You showed that even the paid subscription couldn't do the example on their own ad.
2k per month for research that you can't trust is really bad.
Artificial Intelligence thread
- Thread starter 9dashline
- Start date
You showed that even the paid subscription couldn't do the example on their own ad.
2k per month for research that you can't trust is really bad.
Yup from the horses mouths
Attachments
-
qwq-full-version-open-source-o3-v0-suvl2l7mm58e1.jpeg
And have they released sth like Kling or they just basing their whole strategy on Douyin? For now, for me Bytedance is another Lenovo. Until they show sth good I won't keep my faith in one kneel Zhang's company.
A basic idea in philosophy: something can be necessary but not sufficient. As in: Passing ARC may be necessary but certainly not sufficient for AGI. (Chollet was super clear about this.)
A basic idea in cognitive science and engineering: a mechanism can be effective in some domain, and not others. As in O3 may work for problems where massive data augmentation is possible, but not in others where it’s not.
Most commentary has ignored one or both of these ideas.
Even setting aside the 1000x costs, not one person outside of OpenAI has evaluated o3’s robustness across different types of problems.
If it were truly robust across all or even most problems, they would have called it GPT-5.
The fact that they didn’t is .
An illuminating insight coming from Gary Marcus.
I am a firm believer in Artificial Intelligence and that it'll definitely provided added value as A TOOL TO AUGMENT Human beings to our respective professions and bring about huge productivity dividends as a result.
What I don't buy until convinced otherwise that AGI is closed to being achieve or is just around the corner. People must be underestimating human capacity and overestimating humanity's own creativity and for penchant to OVERHYPE A PRODUCT for the purpose of MONETIZATION.
OAI is now rumor to think about a $20000/mo sub for enterprise use of o3 pro etc
They singlehandedly made symbolic program synthesis mainstream!
o1, o3 are not new AI models. They are AI systems that use LLM's in a loop to extract answers. These experiments show it is possible to extract answers via LLM-guided search, at a great monetary cost.
Currently this search mechanism is wildly inefficient: we need a 100x reduction in search time before even thinking about replacing humans with these systems.
One reason why this could be is the emphasis on natural language as the system language. The system o3 uses looks a bit like this.
LLM generator --> natural language statements --> LLM "interpreter" --> correctness.
While sufficient to solve many reasoning tasks, it is unrobust and inefficient. For such methods to become economically viable, we would need a breakthrough in "natural language interpreters."
My personal opinion is that AI has caused a "crowding out" effect by starving investment into other areas of computer science. There are strengths and weaknesses to LLM-based techniques but betting all investment into LLM's is a foolish allocation of money. Even o3 shows that "pure" LLM prompting is not sufficient, requiring augmentation using search.
Last edited:
Put my thoughts on here
basically as it stands, o3 is completely unusable even if cost comes down to $10 per prompt.
and it’s only solving the hard problems because it got people that can solve them to first train the models. Well, can it get model to train against every menial task it will be expected to automate away?
Natural Language is wildly inefficient for computer systems reasoning
Have a look at this new Meta paper. I think that fundamentally this is the way forward, with the disadvantage that the AI systems will become a complete black box for humans