4 ms·

Hunch: the hype peak is 24-72 months away

by provenance 4y ago

EdwardDiego 4y ago
AI winter 2, electric Boogaloo?
mountainriver 4y ago
Hunch: we are just getting started and AI will reshape society in unimaginable ways
- provenance 4y ago
  While I agree in the long run, I suspect we may run into limitations in generative AI at some point this decade. GPT-4 is incapable of any form of genuine discernment about the veracity of any statement that it makes. This "hallucination" problem might be more difficult to solve, as critical thinking skills are still the exclusive domain of evolved sentinent beings.
  - Clubber 4y ago
    If execs think it will improve their bonus, please give a warm welcome our new GPT-4 development team!
    - provenance 4y ago
      propose a strategy whereby chatgpt recursively writes and applies code to solve the problem of "hallucinations" with chatgpt, as defined by Sam Altman As an AI language model, I do not experience hallucinations, but I can offer a strategy for recursively writing and applying code to solve the problem of "hallucinations" with ChatGPT, as defined by Sam Altman. Here's a possible strategy: Define the problem: The first step is to define what we mean by "hallucinations" in the context of ChatGPT. Based on Sam Altman's definition, we can assume that he is referring to instances where ChatGPT generates text that is inaccurate or misleading, potentially leading to erroneous conclusions or actions. Collect training data: The next step is to collect training data that includes examples of "hallucinations" that ChatGPT has generated in the past. This data will be used to train the code that will help identify and correct such errors in the future. Develop a neural network: We can use the training data to develop a neural network that is specifically designed to identify "hallucinations" in ChatGPT's generated text. This neural network can be trained to recognize patterns in the data that correspond to inaccurate or misleading information. Implement the code: Once the neural network is developed, we can implement the code that will apply the neural network to ChatGPT's output. This code will work recursively, meaning that it will continually check and correct ChatGPT's output as new text is generated. Test and refine: After implementing the code, we will need to test it to ensure that it is effective in identifying and correcting "hallucinations" in ChatGPT's generated text. We may need to refine the code based on the results of testing and additional training data. Continuous improvement: As with any AI system, the effectiveness of the code will depend on ongoing training and improvement. We will need to continually monitor ChatGPT's output and update the code as needed to ensure that it remains accurate and reliable. Overall, the strategy involves developing a neural network that can identify "hallucinations" in ChatGPT's generated text and implementing code that will apply this neural network recursively to continually improve the accuracy and reliability of ChatGPT's output.
- JohnFen 4y ago
  Could be true. That's unrelated to the hype cycle, though, except that we won't see any of the future you're envisioning until we get past the hype stage.

Hunch: the hype peak is 24-72 months away

EdwardDiego 4y ago

AI winter 2, electric Boogaloo?

mountainriver 4y ago

Hunch: we are just getting started and AI will reshape society in unimaginable ways

provenance 4y ago
While I agree in the long run, I suspect we may run into limitations in generative AI at some point this decade. GPT-4 is incapable of any form of genuine discernment about the veracity of any statement that it makes. This "hallucination" problem might be more difficult to solve, as critical thinking skills are still the exclusive domain of evolved sentinent beings.
- Clubber 4y ago
  If execs think it will improve their bonus, please give a warm welcome our new GPT-4 development team!
  - provenance 4y ago
    propose a strategy whereby chatgpt recursively writes and applies code to solve the problem of "hallucinations" with chatgpt, as defined by Sam Altman As an AI language model, I do not experience hallucinations, but I can offer a strategy for recursively writing and applying code to solve the problem of "hallucinations" with ChatGPT, as defined by Sam Altman. Here's a possible strategy: Define the problem: The first step is to define what we mean by "hallucinations" in the context of ChatGPT. Based on Sam Altman's definition, we can assume that he is referring to instances where ChatGPT generates text that is inaccurate or misleading, potentially leading to erroneous conclusions or actions. Collect training data: The next step is to collect training data that includes examples of "hallucinations" that ChatGPT has generated in the past. This data will be used to train the code that will help identify and correct such errors in the future. Develop a neural network: We can use the training data to develop a neural network that is specifically designed to identify "hallucinations" in ChatGPT's generated text. This neural network can be trained to recognize patterns in the data that correspond to inaccurate or misleading information. Implement the code: Once the neural network is developed, we can implement the code that will apply the neural network to ChatGPT's output. This code will work recursively, meaning that it will continually check and correct ChatGPT's output as new text is generated. Test and refine: After implementing the code, we will need to test it to ensure that it is effective in identifying and correcting "hallucinations" in ChatGPT's generated text. We may need to refine the code based on the results of testing and additional training data. Continuous improvement: As with any AI system, the effectiveness of the code will depend on ongoing training and improvement. We will need to continually monitor ChatGPT's output and update the code as needed to ensure that it remains accurate and reliable. Overall, the strategy involves developing a neural network that can identify "hallucinations" in ChatGPT's generated text and implementing code that will apply this neural network recursively to continually improve the accuracy and reliability of ChatGPT's output.
JohnFen 4y ago
Could be true. That's unrelated to the hype cycle, though, except that we won't see any of the future you're envisioning until we get past the hype stage.