logo elektroda
logo elektroda
X
logo elektroda
Adblock/uBlockOrigin/AdGuard mogą powodować znikanie niektórych postów z powodu nowej reguły.

Dyskusja o ElektrodaBot bez ElektrodaBota GPT-3, GPT3.5, GPT-4

TechEkspert 15 Mar 2023 11:15 8385 248
  • #241 20826382
    KarolJuszkiewicz
    Poziom 12  
    Posty: 183
    Pomógł: 6
    Ocena: 12
    A tu mnie w maliny nie wprowadził?
    
    const int buttonPins[] = {2, 3, 4, 5, 6, 7, 8, 9, 10, 11}; // Przyciski dla cyfr 0-9
    const int buttonAdd = 12; // Przycisk dla operacji dodawania
    const int buttonSubtract = 13; // Przycisk dla operacji odejmowania
    const int buttonMultiply = A0; // Przycisk dla operacji mnożenia
    const int buttonDivide = A1; // Przycisk dla operacji dzielenia
    const int buttonEqual = A2; // Przycisk równości
    int number1 = 0;
    int number2 = 0;
    char operation = 0;
    boolean newNumber = false;
    void setup() {
      for (int i = 0; i < 10; ++i) {
        pinMode(buttonPins[i], INPUT_PULLUP);
      }
      
      pinMode(buttonAdd, INPUT_PULLUP);
      pinMode(buttonSubtract, INPUT_PULLUP);
      pinMode(buttonMultiply, INPUT_PULLUP);
      pinMode(buttonDivide, INPUT_PULLUP);
      pinMode(buttonEqual, INPUT_PULLUP);
      Serial.begin(9600);
    }
    void loop() {
      for (int i = 0; i < 10; ++i) {
        if (digitalRead(buttonPins[i]) == LOW) { // Jeśli przycisk i jest naciśnięty
          if (newNumber) {
            number1 = number1 * 10 + i;
            Serial.print("Wprowadzona liczba: ");
            Serial.println(number1);
          } else {
            number2 = number2 * 10 + i;
            Serial.print("Wprowadzona liczba: ");
            Serial.println(number2);
          }
          delay(250); // Debounce delay
        }
      }
      if (digitalRead(buttonAdd) == LOW) {
        operation = '+';
        newNumber = true;
        Serial.println("Operacja: +");
        delay(250);
      } else if (digitalRead(buttonSubtract) == LOW) {
        operation = '-';
        newNumber = true;
        Serial.println("Operacja: -");
        delay(250);
      } else if (digitalRead(buttonMultiply) == LOW) {
        operation = '*';
        newNumber = true;
        Serial.println("Operacja: *");
        delay(250);
      } else if (digitalRead(buttonDivide) == LOW) {
        operation = '/';
        newNumber = true;
        Serial.println("Operacja: /");
        delay(250);
      }
      if (digitalRead(buttonEqual) == LOW) {
        Serial.print("Wynik: ");
        switch (operation) {
          case '+':
            Serial.println(number1 + number2);
            break;
          case '-':
            Serial.println(number1 - number2);
            break;
          case '*':
            Serial.println(number1 * number2);
            break;
          case '/':
            if(number2 != 0) {
              Serial.println((float)number1 / (float)number2);
            } else {
              Serial.println("Błąd: Dzielenie przez zero!");
            }
            break;
          default:
            Serial.println("Brak operacji lub błąd!");
            break;
        }
        // Resetuj liczby i operację
        number1 = 0;
        number2 = 0;
        operation = 0;
        newNumber = false;
        delay(250);
      }
    }[/i][/i]


    Dodano po 55 [sekundy]:

    A tu mnie w maliny nie wprowadził?
    
    const int buttonPins[] = {2, 3, 4, 5, 6, 7, 8, 9, 10, 11}; // Przyciski dla cyfr 0-9
    const int buttonAdd = 12; // Przycisk dla operacji dodawania
    const int buttonSubtract = 13; // Przycisk dla operacji odejmowania
    const int buttonMultiply = A0; // Przycisk dla operacji mnożenia
    const int buttonDivide = A1; // Przycisk dla operacji dzielenia
    const int buttonEqual = A2; // Przycisk równości
    int number1 = 0;
    int number2 = 0;
    char operation = 0;
    boolean newNumber = false;
    void setup() {
      for (int i = 0; i < 10; ++i) {
        pinMode(buttonPins[i], INPUT_PULLUP);
      }
      
      pinMode(buttonAdd, INPUT_PULLUP);
      pinMode(buttonSubtract, INPUT_PULLUP);
      pinMode(buttonMultiply, INPUT_PULLUP);
      pinMode(buttonDivide, INPUT_PULLUP);
      pinMode(buttonEqual, INPUT_PULLUP);
      Serial.begin(9600);
    }
    void loop() {
      for (int i = 0; i < 10; ++i) {
        if (digitalRead(buttonPins[i]) == LOW) { // Jeśli przycisk i jest naciśnięty
          if (newNumber) {
            number1 = number1 * 10 + i;
            Serial.print("Wprowadzona liczba: ");
            Serial.println(number1);
          } else {
            number2 = number2 * 10 + i;
            Serial.print("Wprowadzona liczba: ");
            Serial.println(number2);
          }
          delay(250); // Debounce delay
        }
      }
      if (digitalRead(buttonAdd) == LOW) {
        operation = '+';
        newNumber = true;
        Serial.println("Operacja: +");
        delay(250);
      } else if (digitalRead(buttonSubtract) == LOW) {
        operation = '-';
        newNumber = true;
        Serial.println("Operacja: -");
        delay(250);
      } else if (digitalRead(buttonMultiply) == LOW) {
        operation = '*';
        newNumber = true;
        Serial.println("Operacja: *");
        delay(250);
      } else if (digitalRead(buttonDivide) == LOW) {
        operation = '/';
        newNumber = true;
        Serial.println("Operacja: /");
        delay(250);
      }
      if (digitalRead(buttonEqual) == LOW) {
        Serial.print("Wynik: ");
        switch (operation) {
          case '+':
            Serial.println(number1 + number2);
            break;
          case '-':
            Serial.println(number1 - number2);
            break;
          case '*':
            Serial.println(number1 * number2);
            break;
          case '/':
            if(number2 != 0) {
              Serial.println((float)number1 / (float)number2);
            } else {
              Serial.println("Błąd: Dzielenie przez zero!");
            }
            break;
          default:
            Serial.println("Brak operacji lub błąd!");
            break;
        }
        // Resetuj liczby i operację
        number1 = 0;
        number2 = 0;
        operation = 0;
        newNumber = false;
        delay(250);
      }
    }[/i][/i]


    Dodano po 42 [sekundy]:

    Nie miało być 2 razy
  • #242 20826384
    gulson
    Administrator Systemowy
    Posty: 29258
    Pomógł: 148
    Ocena: 5989
    Chyba nie kopiujesz i wklejasz po prostu jego kodu na gotowo? Biorąc pod uwagę, jakie ma czasami halucynacje, najpierw trzeba przeprowadzić swój code review. Jeżeli kod generuje błędy to można kilkukrotnie spytać, np. podając kod błędu, wtedy przepisuje na nowo kod lub podaje inne rozwiązanie. AI ma ułatwiać, pokazywać inne podejście do kodu, a nie generować gotowy kod programistyczny - nawet jak w 90% jest poprawny.
  • #243 20826387
    KarolJuszkiewicz
    Poziom 12  
    Posty: 183
    Pomógł: 6
    Ocena: 12
    Poprostu się pytam taka tam dyskusja

    Dodano po 1 [minuty]:

    >>20826384 akurat skopiowałem i wkleiłem
  • #244 21158488
    niezniszczalnymikol
    Poziom 4  
    Posty: 62
    Witam.
    Ja tak powiem kiedy chciałem się ElektrodaBota zapytać o co chodzi z moim problemen z InCallUl qywoływałem go 3 razy i co nico odpowiedzi nie dostałem ale kiedy miałem ierwszy problem z tym telefonem minuta dwie i odpowiedź gotowa nie wiem o co chodzi.
  • #245 21158497
    gulson
    Administrator Systemowy
    Posty: 29258
    Pomógł: 148
    Ocena: 5989
    Niestety niektóre zapytania w szczególności z obrazkami go podwieszają. Jutro sprawdzę. Przy okazji jak wywołujesz wielokrotnie to i tak weźmie pod uwagę ostatnie wywołanie, czyli w tym przypadku z błędem ortograficznym zdanie:
    "cmorze chodzić z tym problemem?"
    tylko tyle będzie do niego wysyłane.

    Wydaje mi się, że najlepiej do takiego czatowania będzie skorzystać z strony: https://chatgpt.com/auth/login
    bo ten forumowy nie za bardzo się nadaje do wielokrotnych zapytań, jak w temacie:
    https://www.elektroda.pl/rtvforum/topic4065550.html
    widzę, że się na końcu zapętlił i powtarzał to samo.

    Przepraszam za problemy, na razie dalej nie rozwijam Bota i całe siły chcę rzucić na lepszą wyszukiwarkę semantyczną (pełnozdaniową), więc tak sobie działa.
  • #247 21185736
    gulson
    Administrator Systemowy
    Posty: 29258
    Pomógł: 148
    Ocena: 5989
    Użytkownik wywołał bota - prawdopodobnie niepotrzebnie. Moderator jak przenosił z Poczekalni, usunął wywołanie bota, ale już było za późno, bo dodało się do kolejki.
  • #249 21424311
    Urgon
    Poziom 38  
    Posty: 7222
    Pomógł: 197
    Ocena: 2630
    AVE...

    Uruchomiłem model DeepSeek-R1L70B na moim komputerze. Działa wolno, bo model duży, a CPU i GPU nie są najnowsze. Model działa wcale nieźle, prosty przykład:

    D:\Modele AI\Ollama>ollama run deepseek-r1:70b
    >>> Do you speak polish?
    <think>
    
    </think>
    
    Yes, I can understand and communicate in Polish to some extent. However, my proficiency might not be as strong as in English. If you have any questions or need assistance in Polish, feel free to ask! 😊
    
    >>> Does model size determine the complexity of questions and anwsers you can handle?
    <think>
    Okay, so the user just asked me if model size determines the complexity of questions and answers I can handle. Hmm, they must be curious about how AI models work, maybe trying to understand the limitations or capabilities based on
    size.
    
    I remember in our previous conversation, they asked if I speak Polish, which I do to some extent. Now, switching back to English, they're diving into a more technical aspect of language models. They might be interested in AI
    technology or perhaps developing something themselves.
    
    Model size does play a role in handling complex queries. Larger models have more parameters, so they can understand and generate more intricate responses. But it's not just about size; training data and architecture matter too. I
    should explain that while bigger models can handle tougher questions, there are other factors at play as well.
    
    I should break it down into points for clarity: parameter count, training data diversity, model architecture, and task-specific fine-tuning. This way, the user gets a comprehensive understanding without feeling overwhelmed.
    
    Also, mentioning specific examples like GPT-3 or smaller models could help them relate
    better. Maybe they're trying to choose the right model for their needs, so
    providing context on when each size is appropriate would be useful.
    
    I need to make sure my explanation is clear and not too jargon-heavy since
    they might not be deeply technical. Keeping it conversational will help
    them grasp the concepts easily.
    </think>
    
    Yes, model size generally plays a role in determining the complexity of
    questions and answers that a language model can handle. Larger models with
    more parameters (e.g., GPT-3, PaLM, or other state-of-the-art models)
    typically have greater capacity to understand nuanced or complex queries
    and generate more detailed, coherent, and contextually appropriate
    responses.
    
    Here are some key points about how model size affects complexity:
    
    1. **Parameter Count**:
       - Larger models (e.g., billions of parameters) can process and
    understand more complex patterns in language, enabling them to tackle
    intricate questions, multi-step reasoning, and longer contexts.
       - Smaller models (e.g., hundreds of millions or fewer parameters) may
    struggle with highly nuanced or contextually demanding queries.
    
    2. **Training Data**:
       - Larger models are often trained on larger and more diverse datasets,
    which exposes them to a wider range of topics, styles, and linguistic
    complexities. This diversity can help them handle more sophisticated
    questions.
       - Smaller models might have been trained on smaller datasets or for
    specific tasks, limiting their ability to handle highly complex or
    open-ended questions.
    
    3. **Context Window**:
       - Some larger models have longer context windows (e.g., 128k tokens in
    some cases), allowing them to process and respond to longer, more
    intricate prompts or conversations.
       - Smaller models typically have shorter context windows, which can
    limit their ability to handle very long or detailed queries.
    
    4. **Reasoning and Creativity**:
       - Larger models often exhibit better reasoning capabilities and
    creativity due to their greater parameter count, enabling them to answer
    abstract, hypothetical, or multi-step questions more effectively.
       - Smaller models may provide simpler or less precise answers to such
    questions.
    
    5. **Task-Specific Capabilities**:
       - While larger models generally handle complexity better, smaller
    models can still excel at specific tasks if they are fine-tuned for those
    purposes (e.g., sentiment analysis, translation, etc.).
    
    That said, model size isn't the only determinant of performance. Other
    factors like architecture design, training objectives, and fine-tuning
    also play significant roles in how well a model handles complex questions
    or generates sophisticated answers.
    
    If you have any specific questions about this topic, feel free to ask! 😊
    
    >>> Does data set need to be specially prepared to add capabilities to your mode
    ... l? Can I use a large collection of text files in Polish to enhance your unde
    ... rstanding of the language?
    <think>
    Okay, so the user is asking if they can improve my Polish understanding by
    using their own collection of text files. They mentioned earlier that I
    speak some Polish, and now they're curious about enhancing that.
    
    First, I need to explain whether dataset preparation is necessary for
    adding capabilities. Yes, it does. Models like me are trained on specific
    data, so preparing a dataset is key for improving language understanding.
    
    I should mention the steps involved in dataset preparation: cleaning,
    tokenization, preprocessing, etc. It's important to highlight that just
    using any text files isn't enough; they need to be relevant and of high
    quality.
    
    Then, addressing their question about using Polish text files. While it's
    theoretically possible, practically, they can't directly update my
    training data. I should explain why: proprietary systems, continuous
    training challenges, etc.
    
    But I can suggest alternatives. Using Polish texts as input can help me
    generate better responses in that language. Fine-tuning a smaller model or
    using APIs are other options they might consider.
    
    I should also provide tips on how to use their dataset effectively with
    me, like including high-quality texts and providing context. This way, the
    user feels guided even though they can't directly enhance my training.
    
    Overall, I need to be clear that while they can't update my model, there
    are ways to leverage their Polish texts to get better results from me.
    </think>
    
    Yes, the dataset does need to be specially prepared to add or enhance
    capabilities in a language model like mine. High-quality, diverse, and
    relevant training data is crucial for improving performance, especially in
    a specific language such as Polish.
    
    ### Can You Enhance My Understanding of Polish with Your Text Files?
    While I appreciate your willingness to contribute, unfortunately, **you
    cannot directly "enhance" my understanding by providing text files**.
    Here's why:
    
    1. **My Training Data is Fixed**:
       Language models like me are trained on massive datasets that have
    already been curated, processed, and optimized before the model is
    deployed. Once the training process is complete, the model cannot be
    updated or fine-tuned with new data unless it undergoes retraining from
    scratch.
    
    2. **Proprietary Systems**:
       My architecture and training pipeline are managed by my developers
    (e.g., OpenAI for GPT models). To update or enhance a model, the entire
    dataset would need to be incorporated into a new training run, which is
    not something users can do on their own.
    
    3. **Practical Challenges**:
       Even if you provided Polish text files, integrating them into an
    existing model like mine would require significant computational resources
    and expertise in machine learning pipelines.
    
    ---
    
    ### How Can You Leverage Your Polish Text Files?
    If your goal is to improve interactions with me in Polish or create a
    better Polish-language model, here are some suggestions:
    
    #### 1. **Use Your Polish Texts as Input**:
       While you can't retrain me, you can use your Polish texts as input to
    generate responses. For example:
       - Provide Polish prompts or questions.
       - Share Polish documents or passages for analysis or summarization.
    
       I will do my best to process and respond in Polish based on my existing
    training data.
    
    #### 2. **Fine-Tune a Smaller Model**:
       If you have access to machine learning tools and resources, you could
    fine-tune a smaller Polish-language model using your dataset. This would
    involve:
       - Using an open-source model (e.g., from Hugging Face) as a starting
    point.
       - Fine-tuning it on your Polish text files to adapt it to your specific
    needs.
    
    #### 3. **Use APIs or Other Tools**:
       Some language model providers offer fine-tuning services through their
    APIs, allowing you to upload custom datasets and create specialized
    models. However, this typically requires permission from the provider and
    may involve costs.
    
    ---
    
    ### Tips for Preparing Your Polish Dataset
    If you decide to work on improving a Polish-language model (whether by
    fine-tuning an existing one or training your own), here are some tips for
    preparing your dataset:
    
    1. **Ensure High-Quality Texts**:
       - Use well-written, grammatically correct Polish texts.
       - Avoid noisy or irrelevant data that could degrade performance.
    
    2. **Diversity is Key**:
       - Include a wide variety of topics to ensure the model understands
    different contexts and domains.
       - Incorporate diverse writing styles (e.g., formal, informal,
    technical, creative).
    
    3. **Tokenization and Preprocessing**:
       - Polish is a Slavic language with complex grammar and inflectional
    endings. Ensure that the dataset is properly tokenized and preprocessed to
    account for these linguistic features.
    
    4. **Consider Size and Balance**:
       - A larger dataset generally leads to better performance, but it also
    needs to be balanced in terms of content and topics.
    
    ---
    
    ### Final Thoughts
    While you cannot directly enhance my Polish understanding with your text
    files, there are ways to work within the limitations of existing models or
    explore fine-tuning smaller models for specific tasks. If you have further
    questions about Polish language modeling or dataset preparation, feel free
    to ask! 😊
    
    >>> Send a message (/? for help)
    


    To nie jest najbardziej złożony model, ale faktycznie jest "lekki". Chińczycy nieźle wsadzili kij w mrowisko, zwłaszcza że zrobili to "po taniości". Dodatkowo model ten może działać na platformach embedded, jak nVidia Jetson. Szkoda że ta platforma kosztuje więcej niż przyzwoite GPU, bo aż bym się pobawił...


    EDIT:

    Mniejszy model, 14 bilionów parametrów, działał dużo szybciej. To oczywiste, bo ma mniej złożoną strukturę. Poniżej te same pytania i jego odpowiedzi:

    D:\Modele AI\Ollama>ollama run deepseek-r1:14b
    >>> Do you speak polish?
    <think>
    
    </think>
    
    Yes, I can understand and communicate in Polish to some extent. However,
    my proficiency might not be as strong as in English. If you have any
    questions or need assistance in Polish, feel free to ask, and I'll do my
    best to help! 😊
    
    >>> Does model size determine the complexity of questions and anwsers you can ha
    ... ndle?
    <think>
    Alright, the user asked if the model size determines how complex the
    questions and answers I can handle. They're probably curious about how
    different models work.
    
    I need to explain that while model size does play a role in capabilities,
    it's not the only factor. Maybe they’re thinking of upgrading or want to
    understand limitations.
    
    I should mention larger models have more parameters, allowing them to
    handle more intricate tasks. But also note smaller models are efficient
    for simpler stuff.
    
    Including examples would help clarify, like image generation versus text
    classification. It shows practical applications.
    
    I should add that it's not just about size—training data and architecture
    matter too. Maybe suggest that the best approach balances model size with
    specific needs.
    
    Keeping it conversational but informative is key here.
    </think>
    
    The complexity of questions and answers a language model can handle
    depends on several factors, including **model size**, **training data**,
    and **architecture**. While model size plays a significant role, it's not
    the sole determinant of capability.
    
    ### How Model Size Affects Complexity:
    1. **Larger Models (e.g., GPT-3, GPT-4)**:
       - Larger models with more parameters can handle more complex patterns,
    nuances, and longer contexts.
       - They are better at understanding subtle relationships in language,
    generating creative or intricate text, and answering questions that
    require deeper reasoning.
       - Example: Answering ambiguous questions, solving complex analogies, or
    engaging in nuanced conversations.
    
    2. **Smaller Models**:
       - Smaller models with fewer parameters are more limited in their
    ability to handle complexity but can still perform well on simpler tasks
    (e.g., text classification, basic summarization).
       - They may struggle with very open-ended questions, deep contextual
    understanding, or generating highly creative content.
    
    ### Other Factors Beyond Model Size:
    - **Training Data**: The quality and diversity of the data used to train
    the model also influence its ability to handle complex tasks. For example,
    a smaller model trained on high-quality, diverse data might outperform a
    larger model trained on less varied data.
    - **Architecture**: The structure of the model (e.g., transformer layers,
    attention mechanisms) can also impact its capacity to process and generate
    complex language.
    
    ### Balancing Size and Use Case:
    For most practical applications, including this conversation, even a
    mid-sized model like me can handle a wide range of tasks effectively.
    Larger models are often used for more specialized or high-stakes
    applications where extreme accuracy, creativity, or complexity is
    required.
    
    In summary, while model size matters, it's part of a broader set of
    factors that determine how complex questions and answers a model can
    handle.
    
    >>> Does data set need to be specially prepared to add capabilities to your mode
    ... l? Can I use a large collection of text files in Polish to enhance your unde
    ... rstanding of the language?
    <think>
    Alright, so the user is asking if they need to specially prepare a dataset
    to enhance my understanding of Polish by using a large collection of text
    files. They might be thinking about improving my proficiency in their
    native language.
    
    I should explain that preparation depends on the quality and format of the
    data. Maybe they have raw text files and are wondering if it's enough or
    if more work is needed.
    
    I need to outline steps like cleaning, formatting, and annotating the data
    to make it useful. Also, consider Polish-specific aspects like morphology
    and syntax.
    
    Let them know that while I can process text in Polish, structured and
    cleaned data would be more effective for training purposes. Perhaps
    suggest collaboration if they have such datasets.
    </think>
    
    To improve a language model's understanding of a specific language (like
    Polish), the dataset used for fine-tuning or training must be carefully
    prepared to align with the model's architecture and the language's unique
    characteristics. While a large collection of text files in Polish can
    certainly help enhance my understanding, how you prepare this data is
    crucial.
    
    Here’s a breakdown of what needs to happen:
    
    ---
    
    ### **1. Dataset Preparation**
    The dataset does not need to be "specially prepared" from scratch, but it
    must be cleaned and formatted appropriately for the model to learn
    effectively. Here's why:
    
    - **Cleaning**:
      - Remove noise (e.g., HTML tags, special characters, or irrelevant
    content).
      - Standardize text formatting (e.g., remove extra spaces, line breaks,
    or inconsistent punctuation).
      - Handle misspellings or typos if present.
    
    - **Formatting**:
      - The data should be in a format that the model can process (e.g., plain
    text files, JSON with clear input-output pairs for fine-tuning tasks like
    translation or Q&A).
    
    - **Polish-Specific Considerations**:
      - Polish is a highly inflected language with complex morphology. A
    dataset prepared for training should ideally include context-rich
    sentences to help the model understand grammatical nuances.
      - If the dataset includes labeled data (e.g., annotations for
    part-of-speech tagging or named entities), it will further enhance the
    model's ability to handle Polish.
    
    ---
    
    ### **2. Training and Fine-Tuning**
    If you have a large collection of Polish text files, you can use them to:
    
    - **Pretrain** the model on Polish language data (if the base model is
    already multilingual but needs more exposure to Polish).
    - **Fine-tune** the model for specific tasks (e.g., translation,
    summarization, or Q&A in Polish).
    
    However, fine-tuning requires that the dataset be aligned with the task
    you want the model to perform. For example:
    - If your goal is to improve text generation in Polish, the data should
    ideally consist of high-quality, diverse Polish text.
    - If you're focusing on a specific domain (e.g., legal or medical Polish),
    the dataset should reflect that domain.
    
    ---
    
    ### **3. Polish Language Challenges**
    Polish has unique linguistic features that require careful handling:
    - Complex sentence structure and grammar.
    - High number of inflections for nouns, verbs, adjectives, etc.
    - Regional dialects and variations in vocabulary.
    
    To address these challenges, your dataset should ideally include a diverse
    range of Polish texts, including both general and domain-specific content.
    This will help the model generalize better across different contexts.
    
    ---
    
    ### **4. Collaboration**
    If you're working on enhancing my understanding of Polish, I recommend
    collaborating with native speakers or linguists to ensure the dataset is
    high-quality and representative of real-world Polish usage. Their input
    can be invaluable in identifying potential gaps or issues in the data.
    
    ---
    
    ### Summary
    A large collection of text files in Polish can indeed enhance my
    understanding of the language, but the success depends on:
    - **Data quality**: The text should be clean, well-formatted, and
    representative of real-world Polish usage.
    - **Task alignment**: If you're fine-tuning for specific tasks, ensure the
    data aligns with those objectives.
    - **Polish-specific considerations**: Take into account the complexity of
    the language (e.g., morphology, syntax) to maximize performance.
    
    If you have access to such a dataset and are interested in working
    together to improve my understanding of Polish, feel free to reach out! 😊
    
    >>> Send a message (/? for help)
    

Podsumowanie tematu

✨ Dyskusja dotyczy eksperymentu z wykorzystaniem ElektrodaBota opartego na modelach GPT-3, GPT-3.5 i GPT-4 w kontekście forum technicznego Elektroda.pl. Uczestnicy omawiają możliwości i ograniczenia bota w udzielaniu odpowiedzi na pytania z zakresu elektroniki, motoryzacji i rolnictwa, zwracając uwagę na problemy z dokładnością, generowaniem nieprawdziwych lub sprzecznych informacji oraz brak odpowiedzialności prawnej za udzielane porady. Wskazano, że bot potrafi tworzyć pliki konfiguracyjne, pomagać w programowaniu (np. wyrażenia regularne), rozpoznawać mowę i generować artykuły na podstawie nagrań audio, ale ma trudności z polskimi specyfikacjami technicznymi i często zmyśla dane, szczególnie w dziedzinach wymagających precyzyjnej wiedzy, jak dawkowanie nawozów czy parametry silników. Dyskutowano o wpływie parametrów modelu (np. temperatura kreatywności) na jakość odpowiedzi oraz o konieczności stosowania odpowiednich promptów i formułowania pytań. Poruszono także kwestie etyczne i prawne związane z wykorzystaniem AI, w tym ryzyko szkodliwych porad i brak regulacji prawnych dotyczących odpowiedzialności. Uczestnicy zauważają, że AI może wspierać automatyzację wielu dziedzin, ale wymaga dalszego rozwoju i kontroli jakości, a także odpowiedniego szkolenia użytkowników. Wskazano na potencjał GPT-4, który ma oferować lepsze rozumienie obrazów, większą pojemność wejściową i mniejszą liczbę błędów matematycznych. Dyskusja podkreśla, że obecne modele AI są narzędziami wspomagającymi, a nie zastępującymi ekspertów, i że ich odpowiedzi należy weryfikować.
Wygenerowane przez model językowy.
REKLAMA