Hacker News Reader

Comment by linsomniac | Hacker News Reader

I've just been playing with Claude 1.3 this weekend to summarize large texts. It can take 100K tokens of input, enough for a whole Lex Fridman interview! :-) I've been getting pretty good results with it, so I'm excited to see how v2 works.

View on HN

Now that it's entered open beta it's going to iterate rapidly. I had been using it fairly extensively, alongside other LLMs, through Slack and was always most impressed by it's output over the others.

(I do hold investment in Anthropic, but do not base my statements on that)

How are you invested in Anthropic?

I posted about it in another comment but will restate it here too. The ARK Venture Fund has exposure to it (roughly 7% of the fund).

Can you share the prompts you used ? I'm really happy with Claude-100k for summarization, but I wonder if a better prompt would make it even better.

Sure, here it is:

    Human: Here is the transcript of a podcast:

    <transcript>
    [PASTE TRANSCRIPT HERE]
    </transcript>

    You are an expert on writing factual summaries.  Write a summary of the podcast in about 10 sentences.

    Assistant: I'd be happy to, here is the summary:

Reply on news.ycombinator.com