Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

S1: A $6 R1 competitor?

timkellogg.me

Post reply on HN
Print view View original

201–210 of 430 posts

Re: S1: A $6 R1 competitor?

#201

Post by andrewfromx »

koala_man wrote: post #140
mtrovo wrote: post #133

I found the discussion around inference scaling with the 'Wait' hack so surreal. The fact such an ingeniously simple method can impact performance makes me wonder how many low-hanging fruit we're still missing. So weird to think that improvements on a branch of computer science is boiling down to conjuring the right incantation words, how you even change your mindset to start thinking this way?

It feels like we're back in 1900 when anyone's clever idea (and implementation) can give huge performance improvements, such as Ford's assembly line and Taylor's scientific management of optimizing shovel sizes for coal.

yes, it also feels like we are going to lose our just-in-time global shipments of anything to anywhere any day now. It will soon feel like 1900 in other ways.

Re: S1: A $6 R1 competitor?

#202

Post by zoogeny »

bloomingkales wrote: post #91

If an LLM output is like a sculpture, then we have to sculpt it. I never did sculpting, but I do know they first get the clay spinning on a plate. Whatever you want to call this “reasoning” step, ultimately it really is just throwing the model into a game loop. We want to interact with it on each tick (spin the clay), and sculpt every second until it looks right. You will need to loop against an LLM to do just about…

I can't believe this hasn't been done yet, perhaps it is a cost issue.

My literal first thought about AI was wondering why we couldn't just put it in a loop. Heck, one update per day, or one update per hour would even be a start. You have a running "context", the output is the next context (or a set of transformations on a context that is a bit larger than the output window). Then ramp that up ... one loop per minute, one per second, millisecond, microsecond.

Re: S1: A $6 R1 competitor?

#203

Post by spiorf »

teruakohatu wrote: post #193
xg15 wrote: post #175

Earlier quoted context omitted.

I think the fact alone that distillation and quantization are techniques that can produce substantial improvements is a strong sign that we still have no real comprehensive understanding how the models work. If we had, there would be no reason to train a model with more parameters than are strictly necessary to represent the space's semantic structure. But then it should be impossible for distilled models with less p…

> still have no real comprehensive understanding how the models work. We do understand how they work, we just have not optimised their usage. For example someone who has a good general understanding of how an ICE or EV car works. Even if the user interface is very unfamiliar, they can figure out how to drive any car within a couple of minutes. But that does not mean they can race a car, drift a car or drive a car on…

We know how the next token is selected, but not why doing that repeatedly brings all the capabilities it does. We really don't understand how the emergent behaviours emerge.

Re: S1: A $6 R1 competitor?

#204

Post by gessha »

teruakohatu wrote: post #193
xg15 wrote: post #175

Earlier quoted context omitted.

I think the fact alone that distillation and quantization are techniques that can produce substantial improvements is a strong sign that we still have no real comprehensive understanding how the models work. If we had, there would be no reason to train a model with more parameters than are strictly necessary to represent the space's semantic structure. But then it should be impossible for distilled models with less p…

> still have no real comprehensive understanding how the models work. We do understand how they work, we just have not optimised their usage. For example someone who has a good general understanding of how an ICE or EV car works. Even if the user interface is very unfamiliar, they can figure out how to drive any car within a couple of minutes. But that does not mean they can race a car, drift a car or drive a car on…

Your example is somewhat inadequate. We _fundamentally_ don’t understand how deep learning systems works in the sense that they are more or less black boxes that we train and evaluate. Innovations in ML are a whole bunch of wizards with big stacks of money changing “Hmm” to “Wait” and seeing what happens.

Would a different sampler help you? I dunno, try it. Would a smaller dataset help? I dunno, try it. Would training the model for 5000 days help? I dunno, try it.

Car technology is the opposite of that - it’s a white box. It’s composed of very well defined elements whose interactions are defined and explained by laws of thermodynamics and whatnot.

Re: S1: A $6 R1 competitor?

#205

Post by layer8 »

zoogeny wrote: post #202
bloomingkales wrote: post #91

If an LLM output is like a sculpture, then we have to sculpt it. I never did sculpting, but I do know they first get the clay spinning on a plate. Whatever you want to call this “reasoning” step, ultimately it really is just throwing the model into a game loop. We want to interact with it on each tick (spin the clay), and sculpt every second until it looks right. You will need to loop against an LLM to do just about…

I can't believe this hasn't been done yet, perhaps it is a cost issue. My literal first thought about AI was wondering why we couldn't just put it in a loop. Heck, one update per day, or one update per hour would even be a start. You have a running "context", the output is the next context (or a set of transformations on a context that is a bit larger than the output window). Then ramp that up ... one loop per minute…

Same. And the next step is that it must feed back into training, to form long-term memory and to continually learn.

Re: S1: A $6 R1 competitor?

#206

Post by layer8 »

nico wrote: post #171

> In s1, when the LLM tries to stop thinking with " ", they force it to keep going by replacing it with "Wait". It’ll then begin to second guess and double check its answer. They do this to trim or extend thinking time (trimming is just abruptly inserting " ") I know some are really opposed to anthropomorphizing here, but this feels eerily similar to the way humans work, ie. if you just dedicate more time to analyzin…

What’s missing in that analogy is that humans tend to have a good hunch about when they have to think more and when they are “done”. LLMs seem to be missing a mechanism for that kind of awareness.

Re: S1: A $6 R1 competitor?

#207

Post by zoogeny »

layer8 wrote: post #205
zoogeny wrote: post #202

Earlier quoted context omitted.

I can't believe this hasn't been done yet, perhaps it is a cost issue. My literal first thought about AI was wondering why we couldn't just put it in a loop. Heck, one update per day, or one update per hour would even be a start. You have a running "context", the output is the next context (or a set of transformations on a context that is a bit larger than the output window). Then ramp that up ... one loop per minute…

Same. And the next step is that it must feed back into training, to form long-term memory and to continually learn.

I analogize this with sleep. Perhaps that is what is needed, 6 hours offline per day to LoRa the base model on some accumulated context from the day.

Re: S1: A $6 R1 competitor?

#208

Post by zoogeny »

Aperocky wrote: post #121

For all the hype about thinking models, this feels much like compression in terms of information theory instead of a "takeoff" scenario. There are a finite amount of information stored in any large model, the models are really good at presenting the correct information back, and adding thinking blocks made the models even better at doing that. But there is a cap to that. Just like how you can compress a file by a lot…

I think this is probably accurate and what remains to be seen is how "compressible" the larger models are.

The fact that we can compress a GPT-3 sized model into an o1 competitor is only the beginning. Maybe there is even more juice to squeeze there?

But even more, how much performance will we get out of o3 sized models? That is what is exciting since they are already performing near Phd levels on most evals.

Re: S1: A $6 R1 competitor?

#209

Post by mmoustafa »

Love the look under the hood! Specially discovering some AI hack I came up with is how the labs are doing things too.

In this case, I was also forcing R1 to continue thinking by replacing with “Okay,” after augmenting reasoning with web search results.

https://x.com/0xmmo/status/1886296693995646989

Re: S1: A $6 R1 competitor?

#210

Post by int_19h »

zoogeny wrote: post #202
bloomingkales wrote: post #91

If an LLM output is like a sculpture, then we have to sculpt it. I never did sculpting, but I do know they first get the clay spinning on a plate. Whatever you want to call this “reasoning” step, ultimately it really is just throwing the model into a game loop. We want to interact with it on each tick (spin the clay), and sculpt every second until it looks right. You will need to loop against an LLM to do just about…

I can't believe this hasn't been done yet, perhaps it is a cost issue. My literal first thought about AI was wondering why we couldn't just put it in a loop. Heck, one update per day, or one update per hour would even be a start. You have a running "context", the output is the next context (or a set of transformations on a context that is a bit larger than the output window). Then ramp that up ... one loop per minute…

The hard part is coming up with a good way to grade results. Which you need to update the weights based on the outcome, otherwise the model will not actually learn anything.
Post reply on HN