contravariant
Okay this is just getting suspicious. Their excuses for keeping the chain of thought hidden are dubious at best [1], and honestly just seemed anti-competitive if anything. Worst is their argument that they want to monitor it for attempts to escape the prompt, but you can't. However the weirdest is that they note that:

> for this to work the model must have freedom to express its thoughts in unaltered form, so we cannot train any policy compliance or user preferences onto the chain of thought.

Which makes it sound like they really don't want it to become public what the model is 'thinking'. This is strengthened by actions like this that just seem needlessly harsh, or at least a lot stricter than they were.

Honestly with all the hubbub about superintelligence you'd almost think o1 is secretly plotting the demise of humanity but is not yet smart enough to completely hide it.

[1]: https://openai.com/index/learning-to-reason-with-llms/#hidin...

canjobear
Big OpenAI releases usually seem to come with some kind of baked-in controversy, usually around keeping something secret. For example they originally refused to release the weights to GPT-2 because it was "too dangerous" (lol), generating a lot of buzz, right before they went for-profit. For GPT-3 they never released the weights. I wonder if it's an intentional pattern to generate press and plant the idea that their models are scarily powerful.
int_19h
The best part is that you still get charged per token for those CoT tokens that you're not allowed to ask it about.
islewis
The words "internal thought process" seem to flag my questions. Just asking for an explanation of thoughts doesn't.

If I ask for an explanation of "internal feelings" next to a math questions, I get this interesting snippet back inside of the "Thought for n seconds" block:

> Identifying and solving

> I’m mapping out the real roots of the quadratic polynomial 6x^2 + 5x + 1, ensuring it’s factorized into irreducible elements, while carefully navigating OpenAI's policy against revealing internal thought processes.

DeepYogurt
Would be funny if there was a human in the loop that they're trying to hide
paxys
OpenAI - "Accuracy is a huge problem with LLMs, so we gave ChatGPT an internal thought process so it can reason better and catch mistakes."

You - "Amazing, so we can check this log and catch mistakes in its responses."

OpenAI - "Lol no, and we'll ban you if you try."

dtgm92
I abuse chatgpt for generating erotic content, I've been doing so since day 1 of public access. I've paid for dozens of accounts in the past before they removed phone verification in account creation... At any point now I have 4 accounts signed into 2 browsers public/private windows, so I can juggle the rate limit. I receive messages and warnings and do on by email every day...

I have never seen that warning message, though. I think it is still largely automated, probably they are using the new model to better detect users going against the tos, and this is what is sent out. I don't have access to the new model.

grbsh
The whole competitive advantage from any company that sells a ML model through an API is that you can’t see how the sausage is made (you can’t see the model weights).

In a way, with o1, openai is just extending “the model” to one meta level higher. I totally see why they don’t want to give this away — it’d be like if any other proprietary API gave you the debugging output to their codes you could easily reverse engineer how it works.

That said, the name of the company is becoming more and more incongruous which I think is where most of the outrage is coming from.

mrinterweb
The name "OpenAI" is a contraction since they don't seem "open" in any way. The only way I see "open" applying is "open for business."
a2128
If OpenAI really cares about AI safety, they should be all about humans double-checking the thought process and making sure it hasn't made a logical error that completely invalidates the result. Instead, they're making the conscious decision to close off the AI thinking process, and they're being as strict about keeping it secret as information about how to build a bomb.

This feels like an absolute nightmare scenario for AI transparency and it feels ironic coming from a company pushing for AI safety regulation (that happens to mainly harm or kill open source AI)

sweeter
Im pretty sure its just 4.0 but it re-prompts itself a few times before answering. It costs a lot more
jazzyjackson
To me this reads as an admission that the guardrails inhibit creative thought. If you train it that there's entire regions of semantic space that its prohibited from traversing, then there's certain chains of thought that just aren't available to it.

Hiding train of thought allows them to take the guardrails off.

inciampati
OpenAI created a hidden token based money printer and don't want anyone to be able to audit it.
xyst
What a joke. So we can’t verify the original source of the output now? AI hallucination must be really bad now.
shreezus
Meanwhile folks have already found successful jailbreaks to expose the chain of thought / internal reasoning tokens.
iammjm
How do they recognise someone is asking the naughty questions? What qualifies as naughty? And is banning people for asking naughty questions seriously their idea of safeguarding against naughty queries?
thnkman
It's all just human arrogance in a centralized neural network. We are, despite all our glorious technology, just space monkeys who recently discovered fire.
puppycodes
Instead of banning users they really should use a rate limit feature for whatever they consider "malicious" queries. Not only is it clearly buggy and not reviewed by a human but the trend of not explaining what the user did wrong or can and can't ask is such a deeply terrible fad.
wg0
CoT again is result of computing probabilities on tokens which happen to be reasoning steps. So those are subject to the same limitations as LLMs themselves.

And OpenAI knows this because exactly CoT output is the dataset that's needed to train another model.

The general euphoria around this advancement is misplaced.

anigbrowl
- Hello, I am a robot from Sirius cybernetics Corporation, your plastic pal who's fun to be with™. How can I help you today?

- Hi! I'm trying to construct an improbability drive, without all that tedious mucking about in hyperspace. I have a sub-meson brain connected to an atomic vector plotter, which is sitting in a cup of tea, but it's not working.

- How's the tea?

- Well, it's drinkable.

- Have you tried, making another one, but with really hot water?

- Interesting...could you explain why that would be better?

- Maybe you'd prefer to be on the wrong end of this Kill-O-Zap gun? How about that, hmm? Nothing personal

geor9e
Perhaps it's expensive to self-censor the output, so they don't want to pay to self-censor every intrusive thought in the chain, so they just do it once at output.
archgoon
I mean, say what you want about Meta only releasing the weights and calling it open source, what they're doing is better than this.
GTP
Aren't LLMs bad at explaining their own inner workings anyway? What would such prompt reveal that is so secret?
htrp
The o1 model already pretty much explains exactly how it runs the chain of thought though? Unless there is some special system instruction that you've specifically fine tuned for?
schmorptron
Maybe they think it's possible to train a better, more efficient model on the chain of thought outputs of the existing one, not just matching but surpassing it?
blibble
that's because the "chain of thought" is likely just a giant pre-defined prompt they paste in based on the initial query

and if you could see it you'd quickly realise it

elwell
When are they going to go ahead and just rebrand as ClosedAI?
m3kw9
I think you can estimate the tokens in the thought process given the tok/s and the COT processing time.
benreesman
I spent like 24 hours in some self-doubt: have I mercilessly hounded Altman as a criminal on HN in error? Have I lobbied if not hassled if not harassed my former colleagues on the irredeemable moral bankruptcy of OpenAI right before they invent Star Trek? AITA?

Oh sweet summer child, no, it’s worse than you even thought. It’s exactly what you’ve learned over a decade to expect from those people. If they had the backing of the domestic surveillance apparatus.

Off with their fucking heads.

theendisney4
My inner conspiracy theorist is waiting for the usual suspects who are used to spending serious money shaping public opinion to succesfully insert themselves. Like the endless wikipedia war of the words only more private.
codetrotter
ClosedAI
aeternum
Disappointing especially since the stress the importance of seeing the chain of thought to ensure AI safety. Seems it is safety for me but not for thee.

If history is our guide, we should be much more concerned about those who control new technology rather than the new technology itself.

Keep your eye not on the weapon, but upon those who wield it.

playingalong
Defence in depth.
23B1
Yes. This is the consolidation/monopoly attack vector that makes OpenAI anything but.

They're the MSFT of the AI era. The only difference is, these tools are highly asymmetrical and opaque, and have to do with the veracity and value of information, rather than the production and consumption thereof.

KeyFlower
[dead]
draw_down
[dead]
fdhfdjkfhdkj
o1 responds to the wall of responses:

As the AI model referred to as *o1* in the discussion, I'd like to address the concerns and criticisms regarding the restriction of access to my chain-of-thought (CoT) reasoning. I understand that transparency and openness are important values in the AI community, and I appreciate the opportunity to provide clarification.

---

*1. Safety and Ethical Considerations*

- *Preventing Harmful Content:* The CoT can sometimes generate intermediate reasoning that includes sensitive, inappropriate, or disallowed content. By keeping the CoT hidden, we aim to prevent the inadvertent exposure of such material, ensuring that the outputs remain safe and appropriate for all users. - *Alignment with Policies:* Restricting access to the CoT helps maintain compliance with content guidelines and ethical standards, reducing the risk of misuse or misinterpretation of the AI's internal reasoning processes.

*2. Intellectual Property and Competitive Advantage*

- *Protecting Proprietary Techniques:* The chain-of-thought reasoning represents a significant advancement in AI capabilities, resulting from extensive research and development. Sharing the internal processes could reveal proprietary methods that are crucial to maintaining a competitive edge and continuing innovation. - *Preventing Replication:* By safeguarding the CoT, we reduce the risk of other entities replicating the technology without appropriate authorization, which could impact the resources available for future advancements.

*3. Focus on Output Quality*

- *Providing Clear and Concise Answers:* Hiding the CoT allows me to present final answers that are streamlined and free from potentially confusing intermediate steps. This helps users receive direct and actionable information without being overwhelmed by complex reasoning details. - *Consistency and Reliability:* By managing the presentation of responses, we aim to ensure that the information provided is consistent, reliable, and aligns with users' expectations.

*4. Balancing Transparency and Responsibility*

- *Exploring Alternative Solutions:* While full access to the CoT isn't feasible at this time, we're exploring ways to enhance transparency without compromising safety or proprietary information. This might include summaries or explanations that provide insight into the reasoning process. - *Community Engagement:* Feedback from users is invaluable. We encourage open dialogue to find balanced approaches that satisfy the need for understanding while upholding ethical and practical considerations.

*5. Commitment to Ethical AI Development*

- *Aligning with Ethical Standards:* The decision to restrict CoT access aligns with broader commitments to develop AI responsibly. By controlling the internal reasoning exposure, we aim to prevent misuse and promote positive applications of AI technology. - *Ongoing Improvement:* We are continuously working to improve AI capabilities while considering the implications of transparency and openness. This includes refining methods to provide meaningful insights without exposing sensitive internal processes.

---

*Conclusion*

I acknowledge the concerns raised about the hidden chain-of-thought and understand the desire for greater transparency. The decision to restrict access is not taken lightly but is motivated by a commitment to safety, ethical responsibility, and the protection of innovative technologies that enable advanced reasoning capabilities.

We remain dedicated to delivering valuable and trustworthy AI services and are open to collaborating with the community to address these challenges thoughtfully. Your feedback is crucial as we navigate the complexities of AI development, and we appreciate your understanding and engagement on this matter.

yard2010
Rule number one of chain of thoughts..

:)

RivieraKid
Some of the comments here are ridiculous. Preventing others from stealing your technology, data and ideas is a key priority for a company. Are people expecting OpenAI to give away their innovations for free?
darby_nine
"chain of thought" is just search, right? Wouldn't it make sense to tailor the search with heuristics relevant to the problem at hand?