<\/p>\n
OpenAI released a paper last week detailing various internal tests and findings about its o3 and o4-mini models. The main differences between these newer models and the first versions of ChatGPT we saw in 2023 are their advanced reasoning and multimodal capabilities. O3 and o4-mini can generate images, search the web, automate tasks, remember old conversations, and solve complex problems. However, it seems these improvements have also brought unexpected side effects. It’s best to think of GPT-4.5 as a stepping stone to systems OpenAI plans to offer in the future.<\/p>\n<\/p>\n
That limit with GPT-3.5 and the old version of ChatGPT was 4,096 \u201ctokens,\u201d which is around 8,000 words, or roughly four to five pages of a book. So it would sort of lose track of things after they passed that far \u201cback\u201d in its attention function. GPT-4 is a next-generation language model for the AI chatbot, and though OpenAI isn\u2019t being specific about what changes it\u2019s made to the underlying model, it is keen to highlight how much improved it is over its predecessor.<\/p>\n<\/p>\n
<\/p>\n
<\/p>\n
OpenAI has just announced GPT-4, the latest iteration of its deep learning language model that is touted to be smarter and more capable than ChatGPT. The Microsoft-backed company claims that ChatGPT scored in the 10th percentile in the Uniform Bar exam, while GPT managed a staggering 90th. In a biology Olympiad, ChatGPT reached the 31st percentile rank, while GPT-4 raced far ahead to the 99th percentile. One of the early adopters is Microsoft, which says the Bing Chat bot now relies on GPT-4. Notably, OpenAI’s announcement doesn’t mention anything about the bias problem, which we found was a recurring problem with ChatGPT in performing creative tasks.<\/p>\n<\/p>\n
<\/p>\n
Practical experiments also show that GPT-4.5 is better than other general-purpose models at remaining true to facts and following user instructions. However, given that this is OpenAI\u2019s largest and most powerful non-reasoning model, it is worth considering its strengths and the areas where it shines. Wharton professor and AI commentator Ethan Mollick posted on social media that GPT-4.5 is a \u201cvery odd and interesting model,\u201d noting it can get \u201coddly lazy on complex projects\u201d despite being a strong writer. One of the biggest advantages of the models is its reduced latency (lag) even with higher performance on intelligence evaluations, such as the Multilingual (MMLU) benchmark, seen below. The models were purpose-built for developers and, as a result, will only be available via the API. Kimi K2 was trained and optimised for a fraction of the eye-watering sums it is blowing on GPUs and compute for marginal gains.<\/p>\n<\/p>\n
Its successor, GPT-5, will likely integrate many of OpenAI’s latest technologies, including its frontier o3 model. OpenAI reiterated that today, saying it plans to bring GPT-4.5’s “unique strengths, including broader knowledge, stronger intuition, and greater ‘EQ,’ to all users in future models.” Tokens \u2013 the basic units that large language models process \u2013 are now going to be a lot cheaper on several GPT models. OpenAI describes tokens as pieces of words; input tokens are the pieces of words that make up prompts, whereas output tokens make up responses. GPT-4 is the latest language model for the ChatGPT AI chatbot, and despite just being released, it\u2019s already making waves.<\/p>\n<\/p>\n
Set against that track record, the GPT-4 paper marks a new milestone in lack of disclosure. The decision not only to withhold source code and withhold the finished program, but also to withhold technical details that would allow outside researchers to guess about the composition of the program, is a new kind of omission. GPT-4, as its name suggests, is the fourth version of what’s known as a “generative pre-trained transformer,” a program built to manipulate human language.<\/p>\n<\/p>\n
<\/p>\n
GPT2.5 was \u201cskipped\u201d straight into GPT3, which was even more interesting. GPT3.5 crossed the threshold where it was enough to actually ship as a product and sparked OpenAI\u2019s \u201cChatGPT moment\u201d. And GPT4 in turn also felt better, but I\u2019ll say that it definitely felt subtle. Since the new models offer similar or improved performance at a lower cost than GPT-4.5, the company also announced it is deprecating GPT-4.5 and focusing on building future models. To give developers ample time to transition, GPT-4.5 Preview will be turned off on July 14, 2025. GPT-4 held the previous crown in terms of context window, weighing in at 32,000 tokens on the high end.<\/p>\n<\/p>\n
That\u2019s a very approximate description of how the attention mechanism and token count work, but the general idea is of expanded memory and the capabilities that accompany it. Millions of Bing users, however, can already access the new Bing chatbot. If you’re in that select group, Microsoft has confirmed that GPT-4 is already powering your chatbot interactions.<\/p>\n<\/p>\n
<\/p>\n
Testing the two model versions with a data set of 500 problems, researchers observed that GPT-4 had a 97.6% accuracy rate in March with 488 correct answers and a 2.4% accuracy rate in June after GPT-4 had gone through some updates. The AI world is dominated by English speakers, and everything from data to testing to research papers are in that language. But of course the capabilities of large language models are applicable in any written language and ought to be made available in those. Speaking of reduced hallucinations, OpenAI measured how much better GPT-4.5 in that regard. Obviously, the new model doesn’t solve the problem of AI hallucinations altogether, but it is a step in the right direction.<\/p>\n<\/p>\n
For all that today\u2019s chatbots get right, they tend to be easily led astray. A little coaxing can persuade them that they are simply explaining what a \u201cbad AI\u201d would do, or some other little fiction that lets the model say all kinds of weird and frankly unnerving things. People even collaborate on \u201cjailbreak\u201d prompts that quickly let ChatGPT and others out of their pens. As impressive as GPT-4 was at launch, some onlookers have observed that has lost some of its accuracy and power. These observations have been posted online for months now, including on the OpenAI forums.<\/p><\/p>\n","protected":false},"excerpt":{"rendered":"
OpenAI’s GPT-4: How to get access right now OpenAI released a paper last week detailing various internal tests and findings about its o3 and o4-mini models. The main differences between these newer models and the first versions of ChatGPT we saw in 2023 are their advanced reasoning and multimodal capabilities. O3 and o4-mini can generate …<\/p>\n