GPT-4.5 Shocks the World: Is AI Getting Dumber?
In this article, we’ll break down what went wrong with GPT-4.5, its limitations, and what this means for the future of artificial intelligence.
It is official the AI hype train just went on life support with the underwhelming release of GPT 4.5 yesterday open AI unveiled the most expensive AI model ever produced yet it fails to crush any benchmarks win any awards or offer any novel capabilities whatsoever its only real selling point is Vibes and is supposed to chat in a more natural human-like way don’t get me wrong it’s a good model but not good enough to feed the AI hype monster and it looks increasingly likely that we’re not headed into a technological
singularity but rather a sigmoid of sorrow the Sam ultman couldn’t even be bothered to leave his newborn kid in the hospital to show up to the product launch and instead send in a bunch of interns to demo it and that’s crazy because we’re talking about Orion here in 2023 Tech leaders signed a petition to stop training big models like this mman himself begged the government to regulate it and the only thing more disappointing than GPT 4.
4.5 is that it’s extremely expensive if you thought Claude was expensive at $15 per million tokens GPT 4.5 is five times more expensive at $75 per million output tokens actually no correction that’s input tokens it’s $150 per million output tokens and to chat with it it’s it’s currently only available to the $200 per month Pro users I tried it out myself and it does seem to emit Chill Vibes but the problem is that’s highly subjective however in the launch open aai talked about a new Vibes Benchmark that’s supposed to measure creative
In thinking the best way to get a f for the model is to talk to it so let’s jump into a demo a lot of people on the internet criticize this presentation but as an introvert myself I think they did a great job in addition it apparently has a far lower hallucination rate but what I found is that it still makes a lot of silly mistakes it’s not self-aware and has has no idea what gpt 4.5 even is and says its training cut off is October 2023 it was however able to tell me how many RS are in Strawberry that felt like a huge leap forward but I quickly became disappointed when it gave me the wrong number of L’s in laap paloa now when it comes to programming and science I didn’t even try because we already know it’s not going to perform as well as the deep thinking models like 03 then to make matters worse on the AER polyglot coding Benchmark it’s not only worse at programming than deep seek but also hundreds of times more expensive
now if you’re an Elon Musk hater you’ll want take a bong rip of copium right now because currently x’s Gro is the best model in the world that’s not my opinion it’s the opinion of the betting Market although by the end of 2025 open AI is still the favorite to have the best model but its odds are on the decline that’s problematic for open AI though because they’re raising billions and billions of dollars as they transition to for-profit and will need to maintain a massive valuation Alman says there is no wall and believes they can scale.
These models almost infinitely that’s assuming he gets trillions of dollars from soft Bank in the Saudis to build these data centers my theory as an unqualified ship poster is that they failed to train GPT 5 with any significant Improvement despite scaling up the number of parameters in compute GPT 4.5 is the biggest model they’ve ever created and now they’re lowering the bar for gpt 5 which ultimate described a few weeks ago being more like a router that automatically chooses the best model based on your prompt and that’s highly disappointing because I was expecting to be a post-apocalyptic warlord by now the battling robots and barbecuing rats over burning garbage cans for dinner but instead I live in this dystopia where artificial super intelligence never comes and nothing ever happens but if you’re a computer science student the plateau is great news AI coding tools are incredible but they’re most useful to real human programmers who know what they’re doing and I don’t see that changing anytime soon and you can start getting really good at programming for free thanks to this video sponsor brilliant their platform provides interactive Hands-On lessons that demystify the complexity of deep learning with just a few minutes of effort each day you can understand the math and computer science behind this seemingly magic technology I’d recommend starting with python then check out their full how large language models work course if you really want to look under the hood of chat gpt try everything brilliant has to offer for free for 30 days by going to brilliant.
Conclusion
The AI hype train just took a major hit with the release of GPT-4.5. OpenAI unveiled that what is supposedly the most advanced and expensive AI model ever created? Yet, instead of breaking records or introducing groundbreaking innovations, it has left the tech community questioning whether AI is truly evolving or simply plateauing. GPT-4.5 was supposed to be a revolution, but it turned out to be an expensive, underwhelming update.
FAQs About GPT-4.5
1. What is GPT-4.5?
GPT-4.5 is OpenAI’s latest AI language model, designed to improve conversational fluency and reduce hallucinations but failing to introduce any groundbreaking capabilities.
2. Is GPT-4.5 better than GPT-4?
While it offers smoother interactions and marginally fewer hallucinations, GPT-4.5 lacks major improvements in reasoning, coding, or creative problem-solving.
3. How much does GPT-4.5 cost?
GPT-4.5 is priced at $75 per million input tokens and $150 per million output tokens, making it one of the most expensive AI models available.
4. Can GPT-4.5 replace human programmers?
No. AI coding tools are useful assistants but still require human expertise to function effectively. GPT-4.5 has even shown worse performance on certain programming benchmarks.
5. Will GPT-5 be a major improvement?
Initial reports suggest that GPT-5 might function more as an AI router rather than a drastically improved model, which has left many users disappointed.
