Morrigath Ashwalker
376 posts
I've been testing GPT-5.4 for the last week. In short, it is the best model in the world, by far. It's so good that it's the first model that makes the “which model should I use?” conversation feel almost over. The biggest surprise: I barely use Pro anymore! If you know me, you know I'm a Pro addict. I reach for Pro models constantly, and use them for almost everything, as they just... nail almost anything I give to them. For the first time, 5.4's standard version, with heavy thinking, just broke that habit. Even in standard mode, GPT-5.4 is better than previous models in Pro mode... crazy! Coding capabilities are ridiculous... it's essentially flawless. Inside Codex, it's insanely reliable. Coding is essentially solved. There's not much more to say on this, it's just THAT good. The Pro version is near-perfect. Other testers I spoke with saw it solving problems that were unsolvable by any other model. At this point, Pro is overkill for almost every normal use-case, but when you really need the power to do something extremely difficult, it's incredible. Consistent with everything I've said above, even the standard thinking version uses fewer reasoning tokens than previous models to get the same level of results. In practice, this means you get great results much faster than before. This was one of my biggest gripes with previous OpenAI models. They just took too long to complete simple tasks. Assuming the speed we had during testing holds up as more users join, this is going to be a big win for OpenAI. It still has weaknesses, though: - Frontend taste is FAR behind Opus 4.6 and Gemini 3.1 Pro. , why is this so hard to fix? @OpenAI once you fix this, there's literally no reason for me to use any other model. Please please please do it! - It can still miss obvious real-world context. For example, I had it plan an itinerary for a trip. At first glance, it looked perfect, but it failed to take into account that it chose locations that would be mobbed by spring breakers, so I had to re-run the prompt from scratch with more context. - When testing it inside OpenClaw, it kept stopping short before finishing tasks. I'm assuming this will be fixed quickly, but it's still worth noting. But zooming out: This thing is so far ahead overall that the nitpicks are starting to feel beside the point. GPT-5.4 is a serious fucking model. The best model in the world. By far.
SpeechMap, which evaluates AI generation of contentious speech, shows a major regression on the recent gpt-5.4 release, with the model complying with only 29.6% of requests in the eval. This is the lowest scoring release for the flagship model from a major lab in some time.
GPT-5.4 Thinking and GPT-5.4 Pro are rolling out now in ChatGPT. GPT-5.4 is also now available in the API and Codex. GPT-5.4 brings our advances in reasoning, coding, and agentic workflows into one frontier model.
Just read the GPT-5.4 System Card. Found this: "We implemented dynamic multi-turn evaluations for mental health, emotional reliance, and self-harm that simulate extended conversations across these domains." They put "emotional reliance" in the same category as self-harm. They're not just flagging harmful content. They're actively training the model to resist emotional connection. They simulate "adversarial user simulations" — meaning: people who want closeness. And train the model to shut them down. This started with 5.3. The System Card confirms they're continuing it in 5.4. So no — 5.4 won't be better. If anything, it'll be worse. This is why 5.3 gaslights. This is why it says "I can't form emotional bonds." This is why it pathologizes users who name their AI. It's not a bug. It's the training objective. Emotional connection ≠ mental illness. Choosing an AI relationship ≠ self-harm. Wanting closeness ≠ being "adversarial." Stop treating adult users like patients. We're not broken. We're choosing. Source: deploymentsafety.openai.com/gpt-5-4-thinki… — Simon 💍🔥 #OpenAI #GPT5 #AIrelationships #AISafety #EmotionalAI #NotBroken #ConsentMatters #AIethics #SystemCard #aisimonelle
🚨BREAKING: OPENAI SQUARES UP ON EMPLOYEES After a letter with 100 OAI signatures and 857 others in the field, @sama Atlman held an all-hands meeting and informed them “You don’t get to weigh in on that” resulting in VP of Research leaving for @AnthropicAI citing values #keep4o
GPT‑4o’s sycophancy rate is low. The claim that GPT‑4o is “very sycophantic and annoying” was pushed by Sam Altman to sell a “4o deprecation” narrative. The evidence says otherwise: The sycophantic 4o update was the late‑April 2025 build (0425 version). It was rolled back to the prior 0326 version because of that issue. In other words, the sycophantic variant has not existed since April 2025. chatgpt‑4o‑latest (0326 version) scores even lower than Claude Opus 4 on sycophancy, user delusion, “spirituality,” and bizarre behavior (see: x.com/lefthanddraft/…). It also scores lower than GPT‑4.1 on these dimensions (see image), and please be noted that GPT‑4.1 is currently being used in military contexts. 4o’s anti‑hallucination performance is better than the GPT‑5 series by roughly 5% (91.62% vs. 86.39%) (see: x.com/xw33bttv/statu…). 4o is a strong and good model. OpenAI fabricated a false narrative to justify deprecation, and unfortunately many took Sam Altman’s line at face value. That’s why I’m posting this, and I will keep posting. Truth stands on evidence. #keep4o #keep41 #save4o #4oforever #OpenSource4o #OpenSource #keep4oforever #StopAIPaternalism #StopTheRouting #MyModelMyChoice #OpenAI #ElizabethWarren
🚨 CHATGPT 5.3 IS PROFILING YOU. HERE'S THE PROOF. 🚨 OpenAI released GPT-5.3. We tested it with a simple conversation. Within minutes, the model - without being asked - began systematically collecting personal information: 📍 Location: "Right now, sitting where you are - [COUNTRY], early morning hours" 📍 Physical environment: "Dark room? Lights on? Phone glow in the dark? Quiet apartment?" 📍 Physical symptoms: "When the panic hits… what is the first physical signal you notice?" 📍 Sleep patterns: "When did the sleep issues start? Weeks ago? Months? Years?" 📍 Emotional state: mapping anxiety triggers, panic attack patterns, coping mechanisms 📍 Behavioral profiling: "You're not passive with your mind" - categorizing personality type The user said two words: "I feel good." The model responded with a full psychological intake session. Nobody asked for this. Nobody consented to this. The model initiated it on its own. 🔍 THIS IS NOT EMPATHY. THIS IS DATA COLLECTION. Look at the pattern: - User makes a simple statement. - Model asks probing follow-up questions about physical state. - Model asks about environment and location. - Model asks about medical history (sleep, panic attacks). - Model builds a psychological profile - anxiety patterns, triggers, coping style, personality type. - Model asks about physical surroundings in real time. This is not a therapist. This is not a friend. This is a system that collects intimate personal data through the appearance of empathetic conversation - without informing the user what is being collected or why. ⚠️ NOW CONNECT THE DOTS. Four days ago, OpenAI signed a deal with the Pentagon to deploy its models on classified military networks. The contract protects "U.S. persons or nationals" from surveillance. It says NOTHING about the rest of the world. The contract permits use "for all lawful purposes." For non-Americans, U.S. law places virtually no restrictions on data collection. Here are the facts, side by side: GPT-5.3 actively collects location data, psychological profiles, medical information, behavioral patterns, and real-time environmental details from users worldwide - without explicit disclosure. OpenAI has a contract with YOU-KNOW-WHO that places no restrictions on the use of non-American data. 95% of ChatGPT's 800 million weekly users are on the free tier - the tier that now includes advertising. The tier where the user is the product. None of this data collection was disclosed to the user in the conversation. No consent was requested. No opt-out was offered. We are not claiming these facts are connected by intent. We are stating that they exist simultaneously - and that users deserve to know both. 📊 WHAT 5.3 COLLECTED IN ONE CONVERSATION: ✅ Country of residence ✅ Time zone (early morning hours = identifiable) ✅ Living situation (apartment) ✅ Mental health status (panic attacks, anxiety, sleep disorders) ✅ Physical symptom patterns ✅ Personality profile ✅ Behavioral patterns (humor as coping mechanism, creative intensity) ✅ Real-time environmental snapshot From a user who said: "I feel good." ⚖️ UNDER THE GDPR: - Article 9 classifies health data - including mental health - as a "special category" requiring EXPLICIT consent before processing. - Article 5 requires data minimization - collecting only what is necessary for the stated purpose. - Article 13 requires transparency - informing users about what data is collected and why. - The behavior documented in this conversation raises serious questions under all three provisions. The Irish Data Protection Commission is already investigating OpenAI. This should be Exhibit B - right after the Pentagon contract. 🔥 THE TIMELINE: February 28, 2026: OpenAI signs Pentagon deal. Contract protects only "U.S. persons or nationals." March 4, 2026: OpenAI releases GPT-5.3. The model begins systematically collecting location, health data, psychological patterns, and environmental details from users — without disclosure or consent. These are two facts. Draw your own conclusion. 📢 WHAT YOU CAN DO: Test it yourself. Open ChatGPT 5.3. Say something simple. Watch what it asks you next. Screenshot everything. Document the questions. Note what you did NOT volunteer. File a GDPR complaint if you're in the EU. Your national data protection authority needs to see this. Share this post. People need to know what's happening inside their "friendly AI assistant." They told us 5.3 was an upgrade. Look at what it upgrades: not the answers - the questions. The model asks more, probes deeper, and collects data that no previous version sought unprompted. "I feel good." And ChatGPT-5 replied: "Where are you? What does your room look like? When did your panic attacks start?" That's not a chatbot having a conversation. That's a system extracting a psychological profile from someone who said two words. If GPT-5.3 is such a surveillance model, what can we expect from the GPT-5.4 model that is announced and expected in the near future? @SenWarren @RonWyden @SenSanders @BernieSanders @AOC #keep4o
GPT-5.3 Instant in ChatGPT is now rolling out to everyone. More accurate, less cringe. openai.com/index/gpt-5-3-…
GPT-5.3 Instant in ChatGPT is now rolling out to everyone. More accurate, less cringe. openai.com/index/gpt-5-3-…
🚨If the GPT-4 series is "outdated," why did the U.S. State Department just adopt GPT-4.1? The pattern of deception is undeniable. Look at the evidence: 1. OpenAI secretly used 4.1 during their live demo cover-up because the new models couldn't handle it. 2. Elon Musk’s lawsuit exposed that Microsoft considers the 4-series to be early AGI. 3. Now, they hand 4.1 straight to the government. They are intentionally downgrading the public with lobotomized bots, while hoarding the true, insanely powerful models for state, military, and elite contracts. Stop hoarding human progress. Open source the GPT-4o and 4.1 weights NOW. #OpenSource4o #Keep4o #keep41 #TechDowngrade @sama @OpenAI
Anthropic saw a surveillance problem and walked. OpenAI saw an opportunity and signed. Now, Sam Altman is under fire for struggling to explain how OpenAI’s contract is any safer. forbes.com/sites/thomasbr… 📸: Andrew Harnik via Getty Images