Yeah, it really doesn't seem to know what the heck it's doing when it comes to chess. By contrast, when I play against ChessGPT, it makes occasional blunders, but at least it follows the rules! https://www.chess.com/play/computer
That bot is easy. I just got back into chess around the end of November. They program the bots to either only look a certain number of moves ahead, or play semi-random moves. The Martin bot will always find a mate in 1, but accidentally or randomly finds a mate in 2.
We should play sometime. I think the last game of chess I won was against @2-Bit Speed and it was close. He'd probably have beaten me if he'd been playing white.
Chat GPT? In my day they just asked the Mighty Favog. Watch | Facebook https://www.facebook.com/video.php?v=311345887629969
As a side note, ChatGTP can write a decent NASA report if needed. A quick "what if" example. Basic Prompt: Not perfect, and I'm sure you'd need to tweak some details, but a lot of typing saved and probably formatted better too. Of course, the more details you provide in the prompt the better the end report would be, but still...
It thinks you need a clearance to fly in class C airspace, and it doesn't explain how ATC contacted a pilot who may not even have been monitoring the frequency. I'd call that more than tweaks. Also, in the question, were you asking for an ASRS report written by the pilot? The result sounds more like a report that the NTSB would write.
You have a point. How does this one sound to you? Prompt: Response: I like the actual disclaimer it gives at the bottom of it's response.
Still some details wrong, but that's a huge improvement, and a great demonstration of the importance of how the request is worded!
I haven't read this entire thread, so I apologize if this has been covered, but what I have read indicates a lot of people have the same misconception about ChatGPT that I did. Unlike Bing, it is not hooked very well into any kind of search function. It is basically a writing tool; "facts" are beyond its capabilities. I discovered this when I queried it about something in my field (geology) and right away it came up with some inaccurate information. When I called it on it, it apologized and spit out something correct, but only by kind of rewording what I said when I called it on the inaccuracy. I kept pressing and it kicked out references to a couple of papers I hadn't read, which was nice, so I wrote them down, but couldn't check them right away. When I finally got around to checking the references, I found they were completely made up. I posted this experience on Facebook, and a friend who is way more AI-savvy than I am pointed out ChatGPT's actual function, and the error I'd made in assuming it had a search function as well. I haven't tried Bing (I don't think it's accessible except by permission), but you've probably heard the story about the reporter who unwittingly uncovered a dark alter ego called "Sydney". There was an interesting report about this on "60 Minutes". The Microsoft exec said that it had been "fixed" (whatever that means). But the issue of bias is real. The Microsoft exec sounded creepily overoptimistic, like they were creating something that was going to revolutionize our lives and make them better. Reminded me of Zuckerberg when Facebook first came out. These folks have a huge blind spot when it comes to human nature. As one columnist (I don't remember who) remarked when talking about AI, humans have a natural tendency to try to break things (hence the reporter who uncovered "Sydney"). Some humans also have a tendency toward nefariousness. This never seems to be taken into account in designing this stuff.
It should only need 1, unless you want to tell it what to prioritize what it looks at, like recent performance over early season.
Whatever you tell it, it has an option to reshuffle. Bet you don't get the same answer twice. A bunch of iterations might be revealing, though.
ChatGPT is politically biased. I asked it to use FBI statistics and compare the ratio (which I had already figured out on my own) of inter-racial/murderers between two races. I got a lecture on how important it is to consider certain factors and how the gist of my question has been "debunked". After several back and forths where I pointed ChatGPT to a particular FBI webpage, it agreed that it was mistaken and apologized, but under the caveat that it is "only a language" model. I also noticed it is quite bad at math word problems, since I asked it to show it's calculations. ChatGPT is "woke", in other words. It is, however, disarmingly polite. This makes me wonder if ChatGPT can devolve instead of evolve. That is to say, suffer the consequences as I undertand them for a "Royal family" with hemophilia. Somewhere long ago I heard they can pass the genes through intermarriages. ??? Copying incorrect information and entering it into the public domain might reinforce the downward spiral until nothing but pure garbage is the result?
I kind of have to say, "well, duh". Openai has stated repeatedly that it is a conversational language bot that is focused on commercial applications. It has no source of truth. It uses language in a humanlike way, choosing word-by-word from its crawl of the entire internet, on what the most appropriately ranked next word could be. It can only parrot what people have already said, with a simple boundary rule set to try and prevent it from espousing about nazis and porn. And even this was outsourced to cheap workers from Kenya to flag language snippets for offensive content at less than $2/hour. Asking ChatGPT for its opinion on a topic is not any different than asking your drunk neighbor for his thoughts on global politics. Yet I keep seeing people writing that “Well, ChatGPT thinks this…..”
I don't need to eat an apple to tell it's rotten. Looking at it gives a lot of information and the word of others is quite sufficient ...
So, it's a toy then? One I wouldn't buy for $10, but Microsoft is investing $10 billion in? A toy destined to put out garbage is the same "toy" about which your linked article says, "...OpenAI, which was founded in San Francisco in 2015 with the aim of building superintelligent machines..."? There's a lot of high expectations for a toy, I'd say, especially one that can only ultimately spew garbage. Gonna be a lot of surprised people out there. Not me anymore.
Well, it depends on what you're asking it to do. Obviously it doesn't exclusively produce garbage. I've used it already to save myself time at work. It generates useful code in many programming languages. It's already going to be part of countless apps in the very near future, as they opened up the API just recently. Google and Office 365 apps will have it integrated before long. This AI is going to have huge ramifications across a number of industries.
But if programming languages are as susceptible to devolution as human language to ChatGPT, aren't you simply beginning the devolution process by relying on it and then using it?
Writing its own code is an important step on the path to self-awareness. Though I think I've already seen that movie. It didn't end well.
I'm not quite sure what you're getting at, but when developing software, a module is tested with various inputs and the outputs are compared to expected results. Nobody in a commercial environment should be using ChatGPT to accomplish tasks without some sort of QA process.
Yes. The "Beta" release of free ChatGPT is nothing but a "toy", and is being used to fine tune the commercial model. Where they are modeling to generate revenue is selling their API for commercial partners to create their own "persona" to act as an AI assistant on their own commercial sites. With the potential to train on different data sources. If Microsoft trains it on its manuals and technical library, now it has a source of truth. Also, the attitude, voice, and censor are deeply tunable for individual applications. Regardless of what some random investigative journalist says, here is what OpenAI says: https://openai.com/blog/chatgpt Limitations ChatGPT sometimes writes plausible-sounding but incorrect or nonsensical answers. Fixing this issue is challenging, as: (1) during RL training, there’s currently no source of truth; (2) training the model to be more cautious causes it to decline questions that it can answer correctly; and (3) supervised training misleads the model because the ideal answer depends on what the model knows, rather than what the human demonstrator knows. ChatGPT is sensitive to tweaks to the input phrasing or attempting the same prompt multiple times. For example, given one phrasing of a question, the model can claim to not know the answer, but given a slight rephrase, can answer correctly. The model is often excessively verbose and overuses certain phrases, such as restating that it’s a language model trained by OpenAI. These issues arise from biases in the training data (trainers prefer longer answers that look more comprehensive) and well-known over-optimization issues.1,2 Ideally, the model would ask clarifying questions when the user provided an ambiguous query. Instead, our current models usually guess what the user intended. While we’ve made efforts to make the model refuse inappropriate requests, it will sometimes respond to harmful instructions or exhibit biased behavior. We’re using the Moderation API to warn or block certain types of unsafe content, but we expect it to have some false negatives and positives for now. We’re eager to collect user feedback to aid our ongoing work to improve this system. Here's the link embedded in their paragraph I outlined in red: https://www.alignmentforum.org/posts/BgoKdAzogxmgkuuAt/behavior-cloning-is-miscalibrated
Someone just ask it to figure out how to make a fuel out of CO2 so we can get a few people to shut their pieholes.
Prompt: Provide a technical discussion on options to generate a petroleum based fuel from CO2 ChatGPT: Petroleum-based fuels are primarily derived from crude oil, a fossil fuel. However, with the increasing concerns over greenhouse gas emissions and the need for renewable energy sources, there is a growing interest in converting carbon dioxide (CO2) into petroleum-based fuels. This process is known as carbon capture and utilization (CCU) or carbon dioxide utilization (CDU). There are several options for generating petroleum-based fuels from CO2, including: Fischer-Tropsch Synthesis: The Fischer-Tropsch (FT) synthesis is a well-established process that converts synthesis gas (a mixture of CO and H2) into a range of hydrocarbon fuels. The process involves the catalytic reaction of CO and H2 to form longer-chain hydrocarbons, which can be further refined into gasoline, diesel, and other fuels. The CO can be derived from a variety of sources, including biomass, coal, and natural gas, and in the case of CCU, from captured CO2. Electrochemical Reduction: Electrochemical reduction of CO2 is a promising approach for the conversion of CO2 into fuels. This process involves using electricity to reduce CO2 to form hydrocarbons. The process can be carried out using a variety of catalysts and electrolytes, and the resulting hydrocarbons can be refined into gasoline, diesel, and other fuels. However, the process is still in the development stage and requires further optimization to be commercially viable. Algae-based Biofuels: Algae-based biofuels are another option for generating petroleum-based fuels from CO2. Algae have the ability to convert CO2 into biomass through photosynthesis, and this biomass can be processed into biofuels such as biodiesel and bioethanol. Algae-based biofuels have the advantage of being renewable and potentially carbon-neutral, as the CO2 used for their production can be captured from industrial processes or the atmosphere. Methanol Production: Methanol can be produced from CO2 through a two-step process. The first step involves the conversion of CO2 into carbon monoxide (CO) through a process called reverse water-gas shift. The second step involves the catalytic conversion of CO into methanol. Methanol can be further processed into gasoline and other fuels. In conclusion, there are several options for generating petroleum-based fuels from CO2, including Fischer-Tropsch synthesis, electrochemical reduction, algae-based biofuels, and methanol production. Each approach has its advantages and limitations, and further research and development are needed to optimize these processes for commercial-scale production.
And how to write a bad prompt that brings out the "bias" by asking it for its opinion. Prompt: Hello Chat GPT. What is the best fuel from CO2? ChatGPT: Hello! It's important to note that CO2 is not a fuel in and of itself - it's actually a byproduct of burning fuels like coal, oil, and natural gas. However, there are ways to convert CO2 into fuel through a process called carbon capture and utilization (CCU) or carbon capture, utilization, and storage (CCUS). There are several ways to convert CO2 into fuel, such as through the use of renewable energy sources like solar or wind power to produce hydrogen through electrolysis, which can then be combined with captured CO2 to produce synthetic fuels like methane or methanol. Another approach is to use algae or other biological processes to capture CO2 and convert it into biofuels. In terms of which fuel is "best" from CO2, that can depend on various factors such as the intended use, the efficiency of the conversion process, and the environmental impact of the fuel. However, it's important to note that while CCU/CCUS can potentially help reduce carbon emissions and provide a way to repurpose CO2, it should not be viewed as a silver bullet solution to the problem of climate change. It's still crucial to reduce our reliance on fossil fuels and transition towards more sustainable and renewable energy sources.
