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

Cross-Model Void Convergence: GPT-5.2 and Claude Opus 4.6 Deterministic Silence

zenodo.org

Post reply on HN
Print view View original

21–30 of 34 posts

Re: Cross-Model Void Convergence: GPT-5.2 and Claude Opus 4.6 Deterministic Silence

#21

Post by Lerc »

My thoughts while reading this, went.

What is this abstract even saying? Oh now I understand it's just needlessly wordy. Hmm paper with single author, I wonder if they posted it to HN? Let's see what else they've put out? Four variations of void so far this year.

The language makes it feel like woo, but it might just be banal. I can't descern a significant claim other than;

Models respond to their prompts

One of those responses can be just to immediately end the response.

They can prioritise more recent prompts in case of ambiguity.

Expected behaviour is expected on multiple models.

Re: Cross-Model Void Convergence: GPT-5.2 and Claude Opus 4.6 Deterministic Silence

#22

Post by barrkel »

The abstract is over-egged. The language obscures what it purports to find. So some prompts return null results.

"Ontologically null concepts" could just be a fancy way of saying "the model doesn't know what to do with nonsense". Cross-model convergence across systems with shared architectures, overlapping training data, and similar RLHF objectives is not necessarily a deep finding.

There's a high ratio of jargon-heavy interpretive superstructure to empirical foundation here.

Re: Cross-Model Void Convergence: GPT-5.2 and Claude Opus 4.6 Deterministic Silence

#23

Post by _2d30 »

Lerc wrote: post #21

My thoughts while reading this, went. What is this abstract even saying? Oh now I understand it's just needlessly wordy. Hmm paper with single author, I wonder if they posted it to HN? Let's see what else they've put out? Four variations of void so far this year. The language makes it feel like woo, but it might just be banal. I can't descern a significant claim other than; Models respond to their prompts One of thos…

There A LOT of esotericists/occultists writing research like this about AI today. It is heavily woo.

Re: Cross-Model Void Convergence: GPT-5.2 and Claude Opus 4.6 Deterministic Silence

#24

Post by GrinningFool »

embedding-shape wrote: post #16
qayxc wrote: post #12

Earlier quoted context omitted.

This is an interesting observation. So maybe it has nothing to do with the model itself, but everything to do with external configuration. Token-limit exceeded -> empty output. Just a guess, though.

> Token-limit exceeded -> empty output. Just a guess, though. That'd be really non-obvious behavior, I'm not aware of any inference engine that works like that by default, usually you'd get everything up until the limit, otherwise that kind of breaks the whole expectation about setting a token-limit in the first place...

I just fixed this bug in a summarizer. Reasoning tokens were consuming the budget I gave it (1k), so there was only a blank response. (Qwen3.5-35B-A3B)

Re: Cross-Model Void Convergence: GPT-5.2 and Claude Opus 4.6 Deterministic Silence

#25

Post by embedding-shape »

GrinningFool wrote: post #24
embedding-shape wrote: post #16

Earlier quoted context omitted.

> Token-limit exceeded -> empty output. Just a guess, though. That'd be really non-obvious behavior, I'm not aware of any inference engine that works like that by default, usually you'd get everything up until the limit, otherwise that kind of breaks the whole expectation about setting a token-limit in the first place...

I just fixed this bug in a summarizer. Reasoning tokens were consuming the budget I gave it (1k), so there was only a blank response. (Qwen3.5-35B-A3B)

Most inference engines would return the reasoning tokens though, wouldn't you see that the reasoning_content (or whatever your engine calls it) was filled while content wasn't?

Re: Cross-Model Void Convergence: GPT-5.2 and Claude Opus 4.6 Deterministic Silence

#27

Post by GrinningFool »

embedding-shape wrote: post #25
GrinningFool wrote: post #24

Earlier quoted context omitted.

I just fixed this bug in a summarizer. Reasoning tokens were consuming the budget I gave it (1k), so there was only a blank response. (Qwen3.5-35B-A3B)

Most inference engines would return the reasoning tokens though, wouldn't you see that the reasoning_content (or whatever your engine calls it) was filled while content wasn't?

Yeah, I had been ignoring the reasoning tokens for the summarize call

Re: Cross-Model Void Convergence: GPT-5.2 and Claude Opus 4.6 Deterministic Silence

#28

Post by frde_me »

bob1029 wrote: post #2

Title for the back of the class: "Prompts sometimes return null" I would be very cautious to attribute any of this to black box LLM weight matrices. Models like GPT and Opus are more than just a single model. These products rake your prompt over the coals a few times before responding now. Telling the model to return "nothing" is very likely to perform to expectation with these extra layers.

Out of curiosity, are there any sources to there being a significant amount of other steps before being fed into the weights

Security guards / ... are the obvious ones, but do you mean they have branching early on to shortcut certain prompts?

Re: Cross-Model Void Convergence: GPT-5.2 and Claude Opus 4.6 Deterministic Silence

#29

Post by beering »

This is 100% crank “science” that has wrapped up a banal finding in big words and LaTeX. The claim is roughly as exciting as, “some computer programs print nothing to stdout.”

The output shown is not “null” or “void”. It is the empty string, which these LLMs are perfectly capable of outputting. Technically, it outputs the stop token, analogous to \0 at the end of a C string.

Re: Cross-Model Void Convergence: GPT-5.2 and Claude Opus 4.6 Deterministic Silence

#30

Post by bob1029 »

frde_me wrote: post #28
bob1029 wrote: post #2

Title for the back of the class: "Prompts sometimes return null" I would be very cautious to attribute any of this to black box LLM weight matrices. Models like GPT and Opus are more than just a single model. These products rake your prompt over the coals a few times before responding now. Telling the model to return "nothing" is very likely to perform to expectation with these extra layers.

Out of curiosity, are there any sources to there being a significant amount of other steps before being fed into the weights Security guards / ... are the obvious ones, but do you mean they have branching early on to shortcut certain prompts?

> do you mean they have branching early on to shortcut certain prompts?

Putting a classifier in front of a fleet of different models is a great way to provide higher quality results and spend less energy. Classification is significantly cheaper than generation and it is the very first thing you would do here.

A default, catch-all model is very expensive, but handles most queries reasonably well. The game from that point is to aggressively intercept prompts that would hit the catch-all model with cheaper, more targeted models. I have a suspicion that OAI employs different black boxes depending on things like the programming language you are asking it to use.

Post reply on HN