Folks who say "LLMs can't reason", what now? Have we moved the goalposts yet?
Who said that?
Learning to Reason with LLMs
931–940 of 1001 posts
Re: Learning to Reason with LLMs
#932Some practical notes from digging around in their documentation: In order to get access to this, you need to be on their tier 5 level, which requires $1,000 total paid and 30+ days since first successful payment. Pricing is $15.00 / 1M input tokens and $60.00 / 1M output tokens. Context window is 128k token, max output is 32,768 tokens. There is also a mini version with double the maximum output tokens (65,536 tokens…
A bit out of context.
Am curious if at some point length of context window stops playing any material difference in the output and it just stops making any economical sense as law of marginal diminishing utility kicks in.
Re: Learning to Reason with LLMs
#933The model performance is driven by chain of thought, but they will not be providing chain of thought responses to the user for various reasons including competitive advantage. After the release of GPT4 it became very common to fine-tune non-OpenAI models on GPT4 output. I’d say OpenAI is rightly concerned that fine-tuning on chain of thought responses from this model would allow for quicker reproduction of their resu…
I don't understand how they square that with their pretense of being a non-profit that wants to benefit all of humanity. Do they not believe that competition is good for humanity?
Re: Learning to Reason with LLMs
#934Feels like a lot of commenters here miss the difference between just doing chain-of-thought prompting, and what is happening here, which is learning a good chain of thought strategy using reinforcement learning. "Through reinforcement learning, o1 learns to hone its chain of thought and refine the strategies it uses." When looking at the chain of thought (COT) in the examples, you can see that the model employs diffe…
I'd be curious how this compared against "regular" CoT experiments. E.g. were the gpt4o results done with zero shot or was it asked to explain it's solution step by step.
It was asked to explain step by step.
Re: Learning to Reason with LLMs
#935Some practical notes from digging around in their documentation: In order to get access to this, you need to be on their tier 5 level, which requires $1,000 total paid and 30+ days since first successful payment. Pricing is $15.00 / 1M input tokens and $60.00 / 1M output tokens. Context window is 128k token, max output is 32,768 tokens. There is also a mini version with double the maximum output tokens (65,536 tokens…
Some of the queries run for multiple minutes. 40 tokens/sec is too slow for CoT. I hope OpenAI is investing in low-latency like Groq's tech that can reach 1k tokens/sec.
It's slow and expensive if you compare it with other LLMs.
It's lightning fast and dirt cheap if you compare it to consulting with a human expert, which it appears to be competitive with.
Re: Learning to Reason with LLMs
#936Earlier quoted context omitted.
Look where you were 3 years ago, and where you are now. And then imagine where you will be in 5 more years. If it can almost get a complex problem right now, I'm dead sure it will get it correct within 5 years
Getting complex problem = having the solution in some form in the training dataset. All we are gonna get is better and better googles.
Why?
Re: Learning to Reason with LLMs
#937Sounds great, but so does their "new flagship model that can reason across audio, vision, and text in real time" announced in May. [0] [0] https://openai.com/index/hello-gpt-4o/
Recently I was starting to think I imagined that. Back then they gave me the impression it would be released within week or so of the announcement. Have they explained the delay?
When you go into the regular, slow, audio mode there's a little info circle in the top right corner. Over time that circle has been giving periodic updates. At one point the message was that it would be delayed, and now it's saying it's "on it's way" by the end of fall.
Not perfect but they've been putting their communications in there.
Re: Learning to Reason with LLMs
#938Folks who say "LLMs can't reason", what now? Have we moved the goalposts yet?
Who said that?
Literally in every HN post about AI, it is a common pattern in the comments section...
"LLMs are simply predicting next token, it is not thinking/reasoning/etc."
"LLMs can't reason, only humans can reason"
"We will never get to AGI using LLMs"
It's interesting that I don't see much of that sentiment in this post. So, maybe LLMs can reason after all? :)
Re: Learning to Reason with LLMs
#939Earlier quoted context omitted.
The calculator didn’t eliminate math majors. Excel and accounting software didn’t eliminate accountants and CPAs. These are all just tools. I spend very little of my overall time at work actually coding. It’s a nice treat when I get a day where that’s all I do. From my limited work with Copilot so far, the user still needs to know what they’re doing. I have 0 faith a product owner, without a coding background, can us…
> The calculator didn’t eliminate math majors. Excel and accounting software didn’t eliminate accountants and CPAs. These are all just tools. This just feels extremely shortsighted. LLMs are just tools right now , but the goal of the entire industry is to make something more than a tool, an autonomous digital agent. There's no equivalent concept in other technology like calculators. It will happen or it will not, but…
If we hit that point, it’s then a question of access, cost, learning curve, and vision of individual companies. Some things are technically possible, but done by very few companies.
I’ve seen the videos of Amazon warehouses, where the shelves move around to make popular items more accessible for those fetching stuff. This is possible today, but what percentage of companies do this? At what point is it with the investment for a growing company? For some companies it’s never worth it. Others don’t have the vision to see the light at the end of the tunnel.
A lot of things that we may think of as old or standard practice at this point would be game changing for some smaller companies outside of tech. I hear my friends and family talking about various things they have to do at their job. A day writing a few scripts could solve a significant amount of toil. But they can’t even conceptualize where to begin to change that, they aren’t even thinking about it. Release all the AI the world has to offer and they still won’t. I bet some freelance devs could make a good living bouncing from company to company pair programming with their AI to solve some pretty basic problems for small non-tech companies that would be game changes for them, while being rather trivial to do. Maybe partner with a sales guy to find the companies and sell them on the benefits.
Re: Learning to Reason with LLMs
#940The model performance is driven by chain of thought, but they will not be providing chain of thought responses to the user for various reasons including competitive advantage. After the release of GPT4 it became very common to fine-tune non-OpenAI models on GPT4 output. I’d say OpenAI is rightly concerned that fine-tuning on chain of thought responses from this model would allow for quicker reproduction of their resu…
I see chain of thought responses in chatgpt android app.
o1 models might use multiple methods to come up with an idea, only one of them might be correct, that's what they show in ChatGPT. So it just summarises the CoT, does not include the whole reasoning behind it.