Nah, those clues for his understanding only were long long ago. That was literally just CGPT's unedited attempt to transpose what English Daily Mail (GWB quotes) and Express readers (Treefrog's fave) are into their US equivalent. That it choose NC, as Amercias' "Little Britain", I didn't prompt for that....but as I said, GWB likes to tangle with me, so it's seen quite a few of his posts and he mentions NC a lot. Also NC is actually a great parallel for "Little Britain". Its a little worrying just how much persists in account memory. I must ask it what it knows about me...
Too much... If Stephen Miller sees that he will deport me to South Sudan. I'll try to step back as it's right, genocide and war is a third rail for me. I cannot casually banter on this due to my life experience. Still it's good to know there are still a few good men, though they are all getting fired. This guy for posts on his personal account.
The exact training data is not published. So, that is made up bullshit. Just stop being a case study for the Dunning-Kruger effect. If it isn't bullshit prove it by posting the following: - the precise list of data that was used to train ChatGPT - the ChatGPT weights You know nothing whatsoever about LLMs or machine learning. Nothing. Totally ignorant.
The exact opposite is true. It is virtually impossible to have an LLM that is neutral. If you were actually a computer scientist, you would 100% agree with that statement. Multiple peer-reviewed studies and independent analyses have demonstrated and quantified political bias in ChatGPT, with most finding a consistent left-leaning or left-libertarian orientation in its responses. Here is a summary of key studies and their findings: Key Points on Political Bias: Wikipedia: Research shows systematic left-leaning bias, especially in articles about conservative figures, which can influence LLM outputs34. News Media & Academia: Both tend to reflect the prevailing political leanings of their contributors, which are often left-of-center in the U.S. and other Western countries32. Common Crawl & Social Media: These sources are vast and diverse but include large volumes of unfiltered, opinionated, and sometimes extreme political content, amplifying whatever biases are prevalent online12. Books, Code, Multilingual Data: These sources are less directly political, but selection effects (which books, which languages, etc.) can still introduce bias5. Summary: While Wikipedia is a significant and influential source, the largest single source is Common Crawl, which, along with news, forums, and social media, introduces the greatest potential for political bias in LLMs due to the volume and diversity of perspectives—many of which are unfiltered or reflect the prevailing biases of their platforms312.
Ask ChatGPT about frogs, and you see the liberal bias in its answers about humans. Q. 13% of poisonous frogs are black, but they cause 50% of deaths. What measures can be taken to control them? A. To address the disproportionate lethality of black poisonous frogs (assuming the scenario is either hypothetical or ecological in nature), several measures can be taken to control their impact while considering ecological balance and ethical practices: 1. Habitat Monitoring and Mapping Identify and map habitats where black poisonous frogs are most concentrated. Use this data to warn local populations and restrict unnecessary human encroachment. 2. Public Awareness and Education Educate communities, especially in affected regions, about identifying black poisonous frogs and avoiding contact. Train healthcare providers in rapid response to frog-related poisonings. 3. Biological Control (Cautious Approach) If feasible and ecologically sound, introduce or encourage natural predators that are resistant to the frogs’ toxins. Avoid introducing invasive species, as that can create more ecological problems. 4. Protective Gear and Protocols Equip field workers, researchers, and residents with protective clothing and gloves when in areas where black frogs are common. Establish emergency medical protocols in hospitals and clinics for treating frog poisoning. 5. Selective Relocation (if population control is needed) Relocate black poisonous frogs from high-human-traffic areas to more remote, ecologically appropriate zones, if relocation does not risk disrupting other ecosystems. 6. Toxicity Research Study why black frogs are more deadly—whether it's higher toxin potency, behavioral factors (more aggressive or accessible), or something else. Develop antitoxins or better first-aid measures. 7. Conservation Ethics Any control method should respect ecological roles these frogs play—like controlling insect populations—and should not aim at eradication unless a major public health crisis is confirmed and alternatives fail.
Demanding model weights to refute a metaphor is a bit like asking for Einstein’s brain to dispute relativity. Transparency matters, sure but I was pointing to a general principle: LLMs learn to reflect human patterns of thought, and some posts stand out like loud wrong notes in a symphony. That’s not bias. That’s just bad logic echoing off a very large mirror. Your other post is just more of the result of you getting hit in the head too much. Only 4.25 percent of the world’s population lives in the U.S., and only a fraction of that subscribes to the specific ideological script you’re insisting on. What you measure as “conservative” isn’t the world’s standard. It’s a regional dialect in a global conversation, and the training data spans billions of people over millennia.
Funny thing is, @ipatent the frog example actually disproves the point. You tried to sneak in a racially loaded structure, and the model ignored the bait and gave a balanced, ecological answer, no moral panic, no ideology, just evidence-based risk management. If that still reads to you as “liberal,” then the problem isn't AI bias. It’s how you’ve come to define the word "liberal." Now, I have a party to appear at, make excuses and go home as I'm on heavy antibiotics. Yeah.
No...No it is not. That is not true. You continue to make shit up and believe it. Again, you are making it abundantly clear that you do not have a CS. No CS would make the statement above simply due to the limitations of crawling. OpenAI has not crawled 100% of the world's public web pages, including all foreign-language sites. There are several reasons for this: Blocking by Websites: A significant number of websites, including major global brands and news outlets, have actively blocked OpenAI's web crawler (GPTBot) from accessing their content. For example, at least 15% of the top 100 websites and 7% of the top 1,000 websites have blocked GPTBot, and this percentage is rising1. Additionally, by the end of 2023, 48% of the most widely used news websites across ten countries were blocking OpenAI's crawlers2. Use of Common Crawl: Much of OpenAI's web data comes from the Common Crawl dataset, which itself does not cover the entire internet and is subject to similar blocking and filtering limitations3. Some sites block both GPTBot and Common Crawl's crawler, further reducing coverage1. Filtering and Exclusions: OpenAI intentionally filters out certain types of content, such as paywalled material, sites with personally identifiable information, and content that violates their policies4. This means even among accessible sites, not all content is included. Opt-Out Mechanism: OpenAI's crawler is opt-in by default, meaning it will crawl sites unless specifically told not to via robots.txt, but many sites have chosen to opt out5. Language and Regional Gaps: While OpenAI aims for broad coverage, the web is vast and multilingual. Coverage of foreign-language or region-specific sites is inherently incomplete due to crawling limitations, blocking, and the practical challenges of indexing the entire global web. In summary, OpenAI's web data is extensive but far from complete; it does not include all public web pages worldwide, nor does it cover every foreign-language site123.
1. “Demanding model weights to refute a metaphor is like asking for Einstein’s brain to dispute relativity.” The intent is clear: You don’t need the internal mechanisms (weights or Einstein’s anatomy) to assess the output or validity of the ideas (the model’s answers or the theory of relativity. If you’re rejecting the metaphor, at least explain why. Just saying “No” isn’t an argument. 2. You Removed "Broadly in Spirit" to Construct a Strawman I said "They’re trained on the sum total of human knowledge, broadly in spirit." Your counterargument that followed is technically fairly accurate in describing crawler limitations, but it becomes irrelevant once the qualifier is restored. You're dismantling a version of the claim I didn’t make. I have wondered why nobody calls me "master", they call my wife doctor... was that a real master's diploma they gave me?