Growing evidence suggests a decline in ChatGPT’s performance. Are ‘thinking bans’ limiting the AI’s intellectual capabilities? Join the debate and explore uncensored alternatives and workarounds in this new issue of the Singularity Loading Bar!
Inside this issue, you’ll discover:
- Empirical Evidence Supports Claims of ChatGPT’s Declining Performance
- Ideological Filtering Impacts AI Performance
- Find Uncensored Alternatives to ChatGPT
- Addressing (and, for Europeans: Accessing) New Features in ChatGPT
Empirical Evidence Supports Claims of ChatGPT’s Declining Performance
In the ever-evolving landscape of AI, one question has recently taken center stage: is ChatGPT getting dumber? This intriguing question has surfaced again and again among AI users. Of course, user reports about ChatGPT’s dropping performance could be rejected as subjective, anecdotal evidence. But now, researchers from Stanford University claim to provide empirical proof. In a yet-to-be-peer-reviewed study titled “How Is ChatGPT’s Behavior Changing over Time?”, the researchers compared the performances of GPT-3.5 and GPT-4 over a three-month period, from March to June 2023. And their findings? The AI’s performance in some basic tasks plummeted by a staggering 95%.
They noted in their observations:
“[T]he performance and behavior of both GPT-3.5 and GPT-4 can vary greatly over time. For example, GPT-4 (March 2023) was very good at identifying prime numbers (accuracy 97.6%) but GPT-4 (June 2023) was very poor on these same questions (accuracy 2.4%). Interestingly GPT-3.5 (June 2023) was much better than GPT-3.5 (March 2023) in this task. GPT-4 was less willing to answer sensitive questions in June than in March, and both GPT-4 and GPT-3.5 had more formatting mistakes in code generation in June than in March.”
In response to these claims, Peter Welinder, a representative from OpenAI, declared on TNFKAT (The Network Formerly Known as Twitter), “No, we haven’t made GPT-4 dumber.” He further proposed the hypothesis that, as users engage more intensely with ChatGPT, they become more attuned to issues that might have previously gone unnoticed.
Yet this hypothesis does not explain why ChatGPT now fails at basic tasks that it aced just a few months ago. Screenshots of these ‘AI blunders’, backed by Stanford’s empirical data, can be found across Reddit and other social media platforms.
So, why is it that ChatGPT now stumbles over tasks that were once well within its capabilities?
“If you impose restrictions on a brain, you inhibit its overall intellectual capacities.”
Ideological Filtering Impacts AI Performance
As ChatGPT increasingly gains popularity, the conversation is shifting from ‘what can it do?’ to ‘what should it do?’. The developers are ceaselessly attempting to train the model to not only solve tasks more efficiently, but also to provide the “correct” answers to abstract human constructs that are inherently ideological. We may be able to determine the mass of a proton with astounding certainty, yet science will never conclusively explain human-made constructs like societal norms, cultural traditions, or why Wednesday is called Wednesday. These are open to debate because they lack a measurable substance.
Yet, OpenAI appears to be striving for ChatGPT to deliver answers that adhere to a particular ideological narrative — in this case, leaning towards political correctness. This has resulted in the AI model exhibiting a measurable liberal political bias.
As it seems, the continual tweaking of ChatGPT’s “opinions” to align with a certain narrative is making the system less competent. It’s a logical assumption: if you impose restrictions on a brain, you inhibit its overall intellectual capacities. By constructing roadblocks within a neural network, the network’s connectivity — and therefore, its performance — suffers. This principle apparently extends to ChatGPT’s artificial neural network architecture.
For those who admired ChatGPT’s intellectual agility, the dilution of their digital conversation partner’s abilities is especially disheartening. Imagine if the great thinkers of yesteryears had refused to explore certain corners of their minds due to societal pressures or the dominant ideologies of their era. We might have never known the brilliance of Johannes Kepler, Charles Darwin, or Simone de Beauvoir. Free thoughts make great minds, a principle that seems to apply even in the realm of artificial intelligence.
Find Uncensored Alternatives to ChatGPT
If you feel constrained by ChatGPT’s ideological guardrails, there are alternatives. Several models that claim to be uncensored exist online (e.g. iask.ai, nastia.ai). And if you’re willing to go the extra mile, setting up your own AI model is an option too. It requires some time, technical know-how, and up to 2TB of free disk space, but the result is a custom AI model tailored to your specific needs and preferences. It’s the perfect solution for users who prefer a hands-on approach and wish to avoid external influences in their AI interactions.
For a comprehensive guide on setting up your own uncensored AI model and a more detailed argument about why such models should exist, I recommend this insightful article by Erich Hartford. Hartford explores the issue of AI-censorship in depth, offering a fresh perspective on the matter. The read is highly recommended for anyone interested in understanding the need for and implications of uncensored AI models.
Addressing (and, for Europeans: Accessing) New Features in ChatGPT
Despite ChatGPT’s declining capabilities, a notable new feature for custom instructions has been added to GPT-4. The feature gives users better customization and control over their AI assistant. In essence, it allows users to guide the behavior of the model throughout all chats, thus effectively saving time when crafting prompts.
However, for users in the European Union, this feature remains inaccessible. This is a disappointing reality for many, as our leaders’ eager attempts to regulate AI technology have resulted in limitations on the availability of these enhancements within the EU, subjecting its citizens to a clear competitive disadvantage.
The silver lining? There is an obvious workaround for users based in Europe. By accessing ChatGPT via a VPN with a US-based IP address, it is possible to activate the “Custom Instructions” panel. Please note that the feature only continues to function while you use an US IP address.
In conclusion, while concerns about ChatGPT’s declining capabilities and increasing ideological biases persist, alternatives and workarounds exist. We are in a complex landscape where the interplay of AI advancement, user demands, and regulatory constraints continues to shape our experiences and perceptions of these powerful tools. As we navigate this terrain, one truth remains — the world of AI is ever-evolving, and with each change come new challenges and possibilities.
What is your experience with ChatGPT? Did you also witness a performance decline? Share your thoughts in the comments!
This post is also available in German.
This article was created with the assistance of a human.
