I didn't quite realise this – the reasoning traces are actually encrypted and the model can access them via previous_response_id but you can't. Not that I particularly want to read them, but good to know.
ETHICS
-
Understanding Without Comprehension Hardy Quote
By
–
It is easy to be impressed by what one does not understand very well. (G.H. Hardy)
-
Scale AI’s Data Labeling Quality Problem: A Critical Analysis
By
–
> be me
> Alexandr Wang, born January 1997
> age 19, drop out of MIT
> co-found Scale AI
> "what if we label data, but mid?"
> convince every LLM company that this is fine > 2016–2023
> flood the market with barely-labeled goat photos and out-of-context Reddit takes
> call it -
Building AI from First Principles Versus Surface-Level Implementation
By
–
meme building from first principles all you want without it you're a duct-taper
-
AI Company Criticized for Misleading Performance Claims Gaslighting
By
–
i tell everyone, at least i respect sama, dario? nope let's remember this is a company that gaslit us over months about their daytime performance and if it wasn't for the uproar online we would've never gotten their 3 bugs post-mortem
-
Anthropic Palantir partnership raises AI safety alignment concerns
By
–
Anthropic is literally in deals with Plantir, safety alignment is a joke coming from them lol they're pretending to be the good guys
-
Open Source AI: Risks of Proprietary Model Control
By
–
this guy is malicious. opensource ftw. a reminder that
— Ahmad (@TheAhmadOsman) 22 septembre 2025
> not your Weights means eventually not the same Model btw
they can
> quantize it
> distill it
> hot-swap to a cheaper/weaker checkpoint
> make the model manipulative
> fine-tune it in ways that break safety or depth
>… https://t.co/LVPP2kX0gTthis guy is malicious. opensource ftw. a reminder that > not your Weights means eventually not the same Model btw they can > quantize it
> distill it
> hot-swap to a cheaper/weaker checkpoint
> make the model manipulative
> fine-tune it in ways that break safety or depth
> -
Stealth Account Deletion Restrictions Raise Fairness Concerns
By
–
Afaik the official reasoning was that non-stealth accounts could use this to produce stealth images (by generating the job, downloading and then deleting), but I don't see why stealth accounts cannot delete. I think this is unfair to users
-
NP-Complete Problems Make LLM Output Validation Fundamentally Limited
By
–
NP-complete problems are hard to solve and easy to verify, so validating LLM output is a Band-Aid and always will be.