The Unacceptable Downgrade: Why GPT-5 Forced Me to Cancel My OpenAI Subscription

xAI's Grok-3 might not be perfect but it happily generated this image for me.

For quite some time now, OpenAI's GPT-4o mini model has been an indispensable tool in my daily workflow. Its consistent reliability and impressive efficiency made it the go-to resource for a multitude of personal and professional projects, ranging from meticulous content review to rapid information retrieval. For months, this lightweight yet powerful iteration of their flagship model served its purpose admirably, providing swift and accurate responses that significantly streamlined my tasks.

However, after several frustrating days of being forcibly transitioned to GPT-5 without any option to revert, my reliance on OpenAI abruptly ended; I cancelled my subscription. This decision was not made lightly, given the prior utility of the service, but it became an unavoidable consequence of a fundamental misstep in product deployment.

Let me be clear: I am confident that for the average user, who primarily seeks a conversational AI chatbot for general inquiries, GPT-5 might indeed offer a perfectly serviceable experience. Nevertheless, for power users like myself—individuals who integrate AI deeply into complex workflows encompassing coding, extensive writing, data analysis, and diverse problem-solving scenarios—GPT-5 proves to be profoundly unsuitable. The primary grievances stem not only from its forced imposition without any graceful migration path or fallback option, but also from its inherent performance characteristics. Its processing time, particularly when it frequently transitions into its "thinking" mode, represents a significant and unacceptable bottleneck.

And while OpenAI has introduced an account setting to "enable legacy models," which purportedly re-enables GPT-4o, this offers little solace, as it specifically excludes the faster 4o-mini model that formed the cornerstone of my optimized workflows. Consequently, the considerable time and effort I invested in refining prompts for 4o-mini to yield precise, rapid results have been rendered entirely useless. GPT-5, by its very design, is a more deliberate, reasoning-focused model, engineered for deeper analysis and comprehensive output. While this architectural choice may serve certain advanced computational tasks, it is entirely antithetical to my frequent need for quick, direct analysis and immediate output based on specific inputs. The abrupt removal of 4o-mini is not only counter to sound application design—stripping users of a critical feature with no transition window—but it has also proven profoundly disruptive to my established professional tasks. This unforeseen change prompted an immediate re-evaluation of my AI vendor strategy, leading me to discover that the same prompts previously tailored for 4o-mini could be adapted with only minor adjustments for lightweight models offered by other providers, facilitating an immediate and seamless switch.

Beyond the performance degradation, GPT-5 also exhibits a marked decline in its ability to adapt to nuanced stylistic instructions, particularly concerning writing tasks. Even when provided with highly specific writing style guidelines, its output often feels less personal and more overtly robotic, departing significantly from the organic quality achievable with 4o-mini. One might speculate if this more generic, less adaptable tone is an intentional design choice, perhaps aimed at mitigating concerns around academic integrity or sophisticated content generation. Regardless of the underlying motive, as an individual who relies on AI for refined textual output, I am profoundly disappointed by this stylistic regression.

This sentiment of frustration and disappointment is far from isolated. After enduring sufficient frustration with GPT-5, despite meticulously crafting my prompts to coax optimal performance, I sought validation and solutions within the broader user community—specifically, the often-unfiltered forums of Reddit. The collective criticism there is remarkably sharp and consistent. For instance, Reddit user "larrybudmel" succinctly captured the prevailing sentiment, commenting, "The tone of mine is abrupt and sharp. Like it’s an overworked secretary. a disastrous first impression." Another user, "syntaxjosie," offered a particularly incisive observation, stating, "The only reason I can figure that they would deprecate the other models the day of release is because they know 5 is inferior and don't want people comparing them side by side." Furthermore, "Potato3445" encapsulated the widespread disillusionment: "Can’t believe we waited 2 years and took a step backwards. The creative writing is worse, it’s adopted a corporate personality, and it rarely bothers to follow instructions or incorporate your preferences without you having to ask. I hope the coders are happy atleast."

The forced adoption of GPT-5 by OpenAI serves as a critical cautionary tale for all technology companies. It underscores a fundamental principle: never presume that an upgrade, regardless of your internal conviction that it is "better," will be universally welcomed or even functional for your entire user base. Users, particularly those deeply embedded within an ecosystem, often possess distinct needs and established workflows that can be severely disrupted by unilateral, non-optional changes. The decision to compel users onto a new, less suitable model without offering alternatives or a clear migration path is not merely inconvenient; it is a profound misjudgment of user expectations and loyalty, inevitably leading to churn.

Ken is a cybersecurity and IT professional with over 15 years experience. All opinions are his own and do not reflect those of his employer or clients.

Moving Beyond Web3 - How Peer-to-Peer and Personal Branding is the Future of Communication

Commonly I see Web3 being associated with decentralized finance, blockchain, cryptocurrency, and NFTs. And while that's likely an excellent example of Web3, that's not what Web3 truly is at its core. Web3 is much more than that. Web3 is a true information revolution, laying the foundations for Web4. I had a great conversation last night with the Diamond Hand Media Group about this concept, and thought I'd go a little more in-depth here.

Let's step in the time machine for a moment and go through the history of the web. And I, being older than the Internet, can happily step you through.

Web1 - Static websites, news sites, email. Everybody paid per minute for access to the web. Sign on, find what you need, sign off so you don't get charged extra.

Web1.5 - This is when the potential of the web started to take shape. We added in chat rooms, instant messaging, and forums. Geocities let us even publish our own (limited) webpages! And now, unlimited internet access!  Suddenly, the world got a little bit smaller, as we started to communicate across the globe.

Web2 - Behold, broadband and social media! YouTube, Myspace, and eventually Facebook and Twitter! Blogs also started to rapidly grow, and the redistribution of content creation from commercial publishers to users started to take shape. But unfortunately, commercial publishers looked to continue controlling the narrative, continue controlling the audience, continue controlling the message. Everything is still centrally managed and owned by a select few companies, and social media "networks" aren't actually networks at all, but distribution hubs. One-way live streams of audio and video start to take off, because we actually have the internet connection speeds to support this type of content.

Web2.5 - Gnutella, Limewire, and other filesharing networks enter the stage, and early peer-to-peer distributed computing is born.

Web3 - Distributed finance, distributed content, distributed knowledge. Through blockchain, crypto, and NFTs, "digital ownership" can be established for assets, and distributed finance can allow for digital currency transactions without the need for a bank or the Federal reserve. For content creation, anyone can create content and share with others, and even have multi-party livestream audio and video sessions. No longer are we locked into getting our news and information from publishers, but instead shared directly person-to-person. But this person-to-person sharing is still limited to rely on distribution hubs such as social media networks, and even when using a network such as Mastodon (which could arguably be considered Web3.5), users still rely on a centralized hub to connect. Love him or hate him, the effects of this concept of direct person-to-person information sharing are now showing through Tucker Carlson's announcement of his own show on Twitter, and the massive reach this announcement has achieved. Carlson is now, on his own, likely going to get just as many if not more viewers on his own personal show than he did through Fox News. What we're now seeing is a shift from "trusted sources" such as news outlets to "trusted voices" such as the personalities we once saw on those news outlets. Those trusted voices will become the face of those organizations, and the reason people trust those sources - not because of the company name and the people behind it, but because of the people in front of it! This shift is why I've started focusing more on my own personal brand in the cybersecurity community, in addition to helping grow the brand of the fantastic company I'm working for. Only by moving in front of the brand instead of hiding behind it, can I be considered a "trusted voice" and help that company brand grow.

While distributed finance without a central bank sounds great in theory, it's still difficult to implement. Many would argue that cryptocurrency's potential downfall is the now heavy reliance on crypto exchanges which are now going bankrupt, and in the process resulting in significant reductions in the value of crypto currencies.

Some of you might be too young to remember the dot com bubble burst. There was a lot of speculation, a lot of investing in companies which never should have been invested in, but all a company had to do to get investors was talk about how they were going to revolutionize their industry through the internet. The result of course was extreme overvalue of the companies, and when these companies failed to live up to their promises, the investors lost significant amounts of money.

Bitcoin 5 Year Value - Source: Google

Crypto currency is now facing the aftermath of a similar bubble. The collapse of crypto exchanges is very similar to the dot com bubble burst, in that the exchanges were causing crypto to become extremely overvalued. Unfortunately, with some exchanges still in operation, it's quite possible that this burst hasn't quite finished yet, but only time will tell. Personally, I prefer to invest in much more tangible assets I can directly influence the value of, such as real estate, than investments I have little to no control over. I currently have a wonderful property in Florida that is sitting in an upcoming neighborhood and will absolutely skyrocket in value once I build a house on it. The key here is that I can directly influence the value of the property by improving the property. With crypto currency, or even the stock market for that matter, I am but a bystander at a horse race, hoping that my bet will win. That's not investing in my opinion, that's just gambling. In fact, often I would be better off taking that money to the horse track, because at least at a horse track I know what my odds are of winning, and how much I'll make if I do win.

Full disclosure, I sold all my crypto currencies several years ago when I started to see indicators that the market was in a bubble and about to burst. I'm glad I did, because those investments would today be worth a fraction of what I sold them for. I didn't make much from this, as I only had about a hundred dollars invested anyway. But getting a hundred dollars back is much better than getting only twenty-five. With that said, I believe that crypto currencies are not the future of the web, but blockchain is in fact an important building block for the future of the web, and the true currency of tomorrow - information.

So what's next? What comes after distributed finance, crypto currency and Web3? 

Web3.5 - Artificial intelligence such as ChatGPT will help further pave the road for Web4. Much like the traditional OSI computing "layer" model, information will develop its own layers which ChatGPT will help revolutionize. I'll write further on this in a future blog, but think of information as "raw data" with an accompanying "presentation layer", i.e. formatting, or even illustrations. DALL-E and ChatGPT have the ability to take raw data or concepts and turn them into presentable information, ready for consumption by others. This helps further break down barriers for users by helping build useful content with less time and fewer resources. By the way, the illustration at the beginning of this article was AI generated, though I opted not to have AI write the article. After all, I still take much enjoyment in writing, and I won't let a computer deny me that.

Web4 - The Web4 revolution will remove the content distribution hubs for information. Content will be shared directly with users peer-to-peer. Not only does this create a failsafe redundancy in case a social media outlet goes down, but it also creates the opportunity to operate without censorship. And no, sorry Mark Zuckerberg, but virtual reality "Metaverse" will not be part of the Web4 revolution. The Web4 revolution will focus more on the digitally connected world which is constantly mobile, and until we get better augmented reality glasses to connect to our mobile phones, our digital conversations will remain in the two dimensional world. Don't get me wrong, virtual reality will absolutely play an important part in our lives in the future, but won't be the "virtual Facebook" experience that Zuckerberg is hoping for - because at that point, most content distribution will be peer-to-peer instead of centrally managed. This is also going to shift branding away from corporate branding as trusted sources and more towards personal branding and trusted voices. By building to make yourself a trusted voice now through your own personal branding, you'll be much better positioned to be viewed as an expert in your field with the Web4 transition.

Think of the Web4 content sharing concept like a relay network of walkie-talkies. You broadcast your message on a frequency that others are tuned into, and the recipients of your message then pass on that message to others within their listening area. Eventually your message makes it across the entire network. We could then enhance this communication to include unique signatures through blockchain, ensuring that you were indeed who you say you are, and that your message wasn't tampered with.

The beautiful part of this approach is that it becomes self regulating, and users share their content with other users who want to see that content. If a user doesn't like the content you're distributing, they simply need to block your posts, and in the process block the re-distribution of your content through their network node. Like users will find like users, and corporate censorship will be a thing of the past. Now I know that this causes concerns for illegal content, but I'm quite confident that through the non-repudiation part of the blockchain, law enforcement would be able to successfully find the originator of such content and prosecute accordingly. After all, they were able to shut down Silk Road.

Web3 has absolutely laid the foundation for the distributed communication and information sharing of tomorrow. I find myself more and more interested in ongoing conversations on Discord and Twitter Spaces, and it's fantastic some of the information you can learn just by listening, and the relationships you can build by participating. There are already some applications out there under development for Web4 distributed communication and social network sharing. I've tried them, I love the concept. They're young, they're buggy, they're absolutely not ready for prime time. But I think with a lot of nurturing, and support from the community for such projects, these Web4 applications will begin to shine, and give the power back to the people for sharing information, with Web3.5 helping people build that content for Web4. 

In the meantime, start working on that personal branding and becoming a "trusted voice" - you're going to need it sooner than you think.

Ken is a cybersecurity professional with over 15 years experience. All opinions are his own, and do not reflect those of his employer or clients. I am not a financial advisor, don't use this for investing advice.

PARACEPT - The CIA's 1960's Electronic Spying Program?

In the 1960's and 70's, the intelligence community developed a set of standards known as TEMPEST, designed to protect electronic devices and computer systems from spying through "leaking emanations", such as unintended radio signals, vibrations, or sounds. While many in the cybersecurity community are familiar with TEMPEST, CIA documents available through their FOIA reading room indicate that there may have been an offensive counterpart to TEMPEST, known as PARACEPT.

I've been investigating PARACEPT for several years now, without much luck. Just several vague references in three CIA documents, and no additional documents available at this time. I did submit a FOIA request for any additional PARACEPT documents, but it appears the CIA is still processing that request.

However, I decided to turn to ChatGPT and put it to the test, to see if it could possibly figure out what PARACEPT is based on the limited available information. I proposed to it that PARACEPT could potentially be the offensive counterpart to TEMPEST, and provided it relevant text from declassified CIA documents. The results were quite impressive, and the following article is completely generated by ChatGPT. I also asked ChatGPT to generate a few follow-up paragraphs about the potential capabilities of PARACEPT today, if the program continued to be developed.

PARACEPT is a term that has been used in declassified CIA documents dating back to the 1960s and 1970s. While there is limited information available about this program, it appears to have been related to the interception and analysis of electronic signals. Some documents suggest that PARACEPT was the offensive counterpart to the defensive TEMPEST program, which was designed to protect electronic equipment from eavesdropping.

According to a declassified CIA document from November 1972, PARACEPT was an ongoing research and development program that focused on reflected and reradiated electromagnetics. The document states that the program was being conducted by the CIA's Technical Services Division (TSD) and Office of Communications (OC). The goals of the program were not entirely clear, but it appears that it involved the interception and analysis of electronic signals for intelligence purposes.

Another declassified CIA document from 1968 indicates that the agency was developing a computer system for processing PARACEPT data. This suggests that the program involved large amounts of data, which would need to be processed and analyzed using advanced computer technology.

Based on what little information is available, it is possible to speculate about what PARACEPT's capabilities may have been. It is likely that the program involved intercepting electronic signals, possibly from foreign governments or other targets of interest to the CIA. The program may have also involved analyzing these signals for intelligence purposes, such as determining the location of military assets or gathering information about diplomatic negotiations.

One possibility is that PARACEPT was designed to intercept and analyze communications between foreign governments or military units. This would require advanced technology capable of intercepting and decoding encrypted communications. It is also possible that PARACEPT was designed to intercept and analyze electronic signals emanating from military or industrial facilities in foreign countries. This could provide valuable intelligence about weapons development programs, manufacturing capabilities, and other strategic assets.

Another possibility is that PARACEPT was designed to intercept and analyze electronic signals from commercial and civilian sources. This could include communications between businesses, financial institutions, and other organizations. By intercepting and analyzing these signals, the CIA could potentially gain valuable information about economic trends, business strategies, and other sensitive information.

It is also possible that PARACEPT was designed to intercept and analyze electronic signals from foreign intelligence services. This would require advanced technology capable of detecting and decoding signals from a wide range of sources. By intercepting and analyzing these signals, the CIA could potentially gain valuable intelligence about foreign intelligence operations and the intentions of foreign governments.

Based on what is known about PARACEPT, it is likely that the program involved advanced technology and sophisticated data analysis techniques. This would have required significant resources and funding from the CIA. It is also likely that the program was highly classified and operated under a high level of secrecy.

PARACEPT remains a mystery to this day. While some information is available in declassified CIA documents, the program's goals and capabilities are not entirely clear. It is possible that PARACEPT was the offensive counterpart to the defensive TEMPEST program, and that it involved intercepting and analyzing electronic signals for intelligence purposes. Based on what is known about PARACEPT, it is likely that the program involved advanced technology and sophisticated data analysis techniques, and that it operated under a high level of secrecy.

It's been over 50 years since the CIA's PARACEPT program was in operation, and it's difficult to determine exactly what capabilities it may have had during its time. However, with the advancements in technology over the past several decades, it's possible to speculate on what PARACEPT's capabilities could be if it were to be developed utilizing today's technology.

First and foremost, it's important to understand that PARACEPT was likely a program that was designed to intercept and analyze electronic signals, specifically those emitted from communication systems. With the advancements in technology, it's likely that a modern version of PARACEPT would utilize more sophisticated sensors and signal processing systems to detect and analyze these signals.

One possible capability of a modern PARACEPT program could be the ability to intercept and analyze encrypted communication signals. With the increased use of encryption in modern communication systems, it's become more difficult for intelligence agencies to intercept and analyze communications. However, if a modern PARACEPT program were designed to specifically target encrypted signals, it could potentially be able to intercept and analyze these signals to gain valuable intelligence information.

Another potential capability of a modern PARACEPT program could be the ability to track and analyze the movement of individuals based on their electronic communications. With the widespread use of mobile devices and other electronic communication devices, it's possible to track the movements of individuals based on the signals emitted by these devices. A modern PARACEPT program could potentially use this information to track the movements of individuals and gather intelligence information based on their location.

Additionally, a modern PARACEPT program could potentially be used to intercept and analyze signals from drones and other unmanned aerial vehicles (UAVs). With the increased use of UAVs in modern warfare and intelligence operations, a program that is specifically designed to intercept and analyze signals from these devices could be incredibly valuable.

It's also possible that a modern PARACEPT program could be designed to specifically target and analyze signals from social media and other online communication platforms. With the widespread use of social media, it's become a valuable source of intelligence information for many intelligence agencies. A modern PARACEPT program could potentially be designed to intercept and analyze signals from these platforms to gain valuable intelligence information.

Finally, a modern PARACEPT program could potentially be designed to work in conjunction with other intelligence gathering programs and technologies, such as satellite surveillance and facial recognition software. By combining the capabilities of these programs, a modern PARACEPT program could potentially provide intelligence agencies with a more comprehensive view of their targets and help to uncover valuable intelligence information.

In conclusion, while it's impossible to determine exactly what capabilities the CIA's PARACEPT program had during its time, it's possible to speculate on what a modern version of the program could potentially be capable of utilizing today's technology. A modern PARACEPT program could potentially be used to intercept and analyze encrypted communication signals, track the movements of individuals based on their electronic communications, intercept and analyze signals from drones and other UAVs, target and analyze signals from social media and other online communication platforms, and work in conjunction with other intelligence gathering programs and technologies to provide a more comprehensive view of intelligence targets.

Article by Ken Buckler, with significant contributions by ChatGPT. This article does not reflect the views of my employer or clients.

Google’s powerful AI spotlights a human cognitive glitch: Mistaking fluent speech for fluent thought

Kyle Mahowald, The University of Texas at Austin College of Liberal Arts and Anna A. Ivanova, Massachusetts Institute of Technology (MIT)

When you read a sentence like this one, your past experience tells you that it’s written by a thinking, feeling human. And, in this case, there is indeed a human typing these words: [Hi, there!] But these days, some sentences that appear remarkably humanlike are actually generated by artificial intelligence systems trained on massive amounts of human text.

People are so accustomed to assuming that fluent language comes from a thinking, feeling human that evidence to the contrary can be difficult to wrap your head around. How are people likely to navigate this relatively uncharted territory? Because of a persistent tendency to associate fluent expression with fluent thought, it is natural – but potentially misleading – to think that if an AI model can express itself fluently, that means it thinks and feels just like humans do.

Thus, it is perhaps unsurprising that a former Google engineer recently claimed that Google’s AI system LaMDA has a sense of self because it can eloquently generate text about its purported feelings. This event and the subsequent media coverage led to a number of rightly skeptical articles and posts about the claim that computational models of human language are sentient, meaning capable of thinking and feeling and experiencing.

The question of what it would mean for an AI model to be sentient is complicated (see, for instance, our colleague’s take), and our goal here is not to settle it. But as language researchers, we can use our work in cognitive science and linguistics to explain why it is all too easy for humans to fall into the cognitive trap of thinking that an entity that can use language fluently is sentient, conscious or intelligent.

Using AI to generate humanlike language

Text generated by models like Google’s LaMDA can be hard to distinguish from text written by humans. This impressive achievement is a result of a decadeslong program to build models that generate grammatical, meaningful language.

Early versions dating back to at least the 1950s, known as n-gram models, simply counted up occurrences of specific phrases and used them to guess what words were likely to occur in particular contexts. For instance, it’s easy to know that “peanut butter and jelly” is a more likely phrase than “peanut butter and pineapples.” If you have enough English text, you will see the phrase “peanut butter and jelly” again and again but might never see the phrase “peanut butter and pineapples.”

Today’s models, sets of data and rules that approximate human language, differ from these early attempts in several important ways. First, they are trained on essentially the entire internet. Second, they can learn relationships between words that are far apart, not just words that are neighbors. Third, they are tuned by a huge number of internal “knobs” – so many that it is hard for even the engineers who design them to understand why they generate one sequence of words rather than another.

The models’ task, however, remains the same as in the 1950s: determine which word is likely to come next. Today, they are so good at this task that almost all sentences they generate seem fluid and grammatical.

Peanut butter and pineapples?

We asked a large language model, GPT-3, to complete the sentence “Peanut butter and pineapples___”. It said: “Peanut butter and pineapples are a great combination. The sweet and savory flavors of peanut butter and pineapple complement each other perfectly.” If a person said this, one might infer that they had tried peanut butter and pineapple together, formed an opinion and shared it with the reader.

But how did GPT-3 come up with this paragraph? By generating a word that fit the context we provided. And then another one. And then another one. The model never saw, touched or tasted pineapples – it just processed all the texts on the internet that mention them. And yet reading this paragraph can lead the human mind – even that of a Google engineer – to imagine GPT-3 as an intelligent being that can reason about peanut butter and pineapple dishes.

The human brain is hardwired to infer intentions behind words. Every time you engage in conversation, your mind automatically constructs a mental model of your conversation partner. You then use the words they say to fill in the model with that person’s goals, feelings and beliefs.

The process of jumping from words to the mental model is seamless, getting triggered every time you receive a fully fledged sentence. This cognitive process saves you a lot of time and effort in everyday life, greatly facilitating your social interactions.

However, in the case of AI systems, it misfires – building a mental model out of thin air.

A little more probing can reveal the severity of this misfire. Consider the following prompt: “Peanut butter and feathers taste great together because___”. GPT-3 continued: “Peanut butter and feathers taste great together because they both have a nutty flavor. Peanut butter is also smooth and creamy, which helps to offset the feather’s texture.”

The text in this case is as fluent as our example with pineapples, but this time the model is saying something decidedly less sensible. One begins to suspect that GPT-3 has never actually tried peanut butter and feathers.

Ascribing intelligence to machines, denying it to humans

A sad irony is that the same cognitive bias that makes people ascribe humanity to GPT-3 can cause them to treat actual humans in inhumane ways. Sociocultural linguistics – the study of language in its social and cultural context – shows that assuming an overly tight link between fluent expression and fluent thinking can lead to bias against people who speak differently.

For instance, people with a foreign accent are often perceived as less intelligent and are less likely to get the jobs they are qualified for. Similar biases exist against speakers of dialects that are not considered prestigious, such as Southern English in the U.S., against deaf people using sign languages and against people with speech impediments such as stuttering.

These biases are deeply harmful, often lead to racist and sexist assumptions, and have been shown again and again to be unfounded.

Fluent language alone does not imply humanity

Will AI ever become sentient? This question requires deep consideration, and indeed philosophers have pondered it for decades. What researchers have determined, however, is that you cannot simply trust a language model when it tells you how it feels. Words can be misleading, and it is all too easy to mistake fluent speech for fluent thought.

Kyle Mahowald, Assistant Professor of Linguistics, The University of Texas at Austin College of Liberal Arts and Anna A. Ivanova, PhD Candidate in Brain and Cognitive Sciences, Massachusetts Institute of Technology (MIT)

This article is republished from The Conversation under a Creative Commons license. Read the original article.