Lo, it hath come to pass that OpenAI is engrossed in the arduous task of tending to its GPT language models, for there have been murmurs and accusations swirling around GPT-4. Whispers of laziness, of dullness, and of errors aberrant to the usual behavior of the ChatGPT chatbot did traverse the realms of social media in the waning days of November.
Verily, some do conjecture that GPT-4.5, in clandestine fashion, hath been discreetly ushered into the presence of select users, guided by certain responses from the ChatGPT itself. Whether this be truth or mere fiction, behold, there linger within the sanctum of GPT-4 some positive shifts.
## More GPUs, Enhanced Performance?
As tales began to unfurl as early as the last Thursday, they spoke of the metamorphosis in the performance of GPT-4. Ethan Mollick, a sage of Wharton, who once lamented the decline of GPT-4’s acumen in the month of November, now doth herald a revitalization in the model’s capabilities, without bearing witness to the fabled GPT-4.5 switch. Employing a code interpreter to right his code, he doth recount the transformation as “night and day, both in swiftness and in the quality of answers,” after enduring weeks of ChatGPT-4’s unreliability and somnolence.
Whilst this transpired, in the shadows did OpenAI quietly unveil anew its ChatGPT Plus subscriptions on the Wednesday past, when registration had lain dormant since the 14th day of November. In an X post, Altman proclaimed, “Thanks for thy forbearance as we sought more GPUs.”
Whether a link doth exist betwixt the resurrection of ChatGPT Plus enrollments and the enhancement of GPT-4 remaineth shrouded in mystery, yea, the timing is nigh intriguing. Noteworthy is it that the signups for the premium iteration culminated soon after the inaugural DevDay of OpenAI, where a plethora of novel functions for the paid variant of the AI chatbot were unfurled. A waiting list was imposed for ChatGPT Plus subscriptions due to the inundation of post-DevDay signups, surpassing the service’s capacity to process functions.
People bemoaned how GPT-4 would elucidate the execution of commands rather than executing the tasks themselves.
Posthaste, users began to relay peculiar behavior from GPT-4 beyond the standard AI banter known to all. One common grievance articulated was that GPT-4 would “back talk” to users or necessitate multiple explanations of a command ere fulfilling the query. Another lamentation voiced that the model would explicate to users how to enact their command rather than executing the deed.
The decline of GPT-4 hath roots reaching back at least unto July, when a study noted a marked drop in accuracy betwixt the months of March and June. Many, including the Product Vice President of OpenAI, Welinder, hath posited that the quality of responses might appear lacking as a psychological phenomenon heralding further updates of the model. Some have advised that users could benefit from altering their queries to receive desired outcomes.
Though OpenAI hath remained mostly reticent regarding its inner machinations, Altman’s X post concerning GPUs doth serve as a telltale sign of the covert endeavors afoot. Reports from April foretold that OpenAI would require an excess of 30,000 GPU units to sustain its commercial performance for the remaining portion of the year, ere the surge in interest come November.
## Secret GPT-4.5 Trials or Mere Fantasies?
Furthermore, the chitchat regarding GPT-4.5 hath surged forth with sundry tidings of a potential new iteration of GPT leaking into the ether.
Rowan Cheung, founder of the @therundownai bulletin, lately divulged, on X, pricing particulars for a nascent GPT-4.5 model being crafted by OpenAI. These particulars include fresh pricing tiers and intel on advanced multimodal capabilities.
Cheung didst inquire of OpenAI’s CEO Sam Altman via the social forum on the truth of the leak, to which he did quip, “Nay.”
The speculations of GPT 4.5 took flight on the Thursday past, with a ‘leaked’ image showcasing a novel GPT-4.5 model bedecked with advanced multimodal capabilities and novel pricing.
Nevertheless, Sam Altman uttered a firm “nay” when confronted with queries about the verity of such rumors.
But lo, there’s more to this tale…
— Rowan Cheung (@rowancheung) December 18, 2023
Several users who hath handled the proposed update doth swear upon using GPT-4.5, finding it to be unprecedented and superior in its essence. Some have interrogated the chatbot, which they deemed to be powered by GPT-4, about its model — and it hath responded, “GPT-4.5 Turbo.”
This hath led many to believe that OpenAI hath been scrutinizing GPT-4.5, primarily through its mobile applications, aiming to elude the discerning eye of users. Yet not all have been able to replicate such findings, and the model doth inform them that the current version doth remain mere GPT-4. OpenAI’s own Will Depue didst render a commentary on this matter, dubbing it a “most peculiar and oddly consistent hallucination.”
Cheung noted a post from the official ChatGPT X page, bedight with the emojis of brain and clouds, which he perceives as OpenAI’s cryptic method of reaffirming that these responses doth exist solely in the realm of hallucination.
Noteworthy is the bestowal of early and exclusive access by OpenAI unto its esteemed paid users, encompassing the ChatGPT Plus subscribers, the API developers, and the enterprise users. When features are unveiled, these are among the first to partake in the latest versions of GPT and its functions. Many who do tinker with the models and discern the subtle changes likely hath some semblance of API access or engage with the service to test code, thereafter imparting their insights unto the world.
Even if OpenAI be testing GPT-4.5, the hour of a forthcoming update doth elude us, especially when the company doth seek to regain stability in its services and structures. Too, the company and its product remain in their infancy, and the path of routine updates hath yet to emerge. For now, meseems the “version” of ChatGPT matters not, for the improvements doth appear genuine.
Support our work ❤️
If you enjoyed this article, consider leaving a tip to help us keep publishing great content.
