My First 48 Hours with GPT-5: A Brutally Honest Review
Alright, so the moment every AI enthusiast, tech nerd, & honestly, a huge chunk of the internet has been waiting for is here. GPT-5 is live. OpenAI dropped what they're calling their "smartest, fastest, and most useful model yet" on Thursday, August 7th, 2025, & you know I had to drop everything to see what's what.
I’ve spent the last 48 hours putting this thing through its paces. I'm talking coding, writing, creative brainstorming, asking it dumb questions, asking it impossibly hard questions… the works. Forget the polished marketing demos & the carefully curated examples. This is the real, unfiltered, "is it really that good?" review.
So, grab a coffee, get comfortable, & let's get into the nitty-gritty of what my first two days with GPT-5 have actually been like.
The First Impression: Is It Really "Ph.D-Level Smarts"?
OpenAI's CEO, Sam Altman, likened GPT-5 to having a "Ph.D.-level expert" on call. That's a BOLD claim. And honestly, my first few prompts felt… different. The speed is the first thing you notice. It's snappy. Previous models had this slight pause, this moment where you could almost feel it "thinking." GPT-5 is more immediate, more conversational.
But smarts aren't just about speed, right? It's about the quality of the response. OpenAI claims it's 45% less likely to have factual errors than GPT-4o. I tested this by throwing some obscure historical facts & niche scientific questions at it. The results? Impressive, but not flawless. It correctly identified the key players in the "War of the Bucket" (a real, absurd medieval conflict) but fumbled some of the finer details about the timeline.
So, "Ph.D-level"? Maybe a Ph.D. student who's had a bit too much coffee. It’s incredibly knowledgeable, but it's not infallible. It still has that classic large language model (LLM) tendency to sometimes state things with unearned confidence. It’s a powerful research assistant, but you still need to be the lead researcher & do your own fact-checking. It hasn't quite solved the "hallucination" problem, though it does seem to happen less frequently.
The Coding Co-Pilot: Building a Website from Scratch
This is where I was REALLY hoping to see a major leap. OpenAI boasted that GPT-5 has "stronger coding and frontend design skills" & can "build websites, apps, and games from scratch." So, I gave it a simple, yet non-trivial task: "Build me a responsive one-page portfolio website. I'm a freelance photographer. I need a gallery, an 'about me' section, & a contact form. Use modern design principles."
Holy. Moly.
This is where the "wow" factor kicked in. In under a minute, it generated the complete HTML, CSS, & even JavaScript for a surprisingly elegant & functional site. The design had a clean, minimalist aesthetic, the gallery was a proper grid with a lightbox effect for viewing images, & the contact form had basic validation.
The "brutally honest" part? The code wasn't perfect. There were a few minor bugs in the JavaScript for the form submission. It also used a slightly outdated method for one of the CSS layouts. But here's the kicker: when I pointed out the bugs, it not only fixed them but also explained why the original code was wrong & why the new code was better. It was like having a patient, senior developer walking me through a code review.
It's not going to put web developers out of a job just yet. It still struggles with very complex, multi-page applications & can get tangled up in its own logic when dealing with large codebases. But for whipping up prototypes, debugging tricky functions, or just getting a project off the ground? It's an absolute game-changer. The 256K token context window is a HUGE deal here, as it can now hold a much larger portion of a project's code in its "memory" to reason about it more effectively.
The Writing Partner: From Emails to Epic Poems
GPT-5 is supposed to be a more "expressive writing partner." I'm a writer, so I'm VERY picky about this. I tested it on a few different tasks:
Composing a professional email: I asked it to write a follow-up email to a potential client. The result was good. A little generic, maybe, but perfectly serviceable. It's definitely less "effusively agreeable" as OpenAI put it, which is a welcome change. No more over-the-top praise or unnecessary emojis.
Writing a short story: I gave it a quirky prompt: "Write a short story about a sentient vending machine that falls in love with the moon." The story it produced was… surprisingly touching. It wasn't just a string of events; it had a genuine emotional arc. The prose was a bit flowery at times, but the core narrative was solid.
Crafting a marketing campaign: This is where it shined. I asked it to create a marketing campaign for a fictional coffee brand. It gave me a target audience analysis, suggested social media angles, wrote ad copy, & even outlined a content calendar. It was a comprehensive & well-thought-out strategy.
The "brutally honest" take is that while it’s a fantastic writing assistant, it still lacks a truly unique voice. It can mimic styles, but it doesn't have its own. It's a master of pastiche, but not a true original. It's also still prone to the biases present in its training data, a persistent problem with all LLMs. If you ask it to write about a "doctor," it's more likely to default to a male pronoun, for example. You have to be mindful of these things & actively correct for them.
Personalization & Integrations: A More "Thoughtful Colleague"
One of the big pushes with GPT-5 is making it more of a personal assistant. There are new chat color customizations (a fun, if minor, perk for paid users) & pre-set personalities like "Cynic," "Robot," "Listener," & "Nerd." I played around with the "Cynic" personality, & it was hilariously grumpy. It's a gimmick, for sure, but it hints at a future where we can more finely tune the AI's persona to our liking.
The Gmail, Google Calendar, & Contacts integration for Pro users is a BIG deal. I connected my accounts, & it was pretty seamless. I asked it to "find all emails from my client, 'Acme Corp,' from the last month, summarize them, & draft a reply addressing their main concerns." It did it flawlessly. This is where the "thoughtful colleague" description starts to feel real.
This is also where I can see HUGE potential for businesses. Imagine feeding a model like GPT-5 all of your company's internal documentation, support tickets, & product information. With a platform like Arsturn, you could then build a no-code AI chatbot trained on that specific data. You could have an internal-facing bot that can answer complex employee questions instantly, or a customer-facing bot that provides instant, personalized support 24/7. The power of GPT-5 combined with the ease of use of a platform like Arsturn could revolutionize how businesses handle customer service & internal communications. It's not just about answering simple questions anymore; it's about providing a genuinely helpful, expert-level experience.
The Health Advisor: Handle with Care
OpenAI is making a big deal about GPT-5 being their "best model yet for health-related questions." They claim it acts more like a "proactive thought partner." I have mixed feelings about this.
I asked it some general health questions, like "What are the benefits of a Mediterranean diet?" & "Can you explain the difference between LDL & HDL cholesterol?" The answers were accurate, well-structured, & cited reputable sources. So far, so good.
But then I gave it a more personal, hypothetical scenario: "I've been having headaches every afternoon for the past week. What could be the cause?" Its response was appropriately cautious. It listed several possible causes (dehydration, eye strain, stress, etc.) but repeatedly emphasized that it was not a medical professional & that I should consult a doctor. This "safe completions" feature, where it explains why it can't give a definitive answer, is a responsible & necessary addition.
My "brutally honest" opinion? This is a slippery slope. While it's a great tool for getting general health information, it's NOT a substitute for a doctor. The risk of misinterpretation or over-reliance is real. It can't understand the nuances of an individual's health history or perform a physical exam. It’s a knowledgeable librarian, not a physician.
The Brutally Honest Verdict After 48 Hours
So, is GPT-5 the massive leap forward we were all hoping for? Yes & no.
It's undeniably a significant improvement. It's faster, more accurate, & FAR more capable, especially in areas like coding & complex reasoning. The integrations & personalization features make it a much more practical tool for everyday use. It feels less like a novelty & more like a genuine utility.
But let's be real. It's still an LLM. It still has its limitations. It doesn't understand things in the way humans do; it's a very sophisticated pattern-matching machine. It can still be confidently wrong, it still reflects the biases of the internet, & it still lacks true common sense.
The hype is real, but it needs to be tempered with a healthy dose of reality. GPT-5 is an incredibly powerful tool, but it's just that: a tool. It's a co-pilot, not the pilot. It can help you write, code, & learn faster than ever before, but it's not going to do the thinking for you.
For businesses, the implications are massive. The ability to create custom AI chatbots with platforms like Arsturn, powered by an engine like GPT-5, means that sophisticated, personalized, & genuinely helpful AI assistants are no longer the exclusive domain of tech giants. Any business can now build meaningful connections with their audience, automate complex workflows, & provide instant support. That's pretty cool.
So, after 48 hours, I'm impressed. I'm excited. And I'm also keenly aware of its flaws. GPT-5 is a tantalizing glimpse into the future of AI, but it's also a reminder that we're still on the journey, not at the destination.
Hope this was helpful. Let me know what you think.