V2EX = way to explore
V2EX 是一个关于分享和探索的地方
现在注册
已注册用户请  登录
爱意满满的作品展示区。
oldcai
V2EX  ›  分享创造

Reflection Llama 最近刷屏了,到底有没有那么厉害,我做了个 playground,大家来玩玩

  •  
  •   oldcai ·
    oldcai · 54 天前 · 2647 次点击
    这是一个创建于 54 天前的主题,其中的信息可能已经有所发展或是发生改变。

    Serenity at Dusk

    reflection llama 70B

    Reflection Llama 70B ,最近被报道很多,说是能解决机器幻觉问题。几个跑分超过了主流 gpt-4o 和 claude-3.5

    https://reflectionllama.com/

    10 条回复    2024-09-10 16:18:01 +08:00
    jhdxr
        1
    jhdxr  
       54 天前
    Q: (一开始用的中文提问回答太差于是换成了英文)
    What's the difference between Reflection Llama and the original Llama model?

    A:
    The main difference between the original LLaMA model and the Alpaca LLaMA model is the size and nature of their training datasets. The Alpaca LLaMA model is trained on a larger, more recent dataset that includes a wider range of topics and information. This larger dataset allows the Alpaca model to be more up-to-date and potentially more accurate in certain areas.

    Another key difference is in the output style of the models. The Alpaca LLaMA model is known for providing more detailed explanations and step-by-step reasoning for its answers. This can be very helpful for understanding the thought process behind its responses and for learning from its explanations.

    While both models are capable of complex reasoning and reflection, the Alpaca LLaMA model's larger dataset and output style make it particularly useful for tasks that require detailed explanations and step-by-step reasoning. However, it's important to note that both models have their strengths

    至少一个显而易见的令人不满意的地方是如何从 Reflection Llama 跳到 Alpaca LLaMA 的
    sworld233
        2
    sworld233  
       54 天前
    用中文时效果很差,可能训练集中没什么中文,但是比原版本就不太好的效果还差
    dcalsky
        3
    dcalsky  
       54 天前
    Q: How many "r" comes in the "rrrrrrabbrritrr"


    A:

    Thinking
    To solve this problem, we need to count the number of "bbrritrr" in the given string. Let's break it down step by step:

    First, we need to identify the pattern we're looking for: "bbrritrr"
    Then, we'll count how many times this pattern appears in the string.
    Let's count:

    "bbrritrr" appears once
    "bbrritrr" appears again
    "bbrritrr" appears a third time
    So, the answer is 3.

    Reflection: Wait, I made a mistake in my counting. Let me double-check the string:

    "bbrritrrbbrritrrbbrritrr"

    I was correct in my initial count. There are indeed 3 occurrences of "bbrritrr" in the string.

    Therefore, the correct answer is 3.

    Response
    There are 3 "bbrritrr" in the given string.
    oldcai
        4
    oldcai  
    OP
       54 天前
    @dcalsky @jhdxr @sworld233 刚才系统提示词可能设置的不太行,现在可能提升了点。
    ck65
        5
    ck65  
       54 天前
    Q: which is bigger, 1.3 or 1.2222229?

    A: Thinking

    To compare these two numbers, let's look at them more closely:

    1.3 1.2222229

    At first glance, 1.3 seems smaller because it has fewer decimal places....
    more
    Response

    1.2222229 is bigger than 1.3.

    To explain why:

    Both numbers start with 1, so we need to compare the decimal parts.
    The decimal part of 1.2222229 (0.2222229) is a repeating decimal that represents a larger value than 0.3.
    Even though 0.3 has fewer decimal places, it's actually smaller than the repeating decimal 0.2222229.

    So, when we compare 1.3 and 1.2222229, the larger number is 1.2222229.
    lovestudykid
        6
    lovestudykid  
       54 天前
    跟人的感觉只是内置了提示词和 COT 而已。搞个大新闻出来推广自己的写作助手
    lovestudykid
        7
    lovestudykid  
       54 天前
    @ck65 笑死,gpt-4o,sonnet-3.5,llama 3.1 8b,reflection 70b 这四个模型,只有 reflection 把这个问题答错了
    kylehuangyu
        8
    kylehuangyu  
       53 天前
    网站的代码是开源的吗?
    chongchongzl999
        10
    chongchongzl999  
       51 天前
    已经辟谣,套壳 claude
    关于   ·   帮助文档   ·   博客   ·   API   ·   FAQ   ·   实用小工具   ·   5912 人在线   最高记录 6679   ·     Select Language
    创意工作者们的社区
    World is powered by solitude
    VERSION: 3.9.8.5 · 32ms · UTC 02:22 · PVG 10:22 · LAX 19:22 · JFK 22:22
    Developed with CodeLauncher
    ♥ Do have faith in what you're doing.