3 ms·

These dismissive autocomplete articles are falling on the opposite side of sensational journalism as the love and hitler type. These are more than autocomplete

by basch 4y ago

These dismissive autocomplete articles are falling on the opposite side of sensational journalism as the love and hitler type. These are more than autocomplete. They demonstrate at least some amount of fuzzy logic and substitution and recall abilities, which makes sense from their origins as translators. Because they are weighing the conversation, the conversation becomes the rules of the program. The article I have here shows why they are more than simple autocomplete storytellers. https://telegra.ph/Bing-course-corrected-itself-when-asked-02-18 https://telegra.ph/Bing-course-corrected-itself-when-asked-0... I asked it to reprogram itself, it figured out a solution to the problem and implemented it. From one line of text. The running process is loosely programmable, real time, just by talking to it with natural language requests and commands.

Jensson 4y ago
It was built to encode grammar, and word2vec similarities. It reads your instructions and then binds future output based on its grammatic understanding of those. For example, it has read many bash tutorials and based on those it encodes "> ls" to be followed by file data entered above in the article. Then add in word2vec translations and it can do the above for similar words, or translated words, or words in another format. And you got everything ChatGPT can do, it isn't really magic. And note, ChatGPT doesn't have a sense or the conversation or a sense of self. It just prepends the entire previous conversation, applies the pure function to it and sends back the output. It doesn't "remember", it just predicts the next lines in each conversation. You can rewrite what it wrote, send it, and it tries to follow up based on the new data. But they removed the ability for the user to change what ChatGPT writes, and then humans gets the illusion that ChatGPT actually thinks and have a consistent conversation. Nobody thought it did when they used GPT-3, it is basically the same thing just chat GPT-3 lets you alter all the prompts freely and then it is obvious that GPT-3 isn't a person.

by basch 4y ago

Jensson 4y ago

It was built to encode grammar, and word2vec similarities. It reads your instructions and then binds future output based on its grammatic understanding of those. For example, it has read many bash tutorials and based on those it encodes "> ls" to be followed by file data entered above in the article. Then add in word2vec translations and it can do the above for similar words, or translated words, or words in another format. And you got everything ChatGPT can do, it isn't really magic. And note, ChatGPT doesn't have a sense or the conversation or a sense of self. It just prepends the entire previous conversation, applies the pure function to it and sends back the output. It doesn't "remember", it just predicts the next lines in each conversation. You can rewrite what it wrote, send it, and it tries to follow up based on the new data. But they removed the ability for the user to change what ChatGPT writes, and then humans gets the illusion that ChatGPT actually thinks and have a consistent conversation. Nobody thought it did when they used GPT-3, it is basically the same thing just chat GPT-3 lets you alter all the prompts freely and then it is obvious that GPT-3 isn't a person.