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

Learning to Reason with LLMs

openai.com

Post reply on HN
Print view View original

721–730 of 1001 posts

Re: Learning to Reason with LLMs

#721

Post by mitthrowaway2 »

kmeisthax wrote: post #622

>We believe that a hidden chain of thought presents a unique opportunity for monitoring models. Assuming it is faithful and legible, the hidden chain of thought allows us to "read the mind" of the model and understand its thought process. For example, in the future we may wish to monitor the chain of thought for signs of manipulating the user. However, for this to work the model must have freedom to express its thoug…

> For example, in the future we may wish to monitor the chain of thought for signs of manipulating the user.[...] Therefore we have decided not to show the raw chains of thought to users.

Better not let the user see the part where the AI says "Next, let's manipulate the user by lying to them". It's for their own good, after all! We wouldn't want to make an unaligned chain of thought directly visible!

Re: Learning to Reason with LLMs

#722

Post by arunv »

fshbbdssbbgdd wrote: post #544
jazzyjackson wrote: post #448

Earlier quoted context omitted.

Isn't there a big "Share" button at the top right of the chatgpt interface? Or are you using another front end?

I often click on those links and get an error that they are unavailable. I’m not sure if it’s openAI trying to prevent people from sharing evidence of the model behaving badly, or an innocuous explanation like the links are temporary.

They were probably generated using a business account, and the business does not allow public links.

Re: Learning to Reason with LLMs

#723

Post by bamboozled »

jdthedisciple wrote: post #702
bamboozled wrote: post #691

Earlier quoted context omitted.

Haven't used ChatGPT* in over 6 months, not saving my ass at all.

I bet you've still used other models that were inspired by GPT.

I've used co-pilot, I turned it off, kept suggesting nonsense.

Re: Learning to Reason with LLMs

#724

Post by ActionHank »

WXLCKNO wrote: post #203
dinobones wrote: post #4

Generating more "think out loud" tokens and hiding them from the user... Idk if I'm "feeling the AGI" if I'm being honest. Also... telling that they choose to benchmark against CodeForces rather than SWE-bench.

Exploring different approaches and stumbling on AGI eventually through a combination of random discoveries will be the way to go. Same as Bitcoin being the right combination of things that already existed.

Crypto being used as an example of how we have moved forward successfully as a species is backward toilet sitting behaviour.

Re: Learning to Reason with LLMs

#725

Post by thih9 »

dvt wrote: post #639
kmeisthax wrote: post #622

>We believe that a hidden chain of thought presents a unique opportunity for monitoring models. Assuming it is faithful and legible, the hidden chain of thought allows us to "read the mind" of the model and understand its thought process. For example, in the future we may wish to monitor the chain of thought for signs of manipulating the user. However, for this to work the model must have freedom to express its thoug…

It's clear to me that OpenAI is quickly realizing they have no moat. Even this obfuscation of the chain-of-thought isn't really a moat. On top of CoT being pretty easy to implement and tweak, there's a serious push to on-device inference (which imo is the future), so the question is: will GPT-5 and beyond be really that much better than what we can run locally?

Why would a non profit / capped profit company, one that prioritizes public good, want a moat? Tongue in cheek.

Re: Learning to Reason with LLMs

#726

Post by bad_username »

Prompt:

> Alice, who is an immortal robotic observer, orbits a black hole on board a spaceship. Bob exits the spaceship and falls into the black hole. Alice sees Bob on the edge of the event horizon, getting closer and closer to it, but from her frame of reference Bob will remain forever observable (in principle) outside the horizon. > > A trillion year has passed, and Alice observes that the black hole is now relatively rapidly shrinking due to the Hawking radiation. How will Alice be observing the "frozen" Bob as the hole shrinks? > > The black hole finally evaporated completely. Where is Bob now?

O1-preview spits out the same nonsense that 4o does, telling that as the horizon of the black hole shrinks, it gets closer to Bob's apparent position. I realize that.the prompt is essentily asking to solve the famous unsolved problem in physics (black hole information paradox), but there's no need to be so confused with basic geometry of the situation.

Re: Learning to Reason with LLMs

#727

Post by selimthegrim »

worstspotgain wrote: post #436
cs391231 wrote: post #408

Student here. Can someone give me one reason why I should continue in software engineering that isn't denial and hopium?

Well, the fact that you typed this question makes me think that you're in the top X% of students. That's your reason. Those in the bottom (100-X)% may be better off partying it up for a few years, but then again the same can be said for other AI-affected disciplines. Masseurs/masseuses have nothing to worry about.

I am pretty sure there is a VC funded startup making massage robots

Re: Learning to Reason with LLMs

#728

Post by deathanatos »

rovr138 wrote: post #624
avodonosov wrote: post #614

Earlier quoted context omitted.

The screenshot [1] is not readable for me. Chrome, Android. It's so blurry that I cant recognize a single character. How do other people read it? The resolution is 84x800.

When I click on the image, it expands to full res, 1713x16392.3

> it expands to full res, 1713x16392.3

Three tenths of a pixel is an interesting resolution…

(The actual res is 1045 × 10000 ; you've multiplied by 1.63923 somehow…?)

Re: Learning to Reason with LLMs

#729

Post by ActionHank »

sturza wrote: post #686

It seems like it's just a lot of prompting the same old models in the background, no "reasoning" there. My age old test is "draw a hand in ascii" - i've had no success with any model yet.

It seems like their current strat is to farm token count as much as possible.

1. Don't give the full answer on first request. 2. Each response needs to be the wordiest thing possible. 3. Now just talk to yourself and burn tokens, probably in the wordiest way possible again. 4. ??? 5. Profit

Guaranteed they have number of tokens billed as a KPI somewhere.

Re: Learning to Reason with LLMs

#730

Post by holoduke »

cs391231 wrote: post #408

Student here. Can someone give me one reason why I should continue in software engineering that isn't denial and hopium?

Software development just becomes a level tier higer for most developers. Instead of writing everything yourself you will be more like an orchestrator. Tell the system to write this, tell the system to connect that and this etc. You still need to understand code. But maybe in the future even that part becomes unreadable for us. We only understand the high level concepts.
Post reply on HN