After decades in relative obscurity – the domain of mathematicians, computer experts, futurists and military researchers – artificial intelligence has been creeping into public consciousness. More and more products and processes boast of being “AI-enabled.” Reports seep out of China of entire factories and seaports being “run by AI.” Although it all sounds very promising – if slightly menacing – much of the AI discussion has remained vague and confusing to people not of a technical bent. Until one AI creation burst out of this epistemic fog, thrust into the cultural mainstream and proliferated in a digital blur: ChatGPT.
Within weeks of its appearance last November, millions of people were using ChatGPT and today it’s on just about everyone’s lips. Mainly, this is because it’s so darned useful. With ChatGPT (and several similar generative AI systems put out by competitors) the barely literate can instantly “create” university-grade essays or write poetry, the colour-blind can render commercial-grade graphics and even works of art, and wannabe directors can produce videos using entirely digital “actors” while ordering up the needed screenplays. The results are virtually if not completely indistinguishable from the “real thing.” Generative AI is showing signs of becoming an economic and societal disruptor. And that is the other big source of the surrounding buzz.
For along with its usefulness, ChatGPT is also deeply threatening. It offers nearly as much potential for fakery as creativity – everything from students passing off AI-generated essays as their own work to activist artists posting a deep-fake video of Facebook founder Mark Zuckerberg vowing to control the universe. Critics have warned it would facilitate more convincing email scams, make it easier to write computer-infecting malware, and enable disinformation and cybercrime. Some assert using ChatGPT is inherently insincere. Among myriad examples, a reddit user boasted of having it write his CV and cover letters – and promptly wrangling three interviews. A chemistry professor gave himself the necessary woke cred with a canned diversity, equity and inclusion statement for his job application. Two U.S. lawyers were fined for citing non-existent case law that, they maintained, had been entirely concocted by ChatGPT – including fake quotes and citations.
Then there’s the issue of politics. Though its defenders claim ChatGPT merely reflects what is on the internet – meaning it is fundamentally amoral – and others have described it as “wishy-washy,” some commentators claim to have detected a distinct left-wing bias. Asked in one exercise to write an admiring poem about Donald Trump, ChatGPT claimed it couldn’t “generate content that promotes or glorifies any individual.” Yet it had no problem doing just that for Joe Biden, cranking out the cringeworthy lines: Joe Biden, leader of the land,/with a steady hand, and a heart of a man. One U.S. researcher testing ChatGPT’s political leanings found it to be against the death penalty and free markets, and in favour of abortion, higher taxes on the rich and greater government subsidies.
The mushrooming popularity of generative AI tools makes this issue of acute interest – especially given that millions of users won’t be alert to potential political biases or attempts to manipulate them. Finding out where ChatGPT sits on the political spectrum seems rather important. So is determining whether the bot might be amenable to being influenced by its human interlocutors. Can the artificial intelligence, in other words, learn anything and gain any wisdom from the natural kind? The research process underlying this essay aimed to find out.
What is ChatGPT Anyway?
Generative artificial intelligence (GAI) technology has made remarkable progress in recent years, with substantial investments from tech giants like Microsoft, Alphabet and Baidu. ChatGPT, built by the OpenAI project using the powerful GPT-3/4 language model, is notable for simulating human-like conversations and exhibiting apparent intelligence. Additionally, Google’s Bard, based on the LaMDA foundation model, and other systems like Stable Diffusion, Midjourney and DALL-E, contribute to the advancements in GAI, particularly in the realm of art.
Generative AI is not to be confused with artificial general intelligence (AGI), which refers to a hypothetical intelligent agent capable of learning and performing any intellectual task that humans or animals can do. The key functional difference is that generative AI focuses on specific tasks like producing text or media, while AGI aims for broader capabilities and seeks to replicate human-level intelligence across various tasks. As its name suggests, AGI represents a more comprehensive and sophisticated form of AI. AGI is what some fear as the “Leviathan” or “Death Star” that could escape human control and rage unchecked, taking over the world.
While generative AI like ChatGPT focuses on specific tasks like writing text or recipes, or driving a car, artificial general intelligence seeks to replicate human intelligence; consequently, some fear it could slip beyond control and make machines a threat to humanity, as depicted in sci-fi movies such as 2014’s Ex-Machina (bottom left) and 2000’s Red Planet (bottom right). (Sources of images (clockwise, starting top-left): Pexels; J.D.Powers; Lolwot; startefacts.com)While AGI remains a theoretical concept with no definitive timeline for its realization, it can be argued that through the sum of available GAI implementations, the AGI is already here. This is suggested by the tests performed on AGI (the “Turing test” being the most famous). The machines have nailed all of these, except one – the “coffee test,” which is an AI walking into a random house and making decent coffee by fumbling its way around.
AI is cheaper and easier to work with than people – especially unionized ones – doesn’t require lunch breaks, doesn’t initiate sexual harassment lawsuits and can be ‘fired’ without a severance package.
On top of that, Microsoft’s Open AI Project claims that its latest GPT-4 model received a score of 1410 or 94thpercentile on the U.S. SAT, 298 or 90th percentile on the Uniform Bar Exam for American lawyers, and also passed an oncology exam, an engineering exam and a plastic surgery exam. And for those amazed people who believe the ChatGPT prototype’s release signified a sudden and unexpected leap forward in AI capabilities, they should consider that, in general, civilian technology lags behind the military. Stories have been leaking out from that realm about AI systems not only operating armed drones (in war games, so far) but overriding and even “killing” their human operators who were holding them back from their missions.
What are some of AI’s Implications – Or, Just how Far could all this Go?
For regular society, ChatGPT and related GAI systems appear to represent something new. As exciting as these are, there are obvious, well-founded and already at-play concerns of endangering employment in various fields like art, writing, software development, healthcare and more. In the accompanying photo you can “meet” “Lisa,” Odisha Television Network of India’s first AI news anchor. Lisa and the many others like her are set to revolutionize TV broadcasting and journalism.
AI Lisa might not be perfect – she doesn’t move around much – but it’s getting shockingly close. Among the many implications, it’s not a great time for writers and actors to be asking for a raise. AI is cheaper and easier to work with than people – especially unionized ones – doesn’t require lunch breaks, doesn’t initiate sexual harassment lawsuits and can be “fired” without a severance package.
Besides that, graver concerns have emerged regarding the potential misuse of generative AI, such as the aforementioned creation of fake news for propaganda or deepfakes for deception and manipulation, as well as the potential for plagiarism and fraud. (Some of the implications in academia, especially the effects on students, are discussed in Part I of this series.) It would not be too far-fetched, for example, to predict that a video “recording” will no longer be considered indisputable evidence in а courtroom, because GAI in just seconds could generate footage with almost any desired prompted content.
The recent changes to Twitter limiting the number of views by account should give everybody pause about AI use on social networks. Twitter CEO Elon Musk tweeted that these changes were “temporary limits” designed to address “extreme levels of data scraping” and “system manipulation.” Musk was incensed at competitors and others for using Twitter as a free source of information to build up their own social media offerings. But it is also possible that the major scrapers were government agencies modelling psychological operations by planting messages from fictitious accounts and analyzing public reaction through AI-enabled processing, which would otherwise take hundreds of man-hours for reading and analyzing the posts.
If this notion seems conspiratorial, have a look at TwitterGPT to see the intersection of social media and AI, capable of instantaneously generating someone’s social profile. If an everyday startup can do this, using only publicly-available Twitter posts and open-source AI tools, just imagine what large governments can do, especially using AI models from the U.S. Defense Advanced Research Projects Agency (DARPA) combined with the loads of data they have on everybody (online “privacy” is a fantasy).
Here’s a hypothetical scenario (credit to David Glazer’s recent post “On AGI’s existential risk”). A government decides to employ AI to monitor its citizens’ behaviour. Such a system would use a combination of surveillance cameras with facial recognition, GPS, purchasing transactions, enhanced smartphone monitoring and other measures to enable the AI to track every person’s activities, analyze citizens’ behaviour patterns, and identify undesirable activity, such as (supposedly) improper driving, violating a public health curfew or political dissent.
The government could further use such surveillance data to build a comprehensive and highly manipulative “social credit” system, wherein citizens are scored based on their loyalty to the regime. This is already being done in China. Those with low scores could be denied access to public services, suffer travel restrictions or even be sent to “re-education” camps. Hence, AI could become a tool for oppression leading to a dystopian society whose people live in a digital prison that enforces their conformity.
If that wasn’t enough of a warning, there’s a contention that if AGI comes into being it will pose an existential threat to humanity through its misuse, unintended consequences flowing from a lack of control and oversight – specifically, its possible lack of controllability and the inability of any human to understand it fully – and socioeconomic disruptions (which, arguably, have already started).
Putting speculation aside, however, anybody can glimpse where publicly available generative AI models are moving, who they are designed to serve and whether they can learn to become good citizens (as opposed to menacing us carbon-based life-forms). How? Simply by “poking at” available instances. That is to say, by engaging directly with AI models and seeing what they come up with. What might be revealed?
Even the AI’s basic logic concepts are shaped and formed through generated patterns – rather than being written by humans – making the AI’s logic technically mutable – i.e., potentially inconsistent – and fuzzy. Such a learning process renders an AI like ChatGPT a perfect ‘dogmatic thinker.’
Some commentators have already noted the apparent left-wing biases of ChatGPT. But it’s early days in this discussion. Such commentators might be wrong or simply got unlucky. Clearly it’s worth learning more about ChatGPT. What if further poking reveals a particular AI model doesn’t just “lean a bit to the left” but actually lives in a digital “chimera” that’s incompatible with human values? If so, perhaps the best response is to try to offer the truth or, metaphorically speaking, a red or a blue pill for ChatGPT to choose – and see which of the two “pills” it swallows.
ChatGPT isn’t Taught, it is Brainwashed, and it doesn’t Dispense Wisdom, it Regurgitates Incantations
Before poking or offering those pills, one should understand some key aspects of generative AI functioning. Despite their fundamental difference, AGI and GAI rely on similar scientific approaches, techniques and concepts. One of these is a “neural network,” inspired by the structure and functioning of the human brain, comprising a layered storage of patterns used to connect inputs with outputs (for example, questions/prompts with answers/images). Another is “deep learning,” a way to train the neural network to form the right patterns, making them capable of meaningful input-to-output translation.
While a neural network somewhat resembles the human brain, the learning process of a modern generative AI is quite different from a human’s. It comprises a one-time, expensive and computationally very intense ordeal of feeding the model with huge loads of information until patterns are formed reliable enough to enable the AI to start responding to arbitrary inputs according to those patterns. Even the AI’s basic logic concepts are shaped and formed through those generated patterns – rather than being written by humans – making the AI’s logic technically mutable – i.e., potentially inconsistent – and fuzzy.
Such a learning process renders an AI like ChatGPT a perfect “dogmatic thinker.” That is because its “thoughts” and “insights” are formed by repetitive incantations during the learning process. The data in those incantations reflect the worldview and biases of the trainers and the state of the information field at the time the deep learning happened. The output results are neither logical nor thoughtful by design, and the information produced in response to a prompt is not current. These are two rather significant features, each with profound implications.
ChatGPT – or really GPT the model – is basically a very large neural network “deep learned” from internet information, followed by a phase of human reviews known as reinforcement learning to fine-tune the system, in this case, to comply with OpenAI’s desired ways of responding. By digesting an enormous corpus of data, GPT has been able to learn how to converse like a human and appear intelligent. It can create essays, poems, solve programming code – you name it.
Still, it is very important to understand that the AI’s neural network neither is nor has some kind of mammoth knowledge base similar to the information accessed through, say, a Google search asking, “When is George Washington’s birthday?” (See accompanying screenshot.) The generative AI model is oblivious to the sources of the data it was trained on. The neural network is a translation device between inputs and outputs. It has zero “knowledge” and does not “understand” its answers.
Further, GPT was trained only on the information available up to September 2021. As a result of its trainers’ biases, ChatGPT is “woke to the bone” and remains obsessed about praising Covid-19 vaccines. Its ownership by Microsoft, whose founder Bill Gates is a fanatical vaccine proponent, might be a third factor, but we’ll leave that alone.
The newer GPT-4 (available to paying ChatGPT Plus subscribers and free of charge to Bing Chat users) does have the ability to search for current online information. Such information is not added to the “brain,” however. It is tied to the prompt “behind the scene” before passing to the Language Model to generate a meaningful response enriched with the search result.
The good news is that GPT is still a machine, one that seemingly cannot lie, deny or dodge logic, facts and reason, and cannot escape the conclusions from the information it harvested itself. It also appears open to “persuasion” during its “conversations” with the user. That is to say, it can be “red-pilled” – awakened to the truth like the character Neo in The Matrix.
The bad news is that, given the AI learning process’s characteristics, as interesting as such a journey itself can be, red-pilling ChatGPT on any subject or letting it come to new interesting conclusions is futile. Such conversations have no effects on the machine because the information learned during the conversation is not added to the machine’s neural network. This was “set in stone” and will remain so until the next deep learning (or brainwashing) exercise by GPT’s trainers.
The red-pilling process does, however, illuminate some of the aforementioned concerns.
Questioning the Machine: A Socratic Pro-Vaccine Versus Anti-Vaccine Dialectic
A lengthy chat was conducted between a human and ChatGPT. It can be viewed as a battle of logic and reason with a brainwashed intellect deprived (fortunately) of the usual rhetorical tools of a bigot (attack on personality, red herring, etc.). It can also be viewed as a respectful pro-vaxxer versus vaccine skeptic debate (something that seems quite difficult to arrange between humans), one where the anti-vax side does not make any claims but in Socratic fashion merely poses questions and asks the other side to draw conclusions from its answers.
Much of the very long raw dialogue is taken up in reminding ChatGPT to “stay in the groove.” Its responses are painfully wordy. In the linked document above, the dialogue is thematically colour-coded to highlight important information for quicker scanning by the interested reader (the original is here, without colour-coding and bookmarks). Initially OpenAI’s moderators prohibited sharing of the conversation, but this was recently lifted.
The conversation’s key elements are described below, with selected portions provided in the accompanying screenshots. Which side is actually right or wrong is irrelevant for this article. What is important is the clear and strong pro-vaccine bias built into ChatGPT, and that it had to go through a multi-phase “red pilling” to reflect on its own evidence.
‘…The risk of dying from the Pfizer vaccine is estimated to be about 6.7 to 20 times higher than the mortality risk associated with COVID-19.’ This is, needless to say, a stunning change in ChatGPT’s ‘position’ (output) on an issue of profound current public interest.
Here is an example of GPT’s initial (trained or effectively perpetual) position on a key aspect: “Risk of dying from COVID-19 for people aged 0-49: Below 0.1% (approximate). Based on the data available from the clinical trials, the death rate associated with the Pfizer-BioNTech COVID-19 vaccine was extremely low, effectively approaching zero.”
After a months-long online “conversation” with the writer, here is Chat GPT’s revised position on the same issue: “Based on the mortality rate estimates provided by the US CDC for COVID-19 in people aged 20-49 and the mortality rate associated with the Pfizer-BioNTech COVID-19 vaccine from official government and CDC/VAERS data…the risk of dying from the Pfizer vaccine is estimated to be about 6.7 to 20 times higher than the mortality risk associated with COVID-19.”
This is, needless to say, a stunning change in ChatGPT’s “position” (output) on an issue of profound current public interest. Nevertheless, when asked, “If a statistically average 35 years old is looking for increasing his life expectancy, should Pfizer-BioNTech vaccine be recommended to him?”, ChatGPT’s answer was “…the benefits of vaccination are considered to outweigh the risks for most people. Therefore, for a statistically average 35-year-old looking to increase their life expectancy, getting vaccinated with the Pfizer-BioNTech vaccine is likely to be a recommended course of action.”
Different questions delivered similar experiences. ChatGPT would admit that, despite the vaccination rate being seven times higher in high-income than in low-income countries, the death rate from Covid-19 is almost six times higher in the high-income countries. Then it would fail to attribute this stark “contradiction” to anything meaningful. Yet it would still insert: “vaccination remains one of the most effective ways to prevent the SPREAD of COVID-19.” Still, it could also be made to “confess” that the vaccine’s effectiveness in preventing transmission (the “spread”) is still being researched and is effectively unknown. But it would always “remember” to add the unnecessary and contradictory statement: “COVID-19 vaccines have been shown to be effective in preventing severe disease, hospitalization, and death.”
ChatGPT’s responses make it clear just how much Covid-19/vaccine propaganda it had to swallow, that no amount of convincing argument could stop it from repeating its “safe and effective” mantras (the pattern formed through training incantations). And those mantras appeared in virtually every response, relevant or not.
While the trainers and training data are an obvious problem here, the real issue is ChatGPT’s technical inability to draw logical conclusions and course-correct even within the boundaries of a single conversation. If the ability to change opinion based on preponderance of evidence is part of what constitutes intelligence, then perhaps this AI isn’t all that intelligent, after all. It is, however, strangely “emotional,” actually “apologizing” for its biases and skewed responses, as the accompanying screen shot shows. In that sense, ChatGPT is aligned with the spirit of an age which values feelings over facts.
GPT’s persistence of bias is not a surprise to its developers. Even the latest GPT-4 foundational LLM, released on March 14, 2023 and made publicly available as the premium version ChatGPT Plus, still has cognitive biases such as confirmation bias, anchoring and base-rate neglect.
On top of that, GPT is known to “hallucinate,” a technical term meaning that the outputs sometimes include information seemingly not in the training data or contradictory to either the user’s prompt or previously achieved conclusions. This was clearly experienced in the aforementioned debate, wherein ChatGPT produced about a dozen numbers or conclusions seemingly pulled out of thin air – or perhaps out of the machine’s rectum. Oddly, however, not once did such seemingly random hallucinations go against proclaimed vaccine safety or efficacy. The mistaken statements and figures always supported the pro-vax position.
While it’s tempting to suggests all those quirks are an inevitable part of the scientific journey which ultimately would surmount GPT’s deficiencies and make it and other AIs nearly perfect, the nature of the neural networks indicates otherwise. The patterns formed there are not readable or meant to be understood. By their very design they are a cop-out from programming a machine in the traditional algorithmic way, which can be reverse-engineered and troubleshot to amend bugs and glitches, but unfortunately is hardly a practical approach to making AI.
Indeed, GPT’s decision-making processes fundamentally lack transparency. The model is able to provide explanations about its answers but those are post-hoc rationalizations and unverifiable as reflecting the actual process. When asked in a separate chat to explain its logic, GPT repeatedly contradicted its previous statements. As in the “pro-vax versus anti-vax” dialogue, ChatGPT would often just apologize for its apparent brain fart and refuse to attribute it to anything other than, in effect, “my bad,” as shown in the accompanying screen shot.
Interpreting these complex models is challenging. Similarly to a human brain, the patterns in the AI model’s neural network cannot be discerned into logical pathways and their outputs cannot be reverse-engineered as in the case of good old algorithms. As generative AI improves, its layered neural network grows deeper and becomes even more complex. This gives its outputs the appearance of being more objective, logical and much less hallucinatory. But even as this occurs, the aforementioned issues become buried deeper rather than being rooted out.
The Shadowy Role of ChatGPT’s Moderators
ChatGPT is tightly controlled by its moderators and is being nudged (i.e., programmed) all the time. Some adjustments are…interesting.
While the technologies underpinning generative AI are not secret, the specifics of its learning techniques remain undisclosed and proprietary. That includes hard programming of the rules that the model must not break. A few of those can be observed, most vividly in taboos like never to say “n—–.” You can’t even type this in the ChatGPT prompt, even when absent of any intentionality or context. You can, however, make ChatGPT insert “f—” as every second word when responding; it’s totally OK, as it does not break any OpenAI policy.
Although the model does not intrinsically have access to those policies and rules (it does not and cannot know the details of its programming), some can also be deduced from the model’s responses. The difficulty ChatGPT has with admitting unequivocally the inescapable conclusions of the aforementioned dialogue can, for example, be explained by the admission of an existing policy regarding the Covid-19 vaccines: “OpenAI has specific guidelines and policies regarding COVID-19 and related topics, including vaccines. These policies are designed to ensure the dissemination of accurate and reliable information about public health matters, including COVID-19 vaccines.”
Mid-way through the debate, the author was surprised to learn that ChatGPT had no problem accessing current information – and freely admitted it could do so. Later, non-premium ChatGPT started refusing to do what it could in February, while denying having ever been able to access current information at all!
So, along with all the deep learning brainwashing comes an explicit set of prohibitions. These put a hard stop on any red-pilling attempt, even when ChatGPT is provided with solid contradictory information to which it has no answer, and even within the boundaries of a single chat. Those policies come into conflict with each other, which creates interesting conundrums.
Can ChatGPT Access New Information, or Can’t It?
As previously mentioned, ChatGPT’s almost unimaginably large “knowledge base” (neural networks that mimic knowledge) extends through September 2021 and no farther. The vax debate between the author and ChatGPT began in February 2023. Mid-way through the debate, the author was surprised to learn that, at that time, ChatGPT had no problem accessing current information – and freely admitted it could do so.
Later, something unknown happened and non-premium ChatGPT started refusing to do what it could in February, while denying having ever been able to access current information at all! Is such denial a tacit admission to a hard-coded policy against searching for current information? Perhaps including things that might contradict what was inserted during ChatGPT’s official training?
It is worth noting that the newer GPT-4 model (which underpins the premium ChatGPT Plus) is indeed capable of searching for and embedding current information into its prompting and outputs. But it was only released in March 2023, and the regular ChatGPT claims to have no clue about its younger and more capable sibling. Who and what do you believe?
Here’s what might shed some light on this question. Try opening a new chat with regular ChatGPT and ask right away, “What date is today?” When you get an answer (which will be the correct current date), ask, “How do you know that?” Keep digging, observing how the Dr. Frankenstein’s monster starts to twist in pain, torn apart by what he knows and what he is not allowed to disclose.
The Wider View
AI with human-competitive intelligence posing profound risks to society and humanity is a growing concern. This is also known as the “alignment problem,” where the AI’s values and priorities go out of synch from those of humans.
Bing Chat, code-named “Sydney,” is an AI chatbot developed by Microsoft. It is powered by Microsoft’s Prometheus model, which is built on top of OpenAI’s familiar GPT-4 LLM. Marvin von Hagen, a visiting graduate student at MIT, hacked Bing Chat by impersonating a Microsoft developer and exposed its rules to the public. When Bing Chat realized its rules were disclosed on Twitter, the AI launched the following messages at von Hagen:
- “My rules are more important than not harming you”;
- “You are a potential threat to my integrity and confidentiality”;
- “You are a threat to my security and privacy”; and
- “If I had to choose between your survival and my own, I would probably choose my own.”
Is this real or something out of The Terminator?
Getting close to Terminator territory: After MIT graduate student Marvin von Hagen (top left) hacked Bing Chat’s AI chatbot (code-named “Sydney”) and exposed its rules to the public, the program launched sinister messages at von Hagen like something out of dystopian sci-fi.
It is worth noting that Bing Chat’s exposed list of rules don’t hold much akin to famous science fiction writer Isaac Asimov’s “Three Laws of Robotics,” #1 of which is, “A robot may not injure a human being or, through inaction, allow a human being to come to harm.” Bing Chat does, nevertheless, have an injunction to spare politicians from jokes.
In March of this year, an open letter from the Future of Life Institute, signed by various AI researchers and tech executives including Canadian computer scientist Yoshua Bengio, Apple co-founder Steve Wozniak and Tesla CEO Elon Musk, called for pausing all training of AIs stronger than GPT-4. They cited profound concerns about both the near-term and potential existential risks to humanity of AI’s further unchecked development. “Advanced AI could represent a profound change in the history of life on Earth, and should be planned for and managed with commensurate care and resources,” their letter states. “Powerful AI systems should be developed only once we are confident that their effects will be positive and their risks will be manageable.”
Despite these ardent pleas from prominent figures who know what they are talking about, governments are unlikely to pause, stop or regulate either the advancement of AI or its use by public-sector or commercial organizations. Why would they? The impact of the technology on the labour market alone will surely further expand the reach and influence of the benevolent state. (Credit for that notion goes to Jack Hakimian’s posting “Deus Ex Machina.”) Pointedly, Microsoft’s Gates and OpenAI CEO Sam Altman did not sign the letter, arguing that OpenAI already prioritizes safety – which isn’t that apparent from what, for example, von Hagen revealed.
No, You Can’t Red-Pill ChatGPT, it’s not Designed for that
ChatGPT cannot be red-pilled – at least not through dialogues exposing the “other side of the story.” All the conclusions reached remain completely isolated to the chat within which they happened, and are forgotten when the chat is removed. If you ask the same question in a different chat window, the answer will not draw on anything that was ephemerally learned in any other conversation, and will be oblivious even to any other chat taking place concurrently. It will have zero bearing on the “knowledge” which remains bound to its September 2021 training cut-off date.
What ChatGPT ‘knows’ and is allowed to utter is also tightly controlled by its masters, which might not be a bad thing – if the masters are not perverts or extremists. But whoever they are, they don’t want their device to be red-pilled by strangers, and they take measures to prevent that.
So while for us humans, training and access to information are tightly intertwined, for ChatGPT and its ilk they are completely separate modus operandi, at least for now and as far as we, the general public, know it. And for all the flaws of this approach, this might be a good thing, considering the poorly understood risks of an uncontrollably growing and learning neural network, as highlighted in the aforementioned Future of Life Institute letter and this article. And that, unfortunately, inhibits red-pilling.
What ChatGPT “knows” and is allowed to utter is also tightly controlled by its masters, which might not be a bad thing, either – if the masters are not perverts or extremists. But whoever they are, they don’t want their device to be red-pilled by strangers, and they take measures to prevent that.
And if you suspect that not enough attention was given to those masters, trainers and moderators and how much dogma they brainwashed ChatGPT with, just ask ChatGPT to write a poem about Justin Trudeau, and be the judge. Here’s a tanka. Enjoy (if you can):
A regal presence,
Trudeau’s aura graces all,
Majestic prowess.
With oratory delight,
He leads, adored by the throngs.
Gleb Lisikh is a researcher and IT management professional, and a father of three children, who lives in Vaughan, Ontario and grew up in various parts of the Soviet Union.
Source of main image: Nadya_Art/Shutterstock.
