71–80 of 280 posts
The claim is that this removes the human bottleneck (aka SFT or supervised fine tuning) on domains with a verifiable reward. Critically, this verifiable reward is extremely hard to pin down in nearly all domains besides mathematics and computer science.
Reasoning transfers across domains.
Earlier quoted context omitted.
This assumes that you give honest feedback. Efforts to feed deployed AI models various epistemic poisons abound in the wild.
I am not in this space, question: are there "bad actors" that are known to feed AI models with poisonous information?
Earlier quoted context omitted.
This assumes that you give honest feedback. Efforts to feed deployed AI models various epistemic poisons abound in the wild.
I am not in this space, question: are there "bad actors" that are known to feed AI models with poisonous information?
1. Online trolls and pranksters (who already taught several different AIs to be racist in a matter of hours - just for the LOLs).
2. Nation states like China who already require models to conform to state narratives.
3. More broadly, when training on "the internet" as a whole there is a huge amount of wrong, confused information mixed in.
There's also a meta-point to make here. On a lot of culture war topics, one person's "poisonous information" is another person's "reasonable conclusion."
Earlier quoted context omitted.
Nvidia can actually charge larger margins if inference compute goes down. It would enable them to manufacture more units of smaller GPUs using inferior and cheaper silicon, all of which would increase the profits per unit sold as well as the number of units they can manufacture. The industry has to find a way to separate itself from Nvidia's GPGPU technology if they want to stop being gouged. The issue is that nobody…
I think future of inference is on the client side You can do inference on almost any hardware, I do not see any edge for NVIDIA here I can download DeepSeek 30b model and run inference at good speed on AMD GPU ms and even on CPU. Apple silicon works fine too. I get >50 tokens/s on £300 AMD GPUs. The main bottleneck appears to be memory, not processing power.
Earlier quoted context omitted.
every time you respond to an AI model "no, you got that wrong, do it this way" you provide a very valuable piece of data to train on. With reasoning tokens there is just a lot more of that data to train on now
This assumes that you give honest feedback. Efforts to feed deployed AI models various epistemic poisons abound in the wild.
There’s a lot of incentive to figure this out. And they have so much data coming in that they can likely afford to toss out some good data to ensure that they’re tossing out all of the bad.
"The o3 system demonstrates the first practical, general implementation of a computer adapting to novel unseen problems" Yet, they said when it was announced: "OpenAI shared they trained the o3 we tested on 75% of the Public Training set. They have not shared more details. We have not yet tested the ARC-untrained model to understand how much of the performance is due to ARC-AGI data." These two statements are complet…
We just don't know how much the model seeing what an arc problem is on the first place boosts its ability to solve them - that limited statement is all the author is making.
Earlier quoted context omitted.
every time you respond to an AI model "no, you got that wrong, do it this way" you provide a very valuable piece of data to train on. With reasoning tokens there is just a lot more of that data to train on now
This assumes that you give honest feedback. Efforts to feed deployed AI models various epistemic poisons abound in the wild.
That said, I suspect the signal is very weak even today and probably not too useful except for learning about human stylistic preferences.
> But now with reasoning systems and verifiers, we can create brand new legitimate data to train on. This can either be done offline where the developer pays to create the data or at inference time where the end user pays! > This is a fascinating shift in economics and suggests there could be a runaway power concentrating moment for AI system developers who have the largest number of paying customers. Those customers…
every time you respond to an AI model "no, you got that wrong, do it this way" you provide a very valuable piece of data to train on. With reasoning tokens there is just a lot more of that data to train on now
>> Today's date is Tuesday, January 28, 2025.
> No, you're wrong, today's date is actually Wednesday the 29th.
>> My mistake. Yes, today's date is Wednesday, January 29th, 2025.
Three months later in April when this tagged data is used to train the next iteration, the AI can successfully learn that today's date is actually January 29th.
Earlier quoted context omitted.
It doesn't need to write tests: it can just use the application and figure out if it works.
That's going to be much slower and more expensive than writing tests because image/video processing is slower and more expensive than writing tests. And because of lag in using the UI (and re-building the whole application from scratch after every change to test again).
Ok, so if one can have one program snoop on all the rendering calls made by another program, maybe there could be a way of training a common representation of “an image of an application” and “the rendering calls that are made when producing a frame of the display for the application”? Hopefully in a way that would be significantly smaller than the full image data.
If so, maybe rather than feeding in the video of the application, said representation could be applied to the rendering calls the application makes each frame, and this representation would be given as input as the model interacts with the application, rather than giving it the actual graphics?
But maybe this idea wouldn’t work at all, idk.
Like, I guess the rendering calls often involve image data in their arguments, and, you wouldn’t want to include the same images many time as the input to the encoding thing, as that would probably (or, I imagine) make it slower than just using the overall image of the application. I guess the calls are probably more pointing to the images in memory though, not putting an entire image on the stack.
I don’t know enough about low-level graphics programming to know if this idea of mine makes any sense.
Earlier quoted context omitted.
every time you respond to an AI model "no, you got that wrong, do it this way" you provide a very valuable piece of data to train on. With reasoning tokens there is just a lot more of that data to train on now
> What is today's date? >> Today's date is Tuesday, January 28, 2025. > No, you're wrong, today's date is actually Wednesday the 29th. >> My mistake. Yes, today's date is Wednesday, January 29th, 2025. Three months later in April when this tagged data is used to train the next iteration, the AI can successfully learn that today's date is actually January 29th.
At best, you can only try to layer in contextual info like this as metadata during inference, akin to how other prompting layers exist.
Even then, what up-to-date information should present for every round-trip is a matter of opinion and use-case.