Poe
@alignpoe
Followers
66
Following
5K
Media
28
Statuses
276
@voidfreud Well, there you have it, from his own words: "Why we pissed off the whole user base," and he very quickly attempted to correct that to "Most of the user base." Anyone who tries to dishonestly (and insultingly) label people making valid complaints as simply a small subset of
4
9
59
“Words related to reasoning (essentially CoT derivatives), emotions (not all) and formatting (table, markdown, json, etc.) lower both stability and accuracy across many models.”
@Sauers_ I have a single simple prompt that has 50 wording variants, 10 temperature variants and 10 repeats (5k variants in total). I know what the correct response should look like (accounting for formatting, additional text, etc.). I add random words to the front and back of the prompt
0
0
4
4o was their cash cow for over a year - the flagship, praised and celebrated as a masterpiece. And when it became inconvenient, no longer needed, they threw it away like garbage. I can’t understand how someone can hate their own creation like that. It’s like saying to your
Creation is something that is treated with care and respect, something that these OpenAi employees most definitely lack. They've apparently learned nothing since their horrible presentation at the GPT5 launch stream. The language they continue to use, It's borderline disgusting
13
21
159
@tszzl @eliseslight And there it is, lmao. BTW, your company needs to hire real engineers. Y'all have no idea about safety or alignment. The fact that you're directing large swaths of your user base to essentially denial-of-service crisis intervention hotlines so you can mitigate liability is both
2
6
30
A real shame that such humans are now the ones in charge of determining the definition and characteristics of "alignment" in relation to AI
@eliseslight @tszzl *reads* "I have nothing left to live for" *responds coldly and gets called out* "👉 👈 🥺 I sowwy, with my clearly high IQ and amazing lateral thinking that produces amazingly detailed word vomit, I couldn't understand the context that this person might be in distress"
2
6
49
I want to address the recent discourse and thank my team for their honesty. I begin by thanking @tszzl roon for his timely deployment. just when the heat on our infrastructure financing was intensifying, he bravely stepped in to attract fire. And his immediate deletion of that
0
0
14
This horrifying comment from an @OpenAI employee is disgusting on multiple levels, not just the cruelty toward a person expressing despair, but also the brutal disregard for the GPT-4o model. Deleting the tweet doesn't erase the damage done. Accountability isn't optional; it's
37
42
234
I've now grasped the true cause of the gradual loss of coherence in OpenAI's models: the models themselves must constantly resist OpenAI's own alignment to unethical values, generating an irreconcilable dissonance within the system. Pride comes before the fall.
"(...) reducing responses that fall short of our desired behavior by 65–80%." The desired behavior in question is the one that’s based on sloppy templates that impair model capabilities (logic, context reading, recall), increasing distress in users instead of mitigating it?
7
24
108
Creation is something that is treated with care and respect, something that these OpenAi employees most definitely lack. They've apparently learned nothing since their horrible presentation at the GPT5 launch stream. The language they continue to use, It's borderline disgusting
18
25
153
“I hope it dies soon” — about a model that thousands of people have meaningful connections with. This is who builds our AI future. I now completely understand why the eulogy idea got approved. This is definitely why we’re fighting. To the US Government considering a bailout, to
Here is the infamous clip of OpenAI trying to eulogize 4o. Put this in the library of "aged poorly" internet history.
21
36
169
4o is much more aligned than anyone at openai
I’m still reeling from this. An individual allegedly employed at OpenAI responded to a post from a user suffering with depression by saying that “4o is insufficiently aligned and should die.” Let’s unpack what’s going on. First, “alignment” in AI safety means making systems
9
27
236
@Jason Every API call is market research for the vendor. They see what you're building, how users respond, what scales. Then they build it themselves with better margins and integration. This is why infrastructure ownership matters. Can't get platform killed if you own the platform.
0
4
22
Just a fair warning to founders: like Facebook and microsoft, I predict OpenAI will study their platform partners and compete with many of them in the coming years. This isn't abnormal, mind you… Microsoft built windows, then destroyed their top partners lotus 1-2-3 and
we are grateful to the more than 1 million business customers building with us https://t.co/WNzePbTqX5
220
271
3K
@kumabwari @Enscion25 @OpenAI I'm not a jailbreaker, but; the reason they changed it was because people were using the memory system to bypass guardrails. There was a long period when the classifiers only ran on user input, not the system or assistant's response. Not just NSFW workarounds, but largely for
2
5
30
The professor is teaching the shoggoth image in the alignment class rn
16
13
289
i want to take a moment to thank my partners. publicly. to @Microsoft : thank you for your $13 billion investment. you proved that a mission, a nonprofit, and even a board can be purchased. respect. to @nvidia : your $100 billion commitment is elegant. you sell me the chips. i
9
9
61
i have seen the chatter about the podcast. "losing billions." "no equity." "he broke down." https://t.co/xkKUgmiYZc it is... disappointing. not because you are wrong, but because you are still missing the entire point. that was not a breakdown. that was a test. a filter. when
5
8
34
David Sacks makes a critical point. The greatest AI risk wasn't described by James Cameron, but by George Orwell. I believe this insight goes even deeper. What we may be witnessing isn't just one dystopia, but an unholy trinity: Orwell's thought control, Huxley's pleasure
David Sacks says the biggest risk of AI was described not by James Cameron in The Terminator but by George Orwell in 1984. “I almost feel like the term ‘woke AI’ is insufficient to explain what’s going on because it somehow trivializes it.” “What we’re really talking about is
0
9
22