asked if they think AI will get good enough to replace their jobs specifically..."Nooooo AI will never be able to do x field, it just won't, okay!!! We're hitting a plateau!!!"
So already we are ahead of schedule as Aschenbrenner claimed PhD level AI by 2027 :feelsLUL: ; keep in mind we are still in September of 2024. Absolutely over for the wagies, and I can see why Leopold was so fervent in his claims. If he was aware of the progress of 'Strawberry' and 'Orion' then this rate of development must have spooked him; perhaps the real beast is yet to come.
Last edited: Today at 7:45 PM