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

GPT-5.5 hallucinates 3x more than MIT-licensed GLM-5.2

arrowtsx.dev

Post reply on HN
Print view View original

311–318 of 318 posts

Re: GPT-5.5 hallucinates 3x more than MIT-licensed GLM-5.2

#311

Post by palmotea »

jmalicki wrote: post #277
palmotea wrote: post #272

Earlier quoted context omitted.

> There are journalists being hired to write Atlantic-worthy articles that exist only as LLM training data, because they're getting paid more than the Atlantic would pay them for it. I kinda doubt that quality-assertion of "Atlantic-worthy." While I have no doubt such articles are written solely as training data, I'd expect their quality to be much less than the real thing, since there's no public to critique them, p…

That's where the review process comes in. I've found the review process for these things to be far more vicious and demanding than in the real world.

> That's where the review process comes in.

> I've found the review process for these things to be far more vicious and demanding than in the real world.

1. for your software or for these "Atlantic-worthy articles"?

2. Why would the review process hold up to the standards of a first-rate publication? I'd expect the review process to succumb to the same pressures I'd outlined.

"Someone paid me to do it" will only get you so much motivation. And asking someone to write something whose only destiny is to get fed into the maw of a machine and mangled, seems like a way to eliminate all other sources of motivation.

Re: GPT-5.5 hallucinates 3x more than MIT-licensed GLM-5.2

#312

Post by coldtea »

unshavedyak wrote: post #304
coldtea wrote: post #287

Earlier quoted context omitted.

> I suspect (ignorant armchair) we have some ability to signal strength of those facts, via repetition. Without this layer of introspection Isn't that precisely what the LLM training does? It signals strength of those facts, via repetition.

Yea but i don't think the LLM can inspect that. Eg it's similar to a code function , the code executes but the code doesn't have the ability to inquire about the code itself. In that silly example/thought, the LLM would effectively need the ability to query the strength of a fact/spatial region/etc. Right now i believe the LLM is more just the output of those weights. It has no way to inspect the strength of the sign…

>Right now i believe the LLM is more just the output of those weights. It has no way to inspect the strength of the signal. Eg it doesn't know if blue in "The sky is " is a strong or weak signal, it just predicted that next token.

Yes, but the prediction is itself encoding that it's a strong signal.

Meaning that, if given a context (prompt), it predicted that "the sky is blue", that's because the sky being blue is a stronger signal under that context - statistically speaking.

Re: GPT-5.5 hallucinates 3x more than MIT-licensed GLM-5.2

#313

Post by jgalt212 »

jmalicki wrote: post #307
HarHarVeryFunny wrote: post #297

Earlier quoted context omitted.

Could you give a specific example - just the outline/shape of what it consisted of (or more detail if you care to)?

It was something that took me 8 hours of very hard intellectual work. It was, like some of the higher level contracts, something where I had to create a training example for AI out of whole cloth and prove it was roughly at the boundary of things AIs could solve, and I had to write very detailed rubrics to judge potential solutions and apportion partial credit. I'm not comfortable saying more.

I hope you found writing that response as disatisfying as I found reading it.

Re: GPT-5.5 hallucinates 3x more than MIT-licensed GLM-5.2

#314

Post by jmalicki »

palmotea wrote: post #311
jmalicki wrote: post #277

Earlier quoted context omitted.

That's where the review process comes in. I've found the review process for these things to be far more vicious and demanding than in the real world.

> That's where the review process comes in. > I've found the review process for these things to be far more vicious and demanding than in the real world. 1. for your software or for these "Atlantic-worthy articles"? 2. Why would the review process hold up to the standards of a first-rate publication? I'd expect the review process to succumb to the same pressures I'd outlined. "Someone paid me to do it" will only get…

Journalism is not my field, so I can't deeply judge how effective the reviews are.

In software, the reviews for generating things that only exist as training data are far more rigorous than anything I've seen for things going into production at top companies.

I agree with you that it would make sense it would be phoned in. My experience says the opposite.

Re: GPT-5.5 hallucinates 3x more than MIT-licensed GLM-5.2

#315

Post by claytongulick »

coldtea wrote: post #298
claytongulick wrote: post #293

Earlier quoted context omitted.

> but might have zero influence in your thinking at the brain operation level Most of the people that I follow that research this stuff at the doctoral level say that they (and modern science) have very little idea about how consciousness works, the mind, and its relationship to the physical parts of the brain. We know about a lot of relationships, we can damage parts of the brain and see that damage expressed throug…

The gravity in the paper, as far as I can see, is about the influence of microgravity (e.g. much much smaller than regular Earth gravity) influence in blood circulation, bodily fluids etc, and as such on sleep, breathing, heart function, and other such aspects. Not about whether gravity affects the synapses and neurotransmitters in their regular function. We know of course that e.g. bad sleep or lack of water affects…

Not sure if you read the same paper I did? From the abstract:

"Research over the past decade has demonstrated two co-occurring patterns of spaceflight effects on the brain and behavior: dysfunction and adaptive plasticity. Evidence indicates the spaceflight environment induces adverse effects on the brain, including intracranial fluid shifts, gray matter changes, and white matter declines. Past work also suggests that the spaceflight environment induces adaptive neural effects such as sensory reweighting and neural compensation."

But that's just a single example. The effects of gravity (or lack thereof) on the brain are widely studied, it's pretty easy to find.

Besides, that's all sort of missing the point. I chose gravity as an example of a "huh, I never really thought about that" thing that affects cognition to illustrate the point that there are a gazillion things that impact human cognition, from cat parasites [1] to prescription eyeglasses [2] not to mention all the obvious things like gut health, diet and sleep.

Our brains and bodies have an essentially infinite number of inputs that affect our reasoning - I don't think this is practical, or even possible, to model - and I don't think you can achieve human level reasoning without it.

You can't separate the package from the wrapping when dealing with the human mind. Our biology is our cognition.

[1] https://pmc.ncbi.nlm.nih.gov/articles/PMC1635495/

[2] https://mindeye.com/eye-opener-prescription-glasses-can-affe...

Re: GPT-5.5 hallucinates 3x more than MIT-licensed GLM-5.2

#316

Post by goodness4all »

wolttam wrote: post #113

> it is clear that actual intelligence has plateaued significantly. > Moving forward, the industry cannot continue to train bigger and bigger models since their intelligence not only plateaus but often will get worse These are wild claims - why are we concluding that bigger models and more data = more hallucination? That’s actually the opposite of what’s been happening over the last couple years. Some models may stil…

I’m studying the root of probabilities and it’s impossible to have models without probable “hallucinations”. If we hit truth 20 times and 1 miss, we still would not consider it truth. This is the mathematical foundation these models are built & trained upon. Probability is our way of life, yet truth is subjective in life. Why use AI when you could just use a database if humans want determinism? The statistical mirror IS the power in AI.

Re: GPT-5.5 hallucinates 3x more than MIT-licensed GLM-5.2

#318

Post by unshavedyak »

coldtea wrote: post #312
unshavedyak wrote: post #304

Earlier quoted context omitted.

Yea but i don't think the LLM can inspect that. Eg it's similar to a code function , the code executes but the code doesn't have the ability to inquire about the code itself. In that silly example/thought, the LLM would effectively need the ability to query the strength of a fact/spatial region/etc. Right now i believe the LLM is more just the output of those weights. It has no way to inspect the strength of the sign…

> Right now i believe the LLM is more just the output of those weights. It has no way to inspect the strength of the signal. Eg it doesn't know if blue in "The sky is " is a strong or weak signal, it just predicted that next token. Yes, but the prediction is itself encoding that it's a strong signal. Meaning that, if given a context (prompt), it predicted that "the sky is blue", that's because the sky being blue is a…

Stronger, yes, but not necessarily strong. 2 instances of hearing that the sky is green might be stronger than 1 sky is blue, but that might not fit an expected threshold of say 100 instances of learning for a "fact" to be actually strong. At least in my example.

My point was that i feel like humans have these two aspects, the ability to have a fact, and the ability to have a signal to the facts strength. I propose that as an explanation why we can internally analyze our understanding and come to a conclusion that yes, we do "know" it.

We also at times can't figure out how we "know" it, either because we've made up a detail (filling in blanks, assumptions, etc), or because we forgot where we embedded this detail. The lower the signal strength of this validation the more we can be unsure about a "fact".

I feel like LLMs are one half, but not the other. They effectively need a RAG for all of the knowledge they have, and if the counter for a given fact/idea/etc is low enough, then it's an low signal.

The question i have is how to do this efficiently. Of course i'm just speculating too, i don't know any of this.

Post reply on HN